AMD Radeon RX 7000 RDNA3 To Launch Early December

[btarunr]

AMD's next-generation Radeon RX 7000-series graphics cards based on the RDNA3 graphics architecture, are expected to launch in early-December 2022, according to greymon55, a reliable source with AMD leaks. The cards will be unveiled at a media event to be held on November 3, 2022, with market availability following a month after (between 1st to 5th December). The company is expected to take a top-down product-stack release cycle similar to that of NVIDIA, with the release of two of its top SKUs, the Radeon RX 7900 XTX and the RX 7900 XT. Both these cards are based on the 5 nm Navi 31 MCM GPU. This will be AMD's first client-graphics MCM GPU with more than one logic die. The company has a decade of experience with MCMs, but past generations have been one logic die surrounded with on-package HBM. Navi 31 has on-package logic chiplets, but discrete GDDR6 memory, like most other GPUs in the market today. It's rumored that the company is targeting a 100% performance uplift over the previous-generation, which means team-red is on the prowl to compete with NVIDIA's fastest SKUs, including the RTX 4090 and upcoming RTX 4080.



View at TechPowerUp Main Site | Source

 

[ratirt]

I only hope the latency won't be huge and due to the 2 MCMs. I guess AMD has this issue in the back of their heads and they have addressed the potential latency.
Are they really aiming to double the performance? wonder if they can pull this off.

 

[ixi]

Big difference between xtx and xt. Unexpected. No gddr6x, hmm... there should be visible difference in 2k and 4k against npeedia.

 

[Hyderz]

amd has the advantage to put in a good price because so far... nvidia's msrp is terrible

 

[wolf]

Keen to see these hit the market for many reasons

• assess outright performance
• assess RT performance / impact
• the pricing.
• see if it affects pricing of other products
• See how big and power hungry they are
• see if FSR evolves and is accelerated by RDNA3
• get a sense of stock levels and availability

Lots to look forward to, even if you don't want to buy one.

 

[uftfa]

I only hope the latency won't be huge and due to the 2 MCMs. I guess AMD has this issue in the back of their heads and they have addressed the potential latency.
Are they really aiming to double the performance? wonder if they can pull this off.

I don't think latency will be an issue, but frame pacing will almost certainly be affected by MCM design. Even if they get close to 4090 average framerates, the 1% lows are almost sure to suffer because of MCM.

On the bright side, the total die size is smaller than 6900XT (with better yields because the die area is broken into smaller chiplets), so I am hoping for the same $999 MSRP on the 7900XT. AMD could realistically price it lower, but with nvidia's pricing so high up, I don't think AMD will have any incentive to do so.

My realistic guess is

7900XT = $999 to undercut the 4080, while outperforming it in rasterization.
7900XTX = $1299 to undercut the 4090, while being about 5% off in average rasterization performance (similar to 6900XT vs 3090).

 

[seventy]

I only hope the latency won't be huge and due to the 2 MCMs. I guess AMD has this issue in the back of their heads and they have addressed the potential latency.
Are they really aiming to double the performance? wonder if they can pull this off.

It's very unlikely that latency would be a problem. If they implement cache properly, the difference shouldn't be even noticeable.

From the theoretical standpoint, GPU's were always designed around high latency access to memory.
Instead of going the CPU route and optimizing for latency, GPU's just hide it by increasing the bandwidth and amount of threads.
Different tradeoffs for different workloads.

 

[MarsM4N]

They better drop the mid range cards in January. If not folks waste their christmas bucks on something else.

I only hope the latency won't be huge and due to the 2 MCMs. I guess AMD has this issue in the back of their heads and they have addressed the potential latency.
Are they really aiming to double the performance? wonder if they can pull this off.

Guess that's why they'll use 3D V-Cache, to reduce latency & boost performance similar to the 5800X3D. And we all know how this thing rocks.

AMD RDNA 3 Navi 31 Flagship GPU Allegedly Rocks Six MCDs With Up To 384 MB Infinity Cache With 3D V-Cache Packaging

AMD's Navi 31 RDNA 3 Flagship GPU is expected to rock up to 384 MB of Infinity Cache with 3D V-Cache packaging technology.

wccftech.com

 

[beedoo]

Gah! Was hoping to see mention of a 7950XT(X). Can't bring myself to lay any money down until these show up. AMD! If they don't show up, you might miss out on my two sales this iteration

WRT to Ray Tracing; it will be interesting to see the results this time around. Just how seriously is AMD taking RT, and... if they're 20% slower than the 4090 at RT, will people say AMD still can't do RT, or similarly, if it's faster than the 4090 (which it won't be, I know), will people say Nvidia can't do RT - or that Nvidia's RT is fine (bit of a reach with that last bit I know - it's been an afternoon of analysis!!!).

 

[ZoneDymo]

Gah! Was hoping to see mention of a 7950XT(X). Can't bring myself to lay any money down until these show up. AMD! If they don't show up, you might miss out on my two sales this iteration

WRT to Ray Tracing; it will be interesting to see the results this time around. Just how seriously is AMD taking RT, and... if they're 20% slower than the 4090 at RT, will people say AMD still can't do RT, or similarly, if it's faster than the 4090 (which it won't be, I know), will people say Nvidia can't do RT - or that Nvidia's RT is fine (bit of a reach with that last bit I know - it's been an afternoon of analysis!!!).

if it does 120 and drops to 80 with RT on vs Nvidia 120 dropping to 100 with RT on, I dont think it will be such a big deal for many as long as the price difference is correct

also no 7950xtx will show up before the RTX4090ti

 

[ARF]

Gah! Was hoping to see mention of a 7950XT(X).

If RX 7950 will be launched, it will be much later and a refresh. The **50 moniker is always refreshes.

I don't think AMD will have any incentive to do so.

AMD has because very few users buy Radeons and the market share is on the decline.

 

[Luminescent]

Let's not kid ourselves, if AMD has a faster card than 4090 it will be more expensive than Nvidia, just look at recently launched ryzen cpu's, Intel is now the underdog.
As i see it now Nvidia dedicated a lot of silicon for raytracing and they could make up rasterization difference with DLSS 3, AMD could have a lot of silicon dedicated to rasterization but lose in raytracing.
Both AMD and Nvidia sponsor a game or two to showcase new technologies or performance, that is the most you will get to see what a 450W GPU can do until another GPU generation launches but by then driver support is dead.

 

[N/A]

Why would it be faster than 4090, it's half the size 308mm2 vs 608 on the same node and nvidia is using new high density cell libraries.

 

[uftfa]

Why would it be faster than 4090, it's half the size 308mm2 vs 608 on the same node and nvidia is using new high density cell libraries.

Not apples to apples comparison. The 308 mm2 doesn't include AMD's memory controller die (MCD), which is included in nvidia's 608 monolithic mm2. AMD's equivalent number is 533mm2, but they presumably have less area dedicated to RT/Tensor accelerators than nvidia.

 

[BrainCruser]

I don't think latency will be an issue, but frame pacing will almost certainly be affected by MCM design. Even if they get close to 4090 average framerates, the 1% lows are almost sure to suffer because of MCM.

On the bright side, the total die size is smaller than 6900XT (with better yields because the die area is broken into smaller chiplets), so I am hoping for the same $999 MSRP on the 7900XT. AMD could realistically price it lower, but with nvidia's pricing so high up, I don't think AMD will have any incentive to do so.

My realistic guess is

7900XT = $999 to undercut the 4080, while outperforming it in rasterization.
7900XTX = $1299 to undercut the 4090, while being about 5% off in average rasterization performance (similar to 6900XT vs 3090).

There is no reason for 1% lows to suffer. It isn't an MCM in the traditional sense where there are two GPU cores and separate memories working on separate frames. This is multiple dies working together. Think zen 2 chiplets vs two-socket servers. In two socket servers the NUMA-ness of the system is a lot more exposed than with zen 2. Especially since AMD is only putting one compute die, so this is a ryzen 3700X(1 compute 1 io die), not a 3950X.

The performance of the GPU will be similar or identical to existing designs. There will be different tradeoffs, so games will need to be re-optimized for the new architecture, since the bottlenecks are in a different place. So expect performance in existing games to be worse, but in newer games that have access to RDNA3 GPUs to improve. Similar to other generations of GPUs, but more impactful.

 

[Imouto]

Unless the hardware guys left everything tied up (bondage) tightly I expect the software guys to screw up royally.

 

[RedelZaVedno]

Based on the specs alone RX 7900 XT should eat 4080 for breakfast. It will all come down to pricing now. Price it $899 and they fly off the shelves like hot cakes, but I have a bad feeling that Lisa will accept Jensen's price hiking game and charge $1,199 USD maybe minus 50 bucks for this thing.

 

[Daven]

Why would it be faster than 4090, it's half the size 308mm2 vs 608 on the same node and nvidia is using new high density cell libraries.

You have to add up all the chips on the package for the total size.

 

[DeathtoGnomes]

Of course they need to launch just before Christmas, would have been better on Black Friday.

 

[TheinsanegamerN]

12288 cores. Jebus. Dis gon be fun.

Big difference between xtx and xt. Unexpected. No gddr6x, hmm... there should be visible difference in 2k and 4k against npeedia.

GDDR6X is nvidia-micron exclusive.

 

[Punkenjoy]

You have to add up all the chips on the package for the total size.
View attachment 267950

And also, the claim is that AMD removed all the dark silicon they could for RNDA3.

 

[The Quim Reaper]

Well, at least the specification gap between the two cards looks good, unlike Nvidia's absurd, greed driven, 4090/4080 cut down spec.

 

[Dimitriman]

AMD just call them 7900XT and 7800XT ffs, stop doing this dumb naming..

 

[spnidel]

yeah guys I agree, there's no way the 6800 xt will be faster and more efficient than a 2080 ti

 

[mechtech]

Hmmm so about 14% more shaders in xtx and a bit
More memory bus.

will wait for hopefully affordable 7600/xt