Official AMD Radeon RX 7000 Series (RDNA 3) Thread

[pax]

Nvidia also dropped the power for the 4070 and 4080. Sounds like they will now all be in the same ballpark.

Mind you Nvidia is using a smaller node

https://wccftech.com/nvidia-geforce-rtx-4080-at-320w-rtx-4070-at-285w-tbp-tamer-powerful-gaming-gpus/

NVIDIA seems to have modified its internal TBP figures for its upcoming GeForce RTX 4080 & RTX 4070 graphics cards.

NVIDIA GeForce RTX 4080 & RTX 4070 Updated To Lower TBP Specs, Now Rumored at 320W & 285W

[pax]

https://twitter.com/SkyJuice60/status/1558130277360185344

Some interesting info. Seems AMD might be trying to pull another 4870 here. 300mm2 for the gcd and 37 mm2 for the mcd's. Thats very small and could mean AMD pulls a perf to mm2 win. Also means cheap to make so maybe better MSRP.

https://www.angstronomics.com/p/amds-rdna-3-graphics

What AMD has officially detailed so far about RDNA 3 is yet another significant increase in performance per watt over RDNA 2, with contributions from process node and microarchitectural design choices. However, the design philosophy of gfx11 is all about area, area, area. What is the best way to achieve the performance target with minimal area? The rearchitected Compute Unit and Optimized Graphics Pipeline changes are mostly about trimming the fat in pursuit of the lowest area and cost (example: halving relative FP64 rate to 1/32). As a result of this focus, PPA is significantly increased. In fact, at the same node, an RDNA 3 WGP is slightly smaller in area than an RDNA 2 WGP, despite packing double the ALUs.

Navi 31
gfx1100 (Plum Bonito)

Chiplet - 1x GCD + 6x MCD (0-hi or 1-hi)

48 WGP (96 legacy CUs, 12288 ALUs)

6 Shader Engines / 12 Shader Arrays

Infinity Cache 96MB (0-hi), 192MB (1-hi)

384-bit GDDR6

GCD on TSMC N5, ~308 mm²

MCD on TSMC N6, ~37.5 mm²

[Nunz]

Perf to mm2? Some new metric we're going by to make AMD feel good or something

[pax]

It would just mean the engineering is ahead of Nvidias. Nvidia can go big and slightly win top end like it did during the GTX 280 vs HD 4870 but AMD had a smaller gpu at a price that was half that of the GTX 280.

[pax]

Over the top speculation but would be fun to see get done.

https://twitter.com/3DCenter_org/status/1559193445054304256

Return of the dual gpu on one board?

https://www.3dcenter.org/news/news-des-1314-august-2022

Navi 31 Dual Navi 32 Dual Navi 31

6 shader engines, 48 WGP with 12,288 FP32 units

6 shader engines, 60 WGP with 15'360 FP32 units

12 shader engines, 96 WGP with 24,576 FP32 units

memory interface 384 bits 512 bits 768 bits

[0091/2]

Over the top speculation but would be fun to see get done.

https://twitter.com/3DCenter_org/status/1559193445054304256

Return of the dual gpu on one board?

https://www.3dcenter.org/news/news-des-1314-august-2022

How would you keep it cool? More importantly how do you convince developers to support it?

[Trunks0]

How would you keep it cool? More importantly how do you convince developers to support it?

If it's done via MCM the developers don't have to do anything. As it will operate like a monolithic design.

Cooling it however is another question. 4 Slot cooler collaboration with ICEGIANT?

[bill dennison]

How would you keep it cool? More importantly how do you convince developers to support it?

two GCD's on one chip not a duel crossfire card

it would be a big ass chip but still most likely aio water cooling

[Mangler]

Two gpus seen as one is probably not that far off, isn't apple already doing that with the m1 ultra?

[bill dennison]

Two gpus seen as one is probably not that far off, isn't apple already doing that with the m1 ultra?

AMD was going to do it this time then hit a snafu and went with only one big GCD

they may have fixed it by now
they said they would before RDNA 4

if so most likely won't see them till Q1 next year

i can wait

[pax]

Or its the mid gen refresh in late 2023... would make for a hell of a refresh if true.

Wont be cheap.

[bill dennison]

Or its the mid gen refresh in late 2023... would make for a hell of a refresh if true.

Wont be cheap.

Q1 as a ad on card or Q2 as a mid gen refresh

i think RDNA 4 will be Q3 2024 if late 23 might as well wait on RDNA 4

[pax]

Would be too pricey for most as a mid gen refresh true. A single rdna4 gpu could closely match it if its on 3nm. But they are already on a long 2 year cadence for new gpu gens and rdna4 could be pushed even later at the rate things are slowing.

[bill dennison]

Would be too pricey for most as a mid gen refresh true. A single rdna4 gpu could closely match it if its on 3nm. But they are already on a long 2 year cadence for new gpu gens and rdna4 could be pushed even later at the rate things are slowing.

covid mostly

........

TSMC Phoenix

The company expected to have more than 2,000 employees working at the Arizona plant by 2023.

https://www.azcentral.com/story/news/local/phoenix/2022/06/03/see-taiwan-semiconductors-north-phoenix-office/7498673001/

in 2024 it will be in full swing

i think AMD will go back to a 12 to 18 month cadence

[pax]

Hope so.

[Trunks0]

Two gpus seen as one is probably not that far off, isn't apple already doing that with the m1 ultra?

Yup. AMD has already done it for CPU's with Threadripper and Epyc. They also released the CDNA2 Radeon Instinct MI200 that is MCM.

It wouldn't be a leap for them to do it.

[Nunz]

I don't think it'll be difficult for DX12, but for some reason I have a feeling MCM is going to be an issue with older APIs. I understand the driver will have it recognized as a single GPU, but it just can't be that simple to just *work*

[pax]

Question is, is the dual Navi 31 gpu simply 2 GCD's on one package or two packages each with their own set of MCD's and on board memory banks. Other than latencies does it make that big a diff to the driver? R295 seemed to run well under that latter format.

But hope its the former as it would mean one package and one onboard memory bank so overall cheaper to make and sell.

Hope we get a view of the package when they present it and we see a missing spot like we did for ryzen long before they came out with the dual cpu die.

[bill dennison]

Question is, is the dual Navi 31 gpu simply 2 GCD's on one package or two packages each with their own set of MCD's and on board memory banks. Other than latencies does it make that big a diff to the driver? R295 seemed to run well under that latter format.

But hope its the former as it would mean one package and one onboard memory bank so overall cheaper to make and sell.

Hope we get a view of the package when they present it and we see a missing spot like we did for ryzen long before they came out with the dual cpu die.

don't think it will be CFX .

[pax]

https://wccftech.com/roundup/amd-radeon-rx-7800-xt/

650$ for the 7800xt but 999 for the 7900. Like last time but doubt many will pay 350$ for that extra 10%.

Course mSrp is well... Mostly S...

[pax]

don't think it will be CFX .

Always thought it was easier to do a driver for one board than 2 with the pcie bus being the big issue to manage. Read a bit on the R295 back then and some seemed to like it... course I could be way off base here its been a while.

[bill dennison]

Always thought it was easier to do a driver for one board than 2 with the pcie bus being the big issue to manage. Read a bit on the R295 back then and some seemed to like it... course I could be way off base here its been a while.

i had a HD 5970 after that they got crazy with prices on the duel cards

https://www.techpowerup.com/gpu-specs/radeon-hd-5970.c254

but it is the same Crossfire as having two cards and all the same problems

and two cards in two slots are easer to power and cool with water

[Android1]

AMD Teases Next-Gen RDNA3 Graphics Card: Claims to Repeat 50% Perf/Watt Gain

Will AMD go for raw power, or will they try to compete with NVIDIA on TDP?

AMD in its Ryzen 7000 launch event teased its next-generation Radeon graphics card based on the RDNA3 graphics architecture. Built on an advanced process node just like "Zen 4," AMD is hoping to repeat the magic of the RX 6000 series, by achieving a 50% performance-per-Watt gain over the previous generation. which allows it either to build some really efficient GPUs, or consume the power headroom to offer significantly higher performance at power levels similar as the current-gen.

AMD's teaser included a brief look at the air-cooled RDNA3 flagship reference-design, and it looks stunning. The company showed off a live demo of the card playing "Lies of P," a AAA gaming title that made waves at Gamescom for its visuals. The game was shown playing on an RDNA3 graphics card running on a machine with a Ryzen 9 7950X processor at 4K, with extreme settings. AMD CEO Dr Lisa Su confirmed a 2022 launch for RDNA3.

Source: techPowerUp!

[pax]

https://twitter.com/VideoCardz/status/1564675593995788296

VideoCardz.com
@VideoCardz
(PR) JPR: Q2’22 GPU shipment decreased -14.9% from last quarter: AMD: -7.6%, NVIDIA: -25.7%

So if either were not selling to miners would seem AMD's share of sales to gamers went up quite a bit. My bet is its because of Nvidia pricing.

https://videocardz.com/press-release/jpr-q222-gpu-shipment-decreased-14-9-from-last-quarter-amd-7-6-nvidia-25-7

[Nunz]

NV's pricing is better than ever right now. I think the percentages dropped harder due to NV selling more, so with the reduction of buyers, they have more to lose. I can't remember if it was confirmed whether NV was selling to miners or not - I do remember something about that, but can't remember if it was confirmed NV or if it was one of the AIBs.

[acroig]

NV's pricing is better than ever right now. I think the percentages dropped harder due to NV selling more, so with the reduction of buyers, they have more to lose. I can't remember if it was confirmed whether NV was selling to miners or not - I do remember something about that, but can't remember if it was confirmed NV or if it was one of the AIBs.

IIRC nV sold FE cards in volume directly to miners.

[Android1]

IIRC nV sold FE cards in volume directly to miners.

[pax]

https://wccftech.com/amd-rdna-3-navi-3x-gpus-feature-double-the-cache-per-compute-unit-shader-array/

AMD has listed down the latest information regarding the Cache sizes of their upcoming RDNA 3 "Navi 3X" GPUs within Linux patches.

Coming to the details, the AMD RDNA 3 (GFX11) GPU lineup will feature double the L0 vector cache for each Compute Unit or CU and also double the GL1 data cache (RDNA L1 cache for each Shader Array or SA. As per the new information, the Vector Register File per SIMD will increase to 192KB vs 128KB on RDNA 2, the L0 Vector/Texture cache will increase from 16 KB to 32 KB per CU, the GPU L1 Data Cache per Shader Array will increase from 128 KB to 256 KB while the L2 Data Cache will remain the same as RDNA 2.

Coelacanth-Dream also states that all RDNA 3 "Navi 3X" GPUs come with VODP (Dual-Issue Wave32) instructions, WMMA (Wave Matrix Multiply-Accumulate) support and the performance per WGP has been improved a lot. The increased GL1 cache is said to improve pixel processing performance and is amongst the many changes that AMD is bringing within its RDNA 3 Navi 3x GPU family.

Big cache changes usually means good perf improvement...

[pax]

https://videocardz.com/newz/amd-radeon-rx-7900xt-pcb-layout-leaks-out-with-navi-31-gpu-and-three-8-pin-power-connectors

3 8 pin connectors on the big one 7900xt.

[SSXeon]

https://videocardz.com/newz/amd-rade...wer-connectors

3 8 pin connectors on the big one 7900xt.


*snip*

I believe it is rumored to be 375-400w for the 7900XT now (I believe tom said this) so it makes sense. The GCD is so small!

[bill dennison]

450 watts of power plug and 75 watts from the slot

don't think the 7900XT will need that but the 7950XT might

[Mangler]

I hate how amd is so good at keeping things under wraps compared to nvidia, it's no fun at all.

[bill dennison]

I hate how amd is so good at keeping things under wraps compared to nvidia, it's no fun at all.

Lisa Su sends leakers to the mines of Kessel

[acroig]

Lisa Su sends leakers to the mines of Kessel

or Rura Penthe?

[pax]

Lisa Su sends leakers to the mines of Kessel

Leakers I follow do the run in ~14 parsecs...

[bill dennison]

Leakers I follow do the run in ~14 parsecs...

but she has them wanted in 12 systems

[pax]

[Dave]

Do we have a rumored release time for the 7900XT?

[acroig]

Do we have a rumored release time for the 7900XT?

Not yet.

[pax]

https://twitter.com/9550pro/status/1571834170900623360

4 ghz gpu? Thatd be and have to be cool.