More Radeon RX 7000 Series Graphics Cards Spotted in ROCm 5.6

[GFreeman]

A bunch of unreleased AMD Radeon RX 7000 series graphics card have been spotted in ROCm 5.6 pull request, including the Radeon RX 7950 XTX, 7950 XT, 7800 XT, 7700 XT, 7600 XT, and 7500 XT. AMD has not yet launched its mainstream Radeon RX 7000 graphics cards, but according to the latest pull request, there are several unreleased graphics cards in for both high-end and mainstream segments. While the pull request has been removed from GitHub, it has been saved on Reddit. So far, it appears that AMD's RDNA 3 Radeon 7000 series lineup will be based on just three GPUs, Navi 33, Navi 32, and the Navi 31.

According to the list, we can expect a high-end refresh with Radeon RX 7950 XTX/XT version, also based on Navi 31 GPU. The list also shows that the Radeon RX 7800 series will be the only one based on the Navi 32 GPU, at least for now, while the Navi 33 GPU should cover the entire mainstream lineup, including the Radeon RX 7700 series, Radeon RX 7600 series, and the Radeon RX 7500 series. The list only includes XT versions, while non-XT should show up later, as it was the case with the Radeon RX 6000 series graphics cards. AMD's President and CEO, Dr. Lisa Su, already confirmed during Q1 2023 earnings call that mainstream Radeon RX 7000 series GPUs based on RDNA 3 architecture will launch during this quarter, and earlier rumors suggest we might see them at Computex 2023.

• Radeon RX 7950 XTX RDNA3 gfx1100 (Navi 31)
• Radeon RX 7950 XT RDNA3 gfx1100 (Navi 31)
• Radeon RX 7900 XTX RDNA3 gfx1100 (Navi 31)
• Radeon RX 7900 XT RDNA3 gfx1100 (Navi 31)
• Radeon RX 7800 XT RDNA3 gfx1101 (Navi 32)
• Radeon RX 7700 XT RDNA3 gfx1102 (Navi 33)
• Radeon RX 7600 XT RDNA3 gfx1102 (Navi 33)
• Radeon RX 7500 XT RDNA3 gfx1102 (Navi 33)
• Radeon RX 7600M XT RDNA3 gfx1102 (Navi 33)
• Radeon RX 7600M RDNA3 gfx1102 (Navi 33)
• Radeon RX 7700S RDNA3 gfx1102 (Navi 33)
• Radeon RX 7600S RDNA3 gfx1102 (Navi 33)

View at TechPowerUp Main Site | Source

 

[Daven]

I think the third bullet should be an XTX not an XT|.

 

[btk2k2]

If that is actually how the stack plays out the only way I can see it being remotely reasonable is if N33 can actually keep scaling well with more and more power otherwise a 7700XT based on N33 will not at all be faster than the 6750XT currently is.

To my mind the stack would need to be as follows.

7800XT - N32 - 16GB Vram - 6950XT performance - $600 at most.
7700XT - N33 - 16GB Vram - 6800 performance - $400 at most. (and I would be very very very surprised if N33 can reach this level of performance given the laptop N33 testing floating around).
7600XT - N33 - 8GB Vram - 6700 performance - $280 at most.
7500XT - N33 - 6GB Vram - 6600 performance - $170 at most.

 

[GFreeman]

I think the third bullet should be an XTX not an XT|.

Fixed... thanks!

 

[Darksword]

Any card $200 and up that is actually targeted for "gamers" should have 12GB VRAM minimum. $300 and up should have 16GB minimum.

That's just the world we live in now.

 

[Space Lynx]

7950 xtx in spring 2024 on 3nm node or 4nm... and I bet it beats a 4090 in raster games but not ray tracing... $1199?

 

[Icon Charlie]

If that is actually how the stack plays out the only way I can see it being remotely reasonable is if N33 can actually keep scaling well with more and more power otherwise a 7700XT based on N33 will not at all be faster than the 6750XT currently is.

To my mind the stack would need to be as follows.

7800XT - N32 - 16GB Vram - 6950XT performance - $600 at most.
7700XT - N33 - 16GB Vram - 6800 performance - $400 at most. (and I would be very very very surprised if N33 can reach this level of performance given the laptop N33 testing floating around).
7600XT - N33 - 8GB Vram - 6700 performance - $280 at most.
7500XT - N33 - 6GB Vram - 6600 performance - $170 at most.

Well, the pricing... yea... not even going to go there. Lets just say I'm waiting for the 6800 to go down a little more to pick one up. IMHO they will be doing a Nvidia.

But let see what happens.

 

[Chrispy_]

A single, solitary Navi 32 product? That's not how GPU manufacture works;

Any silicon wafer coming out of lithography will have some defective but functional parts. For a 7800XT to exist, there needs to be a lower model that uses dies with the defects fused off - It has always been this way, ever since dGPUs were a thing...

 

[sethmatrix7]

Damn I'm planning on doing a 7900 xtx or xt in the next month and looks like i'll be waiting

 

[Nostras]

Any card $200 and up that is actually targeted for "gamers" should have 12GB VRAM minimum. $300 and up should have 16GB minimum.

That's just the world we live in now.

Yea, no. That's some wishful thinking. If an 8GB card goes over 300$ the card is DOA.
300$ is acceptable if performance is seriously improved, but no more.
I agree with @btk2k2

 

[Icon Charlie]

Damn I'm planning on doing a 7900 xtx or xt in the next month and looks like i'll be waiting

IMHO I would just wait. It's getting close to summer time (then back to school) and it's one of the reasons material is launched regardless of it being a widget or a video card. Save your money as best as you can. Judging by how our economy is and the obscene pricing for poor quality components of recent years... I believe there will be components on the shelves at launch.

 

[Mr. Perfect]

Any card $200 and up that is actually targeted for "gamers" should have 12GB VRAM minimum. $300 and up should have 16GB minimum.

That's just the world we live in now.

Man, I wish. VRAM has been stagnant for a little to long and these new AAA games are eating it up. I suppose 12GB is too much to hope for though. Even 10GB would be a lifeline.

 

[MrDweezil]

If that is actually how the stack plays out the only way I can see it being remotely reasonable is if N33 can actually keep scaling well with more and more power otherwise a 7700XT based on N33 will not at all be faster than the 6750XT currently is.

To my mind the stack would need to be as follows.

7800XT - N32 - 16GB Vram - 6950XT performance - $600 at most.
7700XT - N33 - 16GB Vram - 6800 performance - $400 at most. (and I would be very very very surprised if N33 can reach this level of performance given the laptop N33 testing floating around).
7600XT - N33 - 8GB Vram - 6700 performance - $280 at most.
7500XT - N33 - 6GB Vram - 6600 performance - $170 at most.

I'd be surprised if they really left a $200 gap between the 7800XT and the 7700XT.

 

[ToTTenTranz]

I'm very surprised about the 7700XT using N33. It's supposedly a tiny 200mm^2 chip built on the "ancient" N6 process that shouldn't be able to scale up to Navi 22 6700XT performance.
I think most people were expecting the 7700XT to use a N32 with only 3 MCDs instead of 4, for a total 12GB VRAM.

Maybe AMD found a way to make N33 clock up to 3GHz, even if it's coming at a substantial power consumption cost (250W or more?).


AMD being already prepared to launch the 7950XTX / XT this soon probably means they might have found a way to significantly increase the core clocks and use 24Gbps memory (current Navi 31 uses 20Gbps chips).
The last time we saw something like this was over 10 years ago with the Radeon HD7970 (925Mhz core, 5.5Gbps GDDR5) -> HD7970 GHz Edition (1050Mhz core, 6Gbps GDDR5), where it took them around 5 months between both GPU releases and the former model was quickly phased out.

I'd be surprised if they really left a $200 gap between the 7800XT and the 7700XT.

The difference between the 6800XT and 6700XT was $170. This wouldn't be that much more.
If the N33 chip can indeed scale so much up in frequency (and be paired up with 16GB VRAM) and N32 yields are good, there's no reason to release a $500 cut-down N32 with "just" 12GB VRAM. For the $500 price point they've been mentioning the N21 solutions with 16GB, which might be quite cheap to make nowadays on the old N7 process.

 

[btk2k2]

I'd be surprised if they really left a $200 gap between the 7800XT and the 7700XT.

They would put a 7800 in there probably around the $500 mark.

 

[dragontamer5788]

A single, solitary Navi 32 product? That's not how GPU manufacture works;

Any silicon wafer coming out of lithography will have some defective but functional parts. For a 7800XT to exist, there needs to be a lower model that uses dies with the defects fused off - It has always been this way, ever since dGPUs were a thing...

While your statement is true for maybe 90% of cases, I can think of some notable exceptions. PS3 for example, was designed as 8x SPE on the die. But it was sold as a 7x SPE system, under the assumption that the "perfect dies" are rare enough to not be worth selling.

A similar case could exist for Navi 32? Maybe the AMD marketing / executives decided that its not worth it to sell a perfect die right now, or perhaps they plan to sell a perfect die maybe next year. The overall plan would be to have a relatively slow production rollout for Navi32, save the perfect dies this year, and then sell them next year.

Or, if they go the PS3 route, they just sell the perfect dies as if they were cut-down dies.

 

[Chrispy_]

While your statement is true for maybe 90% of cases, I can think of some notable exceptions. PS3 for example, was designed as 8x SPE on the die. But it was sold as a 7x SPE system, under the assumption that the "perfect dies" are rare enough to not be worth selling.

A similar case could exist for Navi 32? Maybe the AMD marketing / executives decided that its not worth it to sell a perfect die right now, or perhaps they plan to sell a perfect die maybe next year. The overall plan would be to have a relatively slow production rollout for Navi32, save the perfect dies this year, and then sell them next year.

Or, if they go the PS3 route, they just sell the perfect dies as if they were cut-down dies.

Not comparable. The PS3 is a contractual obligation to produce a single product where the cut-down die is chosen as the product for cost/yield benefits. For their own product, AMD will extract the maximum value they can out of any silicon, so perfect dies are the highest-profit dies that AMD will be most eager to sell fast before the market is saturated with competition to drive down sale prices and demand.

AMD could be stockpiling perfect dies for an eventual 7850XTX or Radeon Pro of some description, but stockpiling premium-cost silicon for another whole cycle isn't a good strategy for very long because it's high-investment and ages/depreciates rapidly. When I say another whole cycle, the Pro-series cards are typically half a cycle behind consumer GPUs and the x50 refresh is usually 9-12 months after the initial release, which feels like too long to sit on silicon to me.

Now I say that though, Nvidia *had* that issue of sitting on lots of GA104 silicon, but I feel like that's an overproduction issue driven by high ETH-mining demand and then hampered by board-vendor manufacturing and distribution issues during the height of the pandemic; Those conditions ripe for overproduction don't exist today - in fact TSMC constraint, low demand, and very high production costs at TSMC are the exact opposite conditions to uncontested, cheap Samsung production of GA104 in a period of massive GPU demand.#

I could still be wrong, but empirical data suggests the most likely answer is that the leaked list of dies in ROCm 5.6 has a typo

 

[dragontamer5788]

I could still be wrong, but empirical data suggests the most likely answer is that the leaked list of dies in ROCm 5.6 has a typo

Fair enough. That's a lot of 33 dies and not a lot of 32. Its more believable that someone mistyped 33 instead of 32 on a number of those SKUs.

 

[mama]

7950 xtx in spring 2024 on 3nm node or 4nm... and I bet it beats a 4090 in raster games but not ray tracing... $1199?

I'd hit that!

 

[eidairaman1]

I think the third bullet should be an XTX not an XT|.

And the ram between XTX and XT should be the same, difference should be in the core binning

 

[Space Lynx]

I'd hit that!