[a / b / c / d / e / f / g / gif / h / hr / k / m / o / p / r / s / t / u / v / vg / vm / vmg / vr / vrpg / vst / w / wg] [i / ic] [r9k / s4s / vip / qa] [cm / hm / lgbt / y] [3 / aco / adv / an / bant / biz / cgl / ck / co / diy / fa / fit / gd / hc / his / int / jp / lit / mlp / mu / n / news / out / po / pol / pw / qst / sci / soc / sp / tg / toy / trv / tv / vp / vt / wsg / wsr / x / xs] [Settings] [Search] [Mobile] [Home]
Board
Settings Mobile Home
/g/ - Technology


Thread archived.
You cannot reply anymore.




File: 1655814397970.jpg (182 KB, 1527x342)
182 KB
182 KB JPG
What do anons think of the rx 7000 leaks?
>>
>>87442495
>2X RTX 3090 performance for $1099
seems legit
>>
>>87442527
They say that same shit every generation it only turns about to be at 30 to 40% faster
>>
>>87442495
what retard did this garbage come from? 512bit 7970 is a dead giveaway that this is fake when you look at the rest of the chart
>>
people are going to be massively disappointed when next gen gpus and cpus aren't a massive leap like they thought it would be. they have to truly wait for zen 5 and rdna4.
>>
>>87442557
another thing is that these retards keep putting 2x performance jumps for nvidia/amds leaks
last time it actually happened was with the gtx 1000 series
>>
three 8 pin and one 12 pin power connection needed? does it come with its own nuclear power plant too?
>>
why would the 7800 XT have less VRAM than the 6800 XT
>>
>>87442495
>128bit bus
>3080
>>
File: performance.png (29 KB, 842x456)
29 KB
29 KB PNG
>>87442584
lol no thats not a hard thing to believe, AMD has already doubled themselves gen over gen and they're likely gonna do it again.

I expect nvidia to double as well, seeing as SS8 was a terrible node for clocks/power and they're adding a giant cache to ada lovelace. So if a 3080 keeps up with 6900xt despite no LLC, its gonna be a game changer for the 4080.
>>
>>87442495
Probably going to get a R9 7900x+7900xt
>>
>>87442601
its not an nvidia card and thats 3x8 OR 1x12
>>
>>87442624
5700 XT and 6900 XT were not equivalent products for their respective generations retard
>>
>7970
why are they doing this
>>
>>87442584
We already have 50% increased perf per watt going from RDNA2 to RDNA3 from AMD's own claims. Also consider that it's going to draw 50% more power you're looking at a 2.25x increase in performance.
>>
>>87442642
They are both the best they have to offer. This image and all of the rumors point towards flagships from both companies at being 2x larger. Unless you think nvidia and amd are going to massively fuck up performance scaling, there is no reason to doubt nearly 2x performance increase gen over gen, flagship vs. flagship.
>>
Cool I found a good uv oc for my 2080ti and 6900xt to last Aussie summer until new years
Gonna get a 7950 black
I love both these gpus but vr just beings them to their knees and rt performance is meh
>>
Kopite7komi said the new Ada would do 2.4x 3090
>>
>>87442557
The memory controller and cache are separate chiplets.
>>87442584
It's more than twice the cores and they say it's a 50% perf/watt increase.
6950xt is 335W TDP and the 7950 looks to be 400-475W.
>>87442642
6900xt has double the cores of the 5700xt and almost exactly double the performance.
5120 vs 2560.

>>87442683
Highly unlikely. It's a 60% core increase or so and power isn't doubled neither.

>>87442610
Cost

>>87442654
>he doesn't remember THE most legendary GPU of all time
>>
>>87442683
It will not. It's the same basic architecture as Ampere and adds only 80% more cores. The best they can hope for is 1.8x a 3090. It may be 2.4x better in RTX which is a shit benchmark.
>>
>>87442495
the only thing they've got right is the memory specs, release date, 7700XT/7800XT/7900XT and the L3 cache.
AMD is looking to do dual dies early next year as the initial models will all be single die... and reusing the 7950 and 7970 model numbers is pure wank, but would be cool.
>>
>>87442495
>top die for anything less than $2000
Fake
>>
>>87442495
How many CUDA cores?
>>
>>87442692
>The memory controller and cache are separate chiplets.
Then why also do the 2x 5120 monolithic setup according to the OP?

I believe the original rumors a lot more than this retardation. 32gb 512 bit is going to be above and beyond. I *highly* doubt they're going to waste the die space on n31 for the extra IF links they'd need for that full bandwidth.
>>
>>87442570
Nah nvidia will do just fine even if all they do is die shrink ampere and slap more L3 cache and call it a day. Just samshit 8nm to tsmc 4n would be 50% performance
>>
>>87442740
yeah and every clickbaiting retard on twitter is hyping the new gens as 3x faster, not even MCM Rdna3 is going to get more than 2x
>>
>>87442736
Radeon Pro RDNA3 is 768 bit, anon.
512 bit is just a cut down 3x 192 bit instead of 2x or 4x.

>>87442717
Top die isn't listed as under $2000. It probably is $2000 or $1899 at best.

>>87442755
More than tripling cores, higher TDP, and a node shrink, can certainly get 2x or more even with the losses of infinity fabric.
>>
>>87442698
>It's the same basic architecture as Ampere
Kopite7kimi literally said it's not and he's the source of leaks everyone is using so I'm not sure where you go that from
>>
>>87442782
>Radeon Pro RDNA3 is 768 bit, anon.
>512 bit is just a cut down 3x 192 bit instead of 2x or 4x.
okay, nostaseronx.

chnk retard just wish you'd fuck off already
>>
dont forget anons, Nvidia is getting off glorified 10nm that Samsung likes to call 8 and is heading straight to node parity with AMD this time around
honestly concerned for AMD they didn't demolish Ampere given rdna2 was like 2 nodes ahead
>>
>>87442868
Rdna2 already smacks Ampere in raster in 1080p. The saving grace for Nvidia is they have gddr6X while rdna2 doesn't.
>>
>>87442868
Nvidia is getting the node advantage of 4nm and its going to be slower and run hotter.
AMD is going to be 6nm and 5nm.
>>
>>87442812
>A few more tensor cores and a node shrink count as a new architecture
Even RDNA3 isn't really a new architecture like RDNA was. More of a reshuffling of RDNA2 with minor restructuring to support MCM.
>>
>>87442952
>AMD is going to be 6nm and 5nm.
5nm, and it's basically the same thing as "4nm", they're on the same platform. N4 is just an alternative to the enhanced N5. Like N6 and N7+ there's little difference in performance and power.
>>
>>87442654
The simulation is looping. The new mainstream laptop chip from AMD is codenamed Mendocino, just like the old based Intel Celeron.
>>
File: AMD-Navi-31-GPU.jpg (62 KB, 692x538)
62 KB
62 KB JPG
>>87442965
nope, its both 6nm and 5nm.
Navi 33 is probably going to be 6nm, Navi 34 is probably going to be monolithic 6nm
>>
File: 4chanwordcloud2010.png (500 KB, 1000x1000)
500 KB
500 KB PNG
>>87442495
i bought yesterday a rx 6600
really great card
idle/desktop use only 5 watt power
max 100w power
3d mark 11 performance benchmark 27990 points

when you have an old graphic card dont wait , upgrade now.
>>
File: 000fb0e0828ae8b87a193.png (2.82 MB, 1116x1080)
2.82 MB
2.82 MB PNG
trust me tm in 4 months amd will release card 30% faster than 3090 for one quarter of price ! totally real , will happened !
>>
File: 1560648186297.png (347 KB, 706x795)
347 KB
347 KB PNG
>>87442980
>The simulation is looping
More like mommy su is taking a fat shit on intel and rewriting over all their history
>>
>>87442495
>implying it will get assembled before Taiwan gets into communistic China
>>
>>87442495
Bog standard 30% performance increase, but without any affordable options. I think I don't give a shit.
>>
>>87443103
You will never see anything "affordable" again. There's too much money to be made in high end and not enough TSMC capacity to waste on economy options.
>>
>>87443103
>twice the ALUs per CU
>higher clocks
>higher CU count than prior designs
>can do 2 64 wavefronts concurrently
hurrr its only 30% faster!
>>
>>87442721
>using CUDA
Lol, get back to work, wagie.
>>
>>87442692

srs question. Where the fuck do I learn more about GPU architecture/featuresets
>>
>>87443152
Unironically start at wikipedia, for example https://en.wikipedia.org/wiki/List_of_Nvidia_graphics_processing_units
Every generation has a microarchitecture article link. Go to them and look at the linked sources. NVIDIA publishes whitepapers, AMD has open docs too.
>>
>>87443152
Locuza https://www.youtube.com/channel/UCaFk_ygFCffeQhYouGCcAkQ
Coreteks
https://www.youtube.com/channel/UCX_t3BvnQtS5IHzto_y7tbw
>>
>>87443132
If the prices are accurate and the thing slotting into the 6700xt's price is 30% faster than the 6700xt, then yeah, it's only a 30% improvement.
>>
>>87443201
There's no 6700XT there, but 6800XT as the lowest equivalent. The cheapest 6800XT on newegg is $729 while 7700 is listed as $479
>>
>>87443190
>>87443197
Thx anons
>>
>>87442654
Because it's the 10 year anniversary
>>
>>87442495
Are the power reqs going to be insane like nvidia. Cost of Ownership is starting to become a thing to consider for these cards.
>>
>>87443503
It's literally a non issue unless you're a miner faggot
>>
>>87443523
or a neet playing 24/7
>>
>>87443523
I mean if you play vidya for 6 hours a day on a high end machine, your looking at like $20+ a month just for your PC.
>>
>7970
KINO, it was my first GPU when I was 11 lol, used it myself for 5 years then passed it on to a friend and now it's two friends later and still going strong, just had to change one of the fans on it
legendary gpu, all of my childhood memories were made on it
AMD really got me in the feels with this one
>>
So what's the power consumption like?
>>
>>87443907
Requires an Electrical substation.
>>
File: 1202643726.gif (163 KB, 112x112)
163 KB
163 KB GIF
>>87443064
>>87443064
>when you have an old graphic card dont wait , upgrade now
No.
>>
File: 2or9qlrd72q11.jpg (103 KB, 1034x689)
103 KB
103 KB JPG
>>87443966
And a fire department
>>
>>87442495
EVERY. FUCKING. GEN.
>>
File: 1655496234028.png (1.3 MB, 1280x1062)
1.3 MB
1.3 MB PNG
>>87442495
>5700 XT
>256bit
>$399 launch price

>6700 XT
>192bit
>$479 launch price

>not even XT
>128bit
>same launch price as the already overpriced prev-gen XT

The way thing are progressing the 8700 will manage to be 64bit and have a launch price of $649

Remember the good old days before crypto was a thing?
Mid-range was actually mid-range; Cards were readily available; Prices were fluctuating between 200 and 300 bucks; You didn't need a power plant to run a high end system... too bad those days are long gone and never to return.
>>
>>87442495
I see they adapted the Nvidia model of releasing a 3090 ti super deluxe edition now
>>
>>87443132
At the <$1000 end, yeah it looks to only be around 30% faster.
It's just better price:performance scaling into the high end.

>>87444303
Maybe it'll actually wind up being the 7600 instead of the 7700.
Just hard to imagine them releasing a 7600 that's close to 3090 perf. The die is probably 350-450mm^2.

>>87443907
Better than Nvidia.
>>
>>87442537
6900XT is about 2x 5700XT
they'll do it again (half of it will be due to higher TDP)
>>
>>87444252
not true. my 3070 runs very cool under full load. never above 59°C.

maybe don't buy the cheapest PowerColor card
>>
>>87442584
RDNA3 unironically has a shot at doing it again
node shrink plus architectural efficiencies add up
>>
>>87442584
AMD has rdna3+node shrink+higher power, while Nvidia is moving to a much more efficient node in tsmc n4 and they're cranking up the power. It's believable that they'll both have a x2 increase in their flagship cards
>>
File: file.png (273 KB, 460x677)
273 KB
273 KB PNG
>>87442495
>3x 8-pin power
>12-pin power
what the actual fuck
before long you'll need a dedicated circuit for your computer with how insane things are getting
>>
>>87444894
I only see it getting worse since they'll eventually be unable to shrink the node.
>>
>>87444894
>what the actual fuck
These are top end cards pushing the architecture as far as it can go. These architectures are incredibly power efficient if you're okay settling at RTX 3080-ish levels of performance.
>>
>>87444759
Yeah I remember all the FUD posting about how the 6900xt wouldn't be twice as fast "just" because it has twice the cores. That they can't get 2x scaling.
ofc the TDP is higher as well, and there were architecture improvements, but both were on the same 7nm node.
RDNA3 has architecture improvements as well to make up for the losses in using chiplets.

>>87444894
>2x more powerful card uses 1.5x the power
I don't see the problem as long as they're not 600w nvidia housefires.

If the 7950 is 2x the perf of the 6950xt, and uses 475W instead of 335W, that's still over a 50% perf:watt improvement.

Not like you can't undervolt. Sacrifice 10% performance to save 25% power or so. They sadly can't use 350W cards to compete with Nvidia's 600W cards. They need to push them closer to 500W to compete.

We've already seen this happen with... was it the 5600XT or something?
They initially were going to be efficient cards. But Nvidia pushed out housefires so the AMD cards had to raise TDP to barely be much lower but perform better. Nvidia is the one driving this massive power consumption escalation and only regulators can stop it since consumers are stupid.
>>
>>87444894
First off it's either 3 8pin or 1 12 pin, but yea power consumption is getting out of hand
>>
>>87445006
Stop looking for power efficiency at the highest end you retards. No one expects a Titan to be power efficient besides you two.
>>
>>87445026
The more expensive cards tend to be better fps:watt, actually.
I'm sure someone has a graph.
>>
>>87445002
why would regulators stop it? just don't buy a fucking TITAN card from nvidia or amd and you're done
>>
>>87445002
>They initially were going to be efficient cards. But Nvidia pushed out housefires so the AMD cards had to raise TDP to barely be much lower but perform better. Nvidia is the one driving this massive power consumption escalation and only regulators can stop it since consumers are stupid.
based anon I hate the revisionism taking place. Nvidia has been doing this for a decade now.
>>
>>87444759
The 6900xt was twice the performance of the 5700xt for more than twice the price. It wasn't even a gen on gen improvement. If they're going to "do it again" then they'll double it for 2k, not 1100.
>>
>>87445055
They could make them put yearly power cost stickers on the box like appliances have.

It's not just titan cards lmao?
The 3060ti is fucking 200W. People said Vega56 was a housefire when it used that much.
The 1060 was a 120W card. GTX 980 was 145W.
Nvidia is going insane pushing TDPs up like this.

>>87445083
Not really. It's namely been since the 2000 series then again with the 3000 series.
GTX 960 and 1060 were both 120W.

>>87445088
AMD is fine with high prices as long as Nvidia doesn't want to compete on prices either.
Intel Arc should at least drive down prices in the <$300 range.
>>
File: 2200.jpg (75 KB, 970x728)
75 KB
75 KB JPG
>>87444894
>before long
seems you're still living in the good old days
>>
>>87442495
I have a 7970 already, it's pretty good
>>
>>87445026
If the top end cards are pushing 600w than that shit will trickle down and we'll have 500w 4080s you retard
>>
>>87445137
1X 8-pin for the 7700 which is equivalent to a 3080. What space age technology do you think AMD has to deliver 500W over a single 8-pin?
>>
>>87445122
not possible for amerimutts
>>
>>87445167
I was talking about Nvidia also the new atx3.0 spec can deliver 609w per connection
>>
>>87445167
Nobody cares about a $479 7700 AMD shill. Shoo shoo.
>>
>>87445185
>no one cares about a 3080 for half the price with better RT performance
retard
>>
>>87445175
It is just gotta have some electrical work done beforehand :^)
>>
>>87445105
>3060ti
thats less than the 5700XT for the same or better performance.
>>
>>87445190
B-but muh COODA and muh dee el ess ess
>>
>>87442495
I think that I really regret getting a 6900xt, because the drivers and VR performance are awful, so I will not fall for it again.
>>
File: amd vs nvidia.jpg (75 KB, 311x837)
75 KB
75 KB JPG
>>87444252
>>
>>87445212
Then sell it and buy a 3070 or whatever you can with the money you get
>>
>>87445246
3070 has worse performance. 3080ti and 3090 literally where do I get those from? I have to wait for the 4000 series and hope the prices are acceptable and the availability exists.
>>
>>87445212
Really? I'm extremely happy with my 6900XT. 1:1 3090 rasterization performance and better Linux drivers for less than half the price of the 3090 after markup.

I got my 6900XT for MSRP ($999). A 3090 at the time cost $2600 and nowadays it's still almost double at $1950
>>
File: wojack crying soyjack.jpg (101 KB, 785x731)
101 KB
101 KB JPG
>>87445191
>nooo my landlord wont let me get a 220v outlet for my gaymerinos, how dare you require such high power GPUs
>>
>>87445269
VR performance on AMD is shittier due to devs only supporting Nvidia since they dominate the market, but they're fine for regular gayming
>>
>>87445278
Why the fuck would I ask them for permission, just get it done.
>>87445260
See if anyone is interested in swapping their 3080 for your 6900, you might get lucky.
>>
>>87445296
I have an Index running at 120hz supersampled and I have no idea wtf you're talking about anon
>>
>>87445269
Linux sure, but I don't use it. I only got it because it was what's available shortly after release and the prices weren't jacked up yet. I was looking for a 3080 but couldn't get one, so I just sniped the first thing I could. Can't play Dragon's Dogma well, since it's a DX9 game, recording stutters no matter what settings or software I use, VR performance is literal dogshit for what I want to do at least. Before that I had a 1660ti and I could play Dragon's Dogma and record with no issues. Why can't I with a 6900xt? I only upgraded because I wanted to get into simracing in VR and my poorfag friend needed a card, so I gave him the 1660ti, otherwise I'm literally stuck here playing HOMM3 on a 6900xt. What a waste. Fuck AMD.

>>87445296
It's actual the bus speed, but whatever.
>>
>>87445313
I'm talking about the quest and all the new meme VR devices being shilled lately
>>
>>87445307
>swapping their 3080 for your 6900

Is that really worth it though? the 3080's VR performance is more or less the same. I guess I'll be able to play Dragon's Dogma, but I feel like I should wait for the 4000 series then sell this when I get a new card. Hopefully the prices aren't fucked and you can actually get one, but who knows, because I live in yurop after all.
>>
>>87445313
I have a Reverb G2 and I'm trying to play simracing games. Obviously an AMD card would be enough for your VR chat on an index.
>>
>>87445338
I'm playing Onward/Pavlov/Elite Dangerous/Into the Radius/Sairento/Project Cars 2 bro
>works on my machine
>>
>>87442495
>7800XT
>192b Bus
They pulled that one out of their arse.
>>
>>87442495
>starting at $500
i do not fucking believe they profit more by excluding half a billion of people from buying their GPUs
>>
>>87445405
Yeah you're playing easy to run games on the best VR headset currently. Good job. Happy for you.
>>
>>87444759
Higher tdp, but it's rumored that they can clock upto or over 3ghz, and judging by the 6950xt and ps5 apu, I'd say that rumor is legitimate
>>
>>87445444
sorry sweetie, if you want more bus you gotta buy the 7950 XT TI super 16GB rev 2.0
>>
>>87445444
gddr6x and 50% more cache should somewhat make up for it.

>>87445469
Hundreds of millions of people don't buy GPUs.
# of GPUs sold in 2021 was 50 million iirc. A lot of those were to miners, I think well over 15 million.
>>
>>87445490
my rx 580 runs pavlov and onward no problem, you're just an idiot that doesn't actually play VR
>>
>>87446030
retard
>>
>>87445787
>Hundreds of millions of people don't buy GPUs.
but they CAN buy GPUs, i don't see why they want to miss out on this big of a market
>>
>>87446167
Exactly, only a fool wouldn't realize the goldmine that has descended
I've ordered several GPUs for when crypto goes back up and mining skyrockets and I'm going to buy more every time I get paid
Given the previous bull run, it's only a matter of time before I can double my money
>>
>>87446167
6600 is perfectly affordable for a billion people.
2.5x the perf of a gtx 1060.
All that 99% of people need.
People don't buy it by choice and ignorance.
>>
>>87446213
it's as expensive as 1070 was when it came out and a lot of people will rather buy a second hand gpu that matches the performance for 1/2 of the price and $400 really is not a price everyone can afford
>2.5x the perf of a gtx 1060.
2.5x faster than a 5 year old GPU and 1.5x as expensive
>6600 is perfectly affordable for a billion people.
i was obviously talking about the bottom 1-3 billion of people, especially people in the developing countries, this is a huge market now and will be an even bigger one in the future i do understand that the margins they can make are much smaller but i really don't understand why are they deciding to just abandon them
>>87446211
wouldn't buying coins be more profitable?
anyway, you probably know a thing or two about mining: where the fuck are all of those post-mining polaris/1060s? I thought no one is going to buy them because muh mining destroyz gpuz!! but even the ones without video output are still more expensive than when they were out
>>
>>87442495
Damn, even GPUs are getting blacked now a days.
>>
>>87446343
People who don't know how inflation works should honestly kill themselves.
1070 was $449 msrp when you ignore the fake MSRP.
$449 adjusted for inflation is almost $600, twice as much as a RX 6600, you COLOSSAL FUCKING MORON.
>>
>>87446343
>where the fuck are all of those post-mining polaris/1060s
Many are for sale in the channels I frequent at good prices, but you won't find these guys selling on ebay or other public sale sites
It's a buyers market, but I don't think it'll last more than a couple months
>>
File: 1655314128285.jpg (103 KB, 729x720)
103 KB
103 KB JPG
>>87442495
>7950 Blacked
>7970 Blacked

why make fake pic and insert an obvious lie, is it some sort of filter for low iq gayman ?
>>
>>87446480
we don't live in the same country so the details are going to vary but you still managed to miss my point quite well i'm impressed by how fucking retarded you are i bet your parents are proud the massive imbecile they've raised
>>
>>87442495
I bought an AMD GPU once, I will never make that mistake again
>>
>>87442495
AMDchads I kneel....
>>
>>87446554
AMD has a tradition of using "Black Edition" as a designation for fully unlocked SKUs. Although I think it just means power in this contenxt, because being able to OC is the norm in GPU world.
>>
File: old.gif (2.31 MB, 360x200)
2.31 MB
2.31 MB GIF
>>87442495
>7970 black
>the hd 7000 series was released 10 (ten) years ago
>>
>>87442495
>low-end GPU at almost $500
They can fuck off.
>>
>>87446583
The MSRPs are announced in USD, fucktard.
Because you pinoys steal GPUs and sell them cheaper doesn't make that the price.
>>
File: nvidia housefires.jpg (114 KB, 1277x364)
114 KB
114 KB JPG
>>87446987
Now look at Nvidia's
>>
>>87447016
you're not very bright are you
>>
mentall illness thread
>>
>>87447104
And yet you're here, curious.
>>
>>87444759
>comparing a high-end card to a mid-range card
what the fuck u doing

if we had a 5800XT then the 6800XT/6900XT would be 50% stronger
>>
>>87442495
>7950
>7970
SOVL
>>
>>87447142
See>>87442665
>>
>>87446343
Crytpo GPUs are expensive because of the people who own them. They're just assholes. Also about 80% of Etherium was mined from China at peak, and it's still in the 70s percentile range, so they're triple assholes.
>>
>>87446987
There are no low end cards on that chart.

No, lowest part of the chart does not mean low end. Same as how Geforce 16 and 20 series have no low end models.
>>
I had a gts 250 and recently upgraded to a gtx 660. Laugh all you want, I'm happy with it. I can play games I wanted to play I never could before. It's a great feeling.
>>
>>87442527
Possible, real world price will be 2K € then
>>87442624
Depends if they really boast so many shaders + more clock + more cache
I expect the top model to be 2x 6900 XT, the main question is what it will cost and if I can obtain one
Maybe I have to scalp me one
If i can get even a midrange NVIDIA for MSRP I will buy and resell it after a few weeks, if the mining hype will be nearly as horrible as last time it may be 50 or 100 % price hike
At this point its very unlikely the top model will be under 1K so I hope the richfags can buy me one
THis time I also don't need a GPU on laucnch so desperately
>>
File: 1621954545138.png (437 KB, 2048x1638)
437 KB
437 KB PNG
>>87442495
Bought a Steam Deck and I don't need more.
>7950 Blacked Edition
>Eleven-hundred Petrodollars
>It's not even the top end SKU
Holy Jesus Lord have mercy, what the fuck are you doing Lisa?
>>
>>87448853
>>Bought a Steam Deck
Don't remind me
>>
original 7970 is usable even to this day
>>
>>87449289
not true now that we live in the always online constantly pushing updates world of 2022, every game has implicitly gotten harder to run as time goes on.
>>
>>87442495
>Completely fake

Big RDNA3 will likely be have 40-50% at best over Big RDNA2 in most cases. It'll be only be much faster at ray-tracing stuff perhaps be about on par with Ampere but falling short of Ada.
Most of the micro-architecture improvements will be focused on ray-tracing stuff.
>MAH MULTI-DIE CHIP
We will have go back to multi-card/GPU rendering again for this to be feasible. CF/SLI support has been neglected for nearly 10 years.
>>
>>87448853
Waiting for Steam Deck 2 here.
Really needs a replaceable OLED screen, 20% larger battery, and to be like at least 15% lighter.

>>87449289
eh there's some lack of features that hold it back. It's still well below a RX 580 or GTX 1060 and the power consumption is AWFUL. It uses 100w minimum with 2 monitors

>>87451683
lmao. You really think 3x the cores, a 30%+ higher TDP, and node shrink is only going to be 50% faster.
You're delusional.
Just a node shrink alone would be a 30% performance increase.
>>
>>87448046
Kid... go to Fiverr or DeviantArt. draw some coom art and you'll be able to at least afford a 1080 ti like a normal human being.
>>
>>87442495
The 3060 is already "let me guess" tier for me. I just wanted a 2060S with HDMI 2.1, and got it. Nugames suck and it runs nugames fine and older games fantastically.
>>
>>87451693
>lmao. You really think 3x the cores, a 30%+ higher TDP, and node shrink is only going to be 50% faster.
>You're delusional.
>Just a node shrink alone would be a 30% performance increase.
You are delusional kiddo. The days of easy gains have been long over since Kepler. The wall of diminishing returns has been scaling higher since.
300W+ chips are non-sellers for the vast majority of the market.
This fake leak implies that multi-chip packages are coming in the customer space (CF/SLI are coming back)
>>
>>87451683
>Most of the micro-architecture improvements will be focused on ray-tracing stuff.
Why talk out of your ass when compiler patches already detailed RDNA3? Why do you choose to be a fuckup retard who posts worthless embarrassing nonsense? 128 ALUs per CU up from 64. Literally a doubling of execution resources that have nothing to do with ray tracing. Dumbfuck.
>>
>shills get increasingly irate as anons rather talk about nvidia in an amd thread: the thread
>>
>>87452008
no one cares about Nvidia anymore
>>
File: tired.jpg (20 KB, 306x306)
20 KB
20 KB JPG
>Next Gen is all 400W+ housefires on all sides
Yawn. I'll stay on my 1080 until they decide to make actual efficiency improvements and they can give me more than double my current performance at 200W or less.
>>
>>87452131
Get a 7700 then
>>
>>87452131
RDNA3 improves perf/watt by over 50% compared to RDNA2.
>>
>>87451966
>Theory-crafting this hard without official white papers
Keep setting yourself for more disappointment for buy into the hype
>>
>>87452201
>a compiler commit is "theory crafting"
Gay retard.
>>
>>87452213
>kiddie getting told this hard and doesn't want his dreams get dashed
>>
>>87451935
Nuuuuuu we just need to reach below 1 nanometer architecture and then the secrets of the universe will unlock for us!! /sarcasm

Realistically what will happen when we go below 1nm ?
>>
File: IMEC roadmap.jpg (46 KB, 970x555)
46 KB
46 KB JPG
>>87452476
There is a lot of room for area and feature size scaling yet, and tens of billions of dollars is being spend ensuring these future processes continue to yield substantial improvements
>>
>>87452131
Isn't a RX 6600 like 30% faster than a 1080 while using half the power?
I know it's 2.5x the performance of the 1060 while only using 20 watts more power.

So the 7600 should be closer to 60% faster.
>>
did anything ever come of those leaked drivers a couple weeks ago? did they actually fix amd's shitty opengl performance on windows?
>>
>>87442527
>$479 for entry level

Guess I'll be sticking to Polaris for a while.
>>
>>87452824
That obviously isn't entry level. Smaller die entry level designs always launch later.
>>
>>87442963
RDNA2/3 is mostly the same as RDNA1, they just put a shitload of SRAM on the die to act as cache. That's the only reason why it's so fast.
>>
>>87452846
It also clocks higher. 5700 XT + 35% higher boost clock = 6700 XT.
>>
>>87442495
Brave of AMD to re-release the HD7970 over a decade later.
>>
>>87452846
>just posting retarded asspulls when reference manuals and programming guides exist
Infinity Cache offsetting the VRAM hit rate is not what makes RDNA2 fast or power efficient.
>>
Can someone explain to me like I'm fucking brain damaged why the memory bus/bandwidth seems to be getting smaller each gen?
>>
>>87452953
Solutions to providing very high bandwidths with DRAM solutions are costly. They're costly in dollar amounts and in power consumption.
AMD's solution to this is what they call Infinity Cache. It is essentially just a large slab of L3 SRAM used to lessen the need for off die dedicated video memory like HBM or GDDR6/X which are all comprised from magnitudes slower DRAM slices.
>>
>>87442495
I can't afford the nuclear reactor required to operate a 2023+ home computer
>>
>>87442495
Fake. No company launches the entry level before the flagship. Also 8gb of ram after they have the 6700 with 12 is also dumb.
>>
>>87453134
7700 is mainstream, not entry level. Words have meanings
>>
>>87442624
They doubled the shaders, clocks, vram and added a bunch of cache. The nickname of the chip is Big Navi. Oh yeah look at the price and how much power it draws. 200% is the bare minimum.
>>
>>87453149
Still makes no sense launching it before the flagship or XT versions. AMD does sandbag but to this point is beyond stupid. Nobody wants the slower product.
>>
>>87453218
They're first to market with chiplet GPUs they have issues that only they are seeing for the first time. It takes time to iron those out before release.
>>
>>87452824
You know the $280 rx 6600 is 2.5x faster than your rx 480, right? Also uses less power.

>>87452953
It's expensive.
Uses die space.
Faster memory requires a narrower bus for a given bandwidth.

Your question is like asking
>why didn't they just put 1024bit bus on a 10 year old GPUs and make them 3x larger dies to make them twice as fast?
But since RDNA3 seems to have 64mb cache+64bit bus chiplets, they may feasibly be able to actually do 512bit and larger bus from now on which will get much better scaling.

It's cost.
As dies get larger for more compute and such, they try to save die space on smaller memory bus and shit like that.
It's easier to save money on compression and shit for higher effective bandwidth and other optimizations and use faster memory than it is to make the gpu itself weaker in exchange for more die space reserved to memory bus.

>>87453218
My guess is it's a way to increase prices without it feeling as bad.
You get a 7700 that's faster and uses less power than the 6700xt. It costs the same but it's still better.
Then they can add a 7700xt later that matches the 3090/6900xt that's say $549 and drop the 7700 price a bit to $449 or so and they seem like the good guy.

Fact is that GPU prices are going up. They're always going up. Inflation exists.
>gtx 970 was $329
>gtx 1070 was $449 (/w fake $379 msrp that the FE and no other cards on launch sold for)
>rtx 2070 was $599 (/w fake $499 msrp that the FE and no other cards on launch sold for)
Nvidia used fake MSRP whch no cards were ever sold at since pascal to hide the rising MSRPs. AMD will probably use repositioning of SKUs in the stack to obfuscate rising MSRPs.

Since Nvidia appears to be releasing housefires, it gives AMD room to release power efficient GPUs on launch and then respond with their own housefire XT versions afterwards.

>>87453187
5700xt is 225w. 6900xt is 300w. That's not double.
>>
>>87442495
>No 200-300$ stuff
Not interested
>>
>>87453812
>You know the $280 rx 6600 is 2.5x faster than your rx 480, right? Also uses less power.
It's 60% faster you retard. It's also not $280 but ~ $300. Also 2.5x faster implies 350% you ungodly abomination, which is even more delusional.
>>87452835
480 was a midrange card for $229 msrp. Whatever abomination you get nowadays for an inflation adjusted price of $265 is a joke considering it has been 6 years. Case in point, aforementioned rx 6600 costs 300 bucks and is 60% faster, a fucking vega 56 is as fast and went for ~ 220 bucks back in 2019. Comparing GPU prices to anything but pre-cough levels is retarded beyond belief.
>>
>>87454825
How are you so wrong?
Why don't you look up benchmarks that show how wrong you are?
You're just going by tflops or something stupid.

And no, 2.5x is +150%. You're terrible at math and need to stop calling others "retard" when you're the most retarded person in this thread.
>>
Don't care
I just want a 6600 to drop to 250€
>>
>>87442495
>release marketing numbers
>but it's a super secret leak ;)
>>
This is without a doubt one of the most retarded threads I've seen on /g/.
Both the OP and most posters are wrong about every word they've posted.
AMD has been on a huge leaker trolling campaign and it has truly gotten out of hand, N31/32 are single GCD, have chiplet MCD's which contain L3 cache and memory controllers on N6, GCD's are N5HPC. N33 is monolithic N6.
RDNA3 has double the ALU's per CU/WGP, so double per unit, RDNA 3 is always 4WGP/SA, unlike N21/22 being 5WGP/SA, the 4 config is better PPW and PPA.
RDNA3, like Zen4 clocks much higher than its predecessor, 3Ghz is comfortable.
RDNA3 is arguably as big as GCN>RDNA1 in regards to the overall changes, though the perf benefits will be larger.
In general, N33 is 2SE, N32 4SE, N31 6SE. PPW is >50%, that means N33 is over 50% and that is quite a bit less efficient than the rest of the lineup due to the node.
The uarch changes are going to benefit a plethora of things, whether that is RT, raster, compute, it is all so much more advanced than RDNA1/2. Software is getting really refined in prep for this launch, with various fixes to longstanding weaknesses, and the full compute stack being supported at launch.
Infinity Cache is larger, with 32+MB and a 64-bit MC per MCD, 1 MCD per SE.
AMD has plenty of options in regards to segmentation and yield management, die sizes are reasonable, the nodes yield brilliantly, and the packaging is advanced, but rather cheap all things considered. AMD will be more efficient on the same node for the first time since goddamn Terascale 2. The peak perf crown is a wash most likely, but one reaches peak perf at far less power.
AMD has the more competitive product, it is their 5870 reborn.
And well, perhaps there is that chance that we do see well above 2X performance this time around, in which case this is their best card since the true GOAT, the legendary R300. To surpass even the almighty G80's ~2.2x perf in a generation, can it really be done today?
>>
>>87455436
Impressive. I might be able to buy a $350 4060ti if amd delivers.
>>
>>87445239
>no driver
kek
>>
>>87455512
Well, if one thing is true, AMD want to increase gross margins with each generation.
Each tier will cost a bit more than the last due to wafer costs increasing, the main thing is AMD will probably add 1-2 more tiers above what they are doing currently on the high end, while basically abandoning the lowest 1-2 tiers.
AMD is supply constrained, they would price more reasonably if they had volume to sell.
Of course we have the normal prepwork happening on both sides, both the 6X50XT and the 3090Ti exist to make the next gen look better than they otherwise would, whether that is perf-per-dollar or perf-per-watt. Nvidia will win this generation on the simple virtue that AMD does not have room to make as many cards, this will be the same way for another 3-4 years.
But AMD can claim a win via consoles and any CPU with an iGPU.
>>
>>87442495
I would be bad ass if they did a run of cards with HBM again, I miss my Radeon VII.
>>
>>87455756
HBM is elegant, but big cache+GDDR is a tad better for most clinet stuff, though it does use more power.
It is more economical to use now that AMD can use EFB instead of CoWoS, but the demand for HBM is nuts, keep it for HPC, that is where it shines.
>>
File: 1624397242601.jpg (107 KB, 394x394)
107 KB
107 KB JPG
>>87442495
>>87442527
https://www.digitaltrends.com/computing/amd-may-be-working-on-an-even-better-rdna-3-gpu/
>>
>>87445088
Nice goalpost moving retard
>>
>>87455857
A Pro Duo of Navi 32 is the best fit if it exists.
But honestly, AMD are just going full deflect mode at this time, whether there truly is a 2GCD card on one package is for time to tell.
>>
>>87455897
what are u talking about radeon has superior arch to deadend arch nvidia has to offer
it is very simple
look at die size
compare two
look at power drawn
compare two
>>
>>87447142
5700XT was RDNA1 flagship stupid faggot. RDNA2 flagship is 2x faster, RDNA3 will have another +50% perf/watt and more cores as well.
>>
>>87446480
>1070 was $449 msrp when you ignore the fake MSRP
>>87453812
>>gtx 1070 was $449 (/w fake $379 msrp that the FE and no other cards on launch sold for)
I’m still using a basic bitch ASUS non-FE GTX 1070 I bought at MSRP. No one forces you to buy the quadruple stacked fan 5X OC FROZR RGB Hyper Ultra MAX cards that add 20MHz to the base clock. And even if that’s the case now, just buy an FE.
>>
>>87446742
AMDCHADS WE'RE GOING HOME
>>
>>87451683
>RDNA3 will likely be have 40-50% at best over Big RDNA2 in most cases
You posted the same cope when RDNA2 was about to launch lmoa.
>>
>>87455936
Best to compare with perf-per-area.
6600XT is 1.11x the speed of the 5700XT with 0.94x the die size. About 1.2x PPA.
However the PPW is far higher, consuming 0.71X the power, does lead to about 1.5x PPW.
PPA gains are all from clock speeds as IPC is the same outside of DX12U stuff or some cache edge cases.
Considering the lower ALU count of the 6600XT, it really is nice.
Meanwhile Navi 21 and Navi 10 have identical PPA, IC and DX12U stuff is less area efficient than more cores and more memory controllers.
>>
>>87453907
Buy 6600 then you poorfag.
>>
>>87456059
And naturally larger GPU's gain diminishing returns in performance, scaling gets worse and clocks/power scale worse.
>>
Nvijeet cope. RDNA2 was just the beginning.
>>
>>87456075
>shit drivers
Sad!
>>
>>87454954
>And no, 2.5x is +150%
2.5x performance is 250%, which is +150% faster.
2.5x faster is +250%, which is 350% baseline performance.
"Faster" is comparative, which means you are comparing it to the baseline of 100% in this case. Learn some fucking basic grammer you imbecile.
>You're just going by tflops or something stupid.
I went with userbenchmark numbers fully aware of them being a shit source, although i was only aware of major problems with the cpu scores of the site - until now. I concede that my numbers are therefore incorrect - except of course the aforementioned comparative grammar nitpicking, i still claim you are in the wrong there.
For the most part, userbenchmark is ranking the v56 as faster than it actually is, a lie that i - as an owner of said card - went along with.
Still, from what other sources i browsed through it looks like the 6600 is approximately "only" twice as fast an rx 480.
>>
>>87456099
>shit cards
Maybe next time nvijeet))
>>
>>87456103
Uselessbenchmark has been into seething overdrive over AMD for at least 2-3 years. It's what happens when you're a coping mentally ill intelvidia fanboy in charge of maintaining a benchmarking site.
>>
>>87442495
rocm won't work
>>
>>87456179
Is this rocm game on steam?
>>
>>87444372
>I see they adapted the Nvidia model of releasing a 3090 ti super deluxe edition now
Did you forget dual gpu cards and the 7970ghz 6gb or the 6950xt that just came out?
>>
>>87442495
https://twitter.com/Kepler_L2/status/1537375541686329345
This is the most rational lineup.
>>
>>87456069
>>87456059
Youre really good at saying nothing relevant to what the other anon is saying
>>87456229
>twitter link
aint clicking that shit nigga, this is an image board, post screenshots or gtfo
>>
>>87456217
He's most likely a zoomer.
>>
>>87456267
The context is the 5700XT is a midrange GPU and the 6900XT is a proper flagship.
A 5900XT would probably be 50% stronger than the 5700XT at a somewhat smaller die size than the 6900XT.
>>
>>87456276
>He's most likely a zoomer.
Zoomers dont use 9yo gpus? Their around a 1060 and cheaper
>>
>>87456137
I have to clarify, i mistakenly used the "effective speed" shit they list at the top instead of the actual results. It's the main flaw of the site and i didn't pay attention like a retard, so i fell for their bs. Really have to update my userbenchmark scraper to use my new scraping library and pay some jeets a few bucks so i can finally host an alternative front-end for this shitshow of a site.
>>
>>87456298
5700XT is the RDNA1 flagship, your imaginary gpus are irrelevant.
>>
YOU SEE IN MY HEADCANON 5999XT IS ON PAR WITH 6900XT
>>
File: 1655698782176.jpg (425 KB, 1028x965)
425 KB
425 KB JPG
>>87456320
>5700XT is the RDNA1 flagship, your imaginary gpus are irrelevant.
No it's not that's what the 900 moniker is for u stupid cunt the 590 Polaris was also cancelled
>>
>>87456267
>RX 7970 XT (Navi31 XTX) 48 WGP/96 CU/12288 FP32, 24 GB GDDR6, 384 MB IC
>RX 7950 XT (Navi31 XT) 42 WGP/84 CU/10752 FP32, 24 GB GDDR6, 192 MB IC
>RX 7900 XT (Navi31 XL) 35 WGP/70 CU/8960 FP32, 20 GB GDDR6, 160 MB IC
>RX 7800 XT (Navi32 XTX) 32 WGP/64 CU/8192 FP32, 16 GB GDDR6, 128 MB IC
>RX 7800 (Navi32 XT) 28 WGP/56 CU/7168 FP32, 16 GB GDDR6, 128 MB IC
>RX 7700 XT (Navi32 XL) 24 WGP/48 CU/6144 FP32, 12 GB GDDR6, 96 MB IC
>RX 7600 XT (Navi33 XT) 16 WGP/32 CU/4096 FP32, 8GB GDDR6, 64 MB IC
>RX 7600 (Navi33 XL) 14 WGP, 28 CU/3584 FP32, 8GB GDDR6, 64 MB IC
I'll personally add that price is probably from $400-$2000 and perf relative to a 6900XT is 0.9-2.2+?x
Those are just safe personal estimates.
>>
File: 1596196651547.jpg (12 KB, 243x208)
12 KB
12 KB JPG
>>87456338
>IESLB
5700XT is the RDNA1 flagship, your imaginary gpus are irrelevant. Stay mad braindead zoomer.
>>
THIS FANFICTION RDNA1 CARD IS ON PAR WITH 6900XT BRO
>>
>>87456335
>YOU SEE IN MY HEADCANON 5999XT IS ON PAR WITH 6900XT
6900xt guy here what's this? My gpu isn't even a flagship anymore now that the xtxh and 6950xt are a thing
>>87456363
What's wrong dickhead I'm 32 my first ati gpu was a 9600 two decades ago don't pull age on me nigger
>>
>>87456363
Look, the point is that if RDNA3 doubles RDNA2's performance, it is far more impressive than RDNA2 doubling RDNA1.
And it is $400 vs $1000 ffs.
>>
>>87456385
XTXH was a nonexistent meme.6950XT is the flagship but that came out later, it's more like RDNA 2.5 since it also uses different memory.
>>
>>87456395
Figured as much it's literally jsut a unlocked bios
Gave up oc after a month and put it back to stock and enjoying 250w 3080 perf
>>87456381
>THIS FANFICTION RDNA1 CARD IS ON PAR WITH 6900XT BRO
He mad
I dunno why anyone would want a rdna1 gpu it was literally a beta test for rdna 2 which was a stepping stone to rdna 3 mcm
>>
>>87456395
I guess the HD7970 Toxic 6GB is GCN 1.5 then.
>>
>>87456410
rdna1 is already 3 years old
>>
>>87456414
1.488
>>
>>87456390
+50% better perf/watt already means that a 200W RDNA3 card will be at least 50% faster than a 200W RDNA2 card. And this is without more cores higher clocks and more cache.
>>
>>87445192
Weird way of saying the 3060ti is simply better.
>>
>>87442584
Chiplet design + better node + higher wattage. 2x performance is entirely reasonable.
>>
>>87456363
>5700XT is the RDNA1 flagship,
Correct
>your imaginary gpus are irrelevant. Stay mad braindead zoomer.
Incorrect.
>>87456298
>The context is the 5700XT is a midrange GPU and the 6900XT is a proper flagship.
Correct.

The 5700xt is the flagship gpu for rdna1 same way the 480/590 were flagships gor polaris. That said, a flagship is usually only the largest and most powerful vessel in its fleet (although there are exceptions), so a flagship in gpu generations is simply the most powerful in that generation at that time. Therefore, while a gpu like a 480 was technically a flagship, it was at best mid-range, while the 5700xt could be comfortably considered to be part of the upper mid-range, but most definitely not high-end. Mostly, these classifications arise from the actual chips and more specifically their dimensions. In Nvidias case it's also quite convenient that they are somewhat consistent in naming their chips.
>>87456506
Well that's an idealized way of computing things.
>And this is without more cores higher clocks and more cache.
I'm very certain that this 50% efficiency improvement they claim factors in clock speed changes and most definitely the cache. Also more cores will be part of the equation as well.
>>87445105
>Nvidia is going insane pushing TDPs up like this.
And it will be called a house fire. I'm already collecting material to make memes.
t.v56 owner
>>87456568
this. Mostly the chiplet design allows for truly huge GPUs. Just compare, the 6900xt has 5120 streaming multiprocessors. The 7900xt has allegedly double. If clock speeds are somewhat lower but the cores have more ipc through u-arch improvements and more cache we're easily looking at 2x performance. If the overall tdp rises by 33% (+100% for double cores - potentially lower clocks - u-arch caused voltage reduction - process node improvements) that would be a 50% improvement in efficiency. Now amd could also increase clock speeds for some designs and sacrifice some of the efficiency.
>>
>>87443503
Higher than current gen, but not as insane as Nvidia.
>>
>>87442495
>7950 BLACKED
dropped
>>
>>87456725
I'm not sure if this is ironic /pol/ posting and I've been baited or if you're actually retarded. Kudos.
>>
>>87455436
okay but if they don't bring back the 7970 I'm not interested
>>
>>87456229
>>87456359
42 to 48 is too small of a jump on the high end.
I don't think the cache bump would be enough to make it scale.

It's pretty pointless to utilize chiplets unless you're getting good scaling.
Why Ryzen was so good if that the 16 core actually scaled 2x over the 8 core (at least if you also had better binning or higher tdp).
If you don't scale the bus with the cores, and more heavily scale the cores, it's dumb.

Also it's 16384 cores maximum. Why the FUCK would they only do 12288 for the 7970?

The bad timeline is that the highend is actually 8192 max and the 2 GPU card is radeon pro only but I've heard that's not true for years.
>>
>>87456662
t. malding zoom
>>
>>87456844
Knowing the state of poltroons it was probably actual retardation.
>>
>>87455756
Reserved for CunnyDNA.
>>
>>87455033
>he doesn't know
>>
File: FQd-BpyXsAAYjpT.png (57 KB, 621x454)
57 KB
57 KB PNG
Where's my RDNA3D?
>>
>>87458771
Look I'm yuropean, I don't know where to look. All I find is these for like 329€.
>>
>>87459370
This would be really cool. Just like a few GB of infinity cache would make for a monsterous card.
>>
>>87460518
>a few GB of SRAM
Anon, you're drunk
>>
>>87460541
The 6900xt already has 128MB. The 7970xt will supposedly have 512MB.
>>
>>87460635
>this one card had 128MB
>therefore several GB is totally feasible
>>
>>87460680
the cache is on chiplets.
64bit bus and 64mb cache per chiplet.
8 of those chiplets = 512bit bus and 512MB cache

384MB and 384bit is more likely though, at least for a consumer card.
>>
>>87456061
>200-300$ stuff
>Posts about 350$ garbage
kys
>>
>>87442495
damn the 7700 sounds kinda based if true. i'd consoom if cryptocels are gonna keep getting fucked, probably not gonna happen though unfortunately
>>
>>87444930
There are already factories being made in Japan that will use Galium to make 2nm chips.
>>
>>87445278
You can buy and install a small 220v transformer the size of a coffee machine at your house. I have one for my AC.
>>
>>87460680
Imagination is the only limit.
>>
>>87460888
>turd worlder
Not my problem.
>>
>>87460635
1488MB when
>>
File: 01103601781l.jpg (242 KB, 800x320)
242 KB
242 KB JPG
Here's your nvidia gpu bro.
>>
>you're... le poor
why do amerimutts think this is a legitimate argument outside of their kaiji-tier dystopia?
>>
>>87462451
Cope
>>
I just wish they fixed their fucking drivers
I dont care about gayming but somethings that a GPU can help (like AI, video editing) usually is more worth with Nvidia since their drivers are actually supported
AMD shouild better up their drivers and increase on marketing because I've never met someone with an AMD GPU (even if for gayming they are more worth it)
>>
>>87443595
>6 hours a day
You mean like a NEET?
>>
>>87463372
stop over sleeping and you'll have time to game
>>
>>87463250
Suddenly everyone on the planet is a video editor.
>>
>>87442495
I did hear that top RDNA3 is intended to be 6x a 5700XT.
That lines up with the ALU count being 6x higher, process, clocks and uarch get the scaling to that level.
>>
>>87464864
It's also 6x cost
>>
>>87464879
Pay up goy.
>>
>>87453812
>You know the $280 rx 6600 is 2.5x faster than your rx 480, right?
In what jewish multiverse is this true?
>>
>>87443523
It is a problem for thermal management. The main reason why 300W+ TDP SKUs are unmarketable deadweight.
You can't get around the laws of physics. Nvidia and AMD are just vainly trying to keep up the illusion of leapfrogging each generation.
>>
>>87456428
>rdna1 is already 3 years old
Good it was just as bad as radeon 7
Hope it gets memory holed
>>87464879
>It's also 6x cost
And who's fault is that
Fucking miners greedy amd tsmc
>>
File: fren-thank-you-king.jpg (69 KB, 1071x561)
69 KB
69 KB JPG
>>87443197
Thank you fren
>>
>>87462890
not a cope + it is your problem since during the climate disaster hundreds of millions of thirdiechads are going to enter your cunt without permission you fucking suburbanoid subhuman
>>
>>87464879
rarted
>>
File: generational gpus.png (131 KB, 970x546)
131 KB
131 KB PNG
>>87465103
This one.
Multiple reviewers show the same thing.
>>
>>87468405
Reality doesn't matter
>>
>>87466851
Nice projection, coping turd worlder.
>>
hi guys, TDPlet here
should i buy one of those intel "k" cpus marketed as 125w or a non k marketed as 65w
will the k result in higher temperatures?
>>
I'm just waiting for the rt performance
>>
>>87468405
sorry but i only believe the stats from the nvidia and amd official websites
>>
>>87465103
It also costs double the price so why bother. The card is only good for 1080p anyway. Everything else it shits the bed.
>>
>>87470120
>costs double
>gets 2.5x the performance for the same power draw
You're actually arguing to get a used card that is worse perf:price and perf:watt. You're an idiot.
>>
>>87442527
>7900/7950
>2~2.2 times the SPs of 6900XT
>Arch+Node gains
>higher freq. mem
I mean its not impossible.
>>
>>87469947
Yes. Buy the 65w
>>
>>87442495
I don't like the lower L3 cache and VRAM for the mid-low end. Same with the bus width.

Also the MSRP is getting out of hand.
>>
>>87468405
That's 2x not 2.5x, also the cheapest RX 6600 where I live is $100 more than what I paid for my RX 480 back in early 2017. To be fair the exchange rate was slightly better back then, but only by about 10%.
>>
>>87442495
Wake me up when there's a 1440p capable card at <$300, with more VRAM than an RX 480.
It won't happen and I'll stay salty about it. Might buy something used if current gen cards get dirt cheap, like <$100 in the coming couple of years. All I need is something capable of running Elden Ring at 1440p, but I don't want to overpay for something that'll be outdated really fast due to lack of VRAM (Ampere) or RT performance (RDNA2).
>>
>>87471799
They are really gimping sub $400/300 GPUs on purpose.
>NVIDIA/AMD : "What are you gonna do about it? not buy a GPU?"
>>
I'm not buying a graphics card using over 250w
>>
File: you.gif (350 KB, 368x450)
350 KB
350 KB GIF
>>87456103
>2.5x performance is 250%, which is +150% faster.
>2.5x faster is +250%, which is 350% baseline performance.
>"Faster" is comparative, which means you are comparing it to the baseline of 100% in this case.
>>
>>87471696
Nope it's more than 2x.
Go back to elementary school.
>>
>>87472102
A very convincing argument anon. Very cool.
>>
>>87469218
gpt-tier response
>>
>>0087472373
>replies with more projection
No more (You)s for NPCs.
>>
>>87442495
If you round up these prices and change the $ to a € you get what these will cost in Europe.
>>
>>87472578
>look look i found a new reddit word to spam
>>
>>0087472634
>NPC IESLB
I accept your concession.
>>
File: 1627241168765.png (227 KB, 1571x910)
227 KB
227 KB PNG
>>87473450
>more buzzwords
>tries to force a "concession" like a debatenigger
>calls others npcs
why do amerimutts lack self-awareness? is there something in the burgers that causes this degree of retardation?
>>
File: 1601809010821.jpg (11 KB, 300x207)
11 KB
11 KB JPG
>>0087473495
>NPC IESLB 2: electric boogaloo
>>
>>87473542
kys dbsshitter
>>
>/g/ posters struggle with basic maths
Impressive. This is the power of CS.
>>
>>87473566
Vacate discorddit
>>
>>87442495
If these are chiplet GPUs then neat.
If these are crossfired GPUs on the same pcb then lol, lmao even.
>>
File: 1635213764518.gif (2.77 MB, 1280x720)
2.77 MB
2.77 MB GIF
>>87460981
Gallium is way too rare for widespread use like silicon is.
>>
>>87473575
We had the same issue with Ryzen 7000 and what Lisa Su showed with workload speed vs a Ryzen 9 5950X with the issue of how the terminology of faster vs quicker and units of time. Same issue here. Kind of sad for self professed tech followers, but it's not surprising since a lot of /g/ is underage.
>>
>>87474673
Chiplet
>>
>>87452476
https://www.youtube.com/watch?v=r3y_cQww-cU
>>
>>87474673
Proper chiplet, single device.
>>
>>87460981
Really? Source? I feel like this would be big news.
>>
>>87442495
I don't care in 1660 soc and thin wallet



Delete Post: [File Only] Style:
[Disable Mobile View / Use Desktop Site]

[Enable Mobile View / Use Mobile Site]

All trademarks and copyrights on this page are owned by their respective parties. Images uploaded are the responsibility of the Poster. Comments are owned by the Poster.