AMD Radeon Pro W7800 GPU Tweaked to Simulate "RX 7800 XT"

[T0@st]

An AMD Navi 32 die was belatedly observed in a Forbes video feature on the company's CEO and President Lisa Su—this small tidbit fired up the PC hardware community once again with speculation about why Team Red has not yet released proper mid-range RDNA 3 game-oriented models. A handful of news sites have recently reported that a Navi 32 GPU sits at the heart of AMD's fairly new workstation-grade AMD Radeon Pro W7800 32 GB GDDR6 graphics card, but fact checkers have quickly pointed out that the $2499 (MSRP) product is actually based on Navi 31. Sites have theorized about the makeup of a possible "Radeon RX 7800" GPU and assumed that a similarly named/numbered workstation model would offer a preview of things to come.

Igor Wallossek (of Igor's Lab fame) has conducted an interesting investigation into this matter. He has put a Radeon Pro W7800 test unit through its paces as a gaming card, but the high-end nature of the Navi 31 GPU leads him to believe that the performance level on tap would be roughly equivalent to a hypothetical "RX 7800 XT." Igor assumes that his simulated gaming card will have access to a smaller pool of VRAM (16 GB instead of 32 GB)—he achieves this via the memtestcl program, since "RDNA 3 no longer allows us to emulate the cards directly." He also sets provisions for differing power consumption due to the workstation card being an efficiency-focused product: "The Radeon Pro W7800 has a TBP of 260 watts, my own extrapolation results in a TBP of around 270 watts for the RTX 7800XT based on the efficiency values of the other two Navi 31 cards."




Igor also made a comparison to AMD's current flagship gaming card: "The Radeon PRO W7800 has been designed for professional users by AMD...It is manufactured in the 5-nanometer process...This graphics card is compatible with DirectX 12 Ultimate, the latest generation of the DirectX interface. With a die area of 529 mm² and 57.7 billion transistors, the Navi 31 GPU is a large and powerful chip. Unlike the Radeon RX 7900 XTX, which uses the same GPU and has all 6144 shader units enabled, AMD has disabled some shader units on the Radeon PRO W7800 to achieve a specific shader count for this product." He outlines the workstation model's slightly lesser specs: "As a result, the Radeon PRO W7800 only has 4480 shader units, 280 texture mapping units and 128 ROPs (Raster Operations Pipelines). Furthermore, 70 ray tracing acceleration cores are integrated. Up to this point, the data also exactly matches the colported RX 7800 XT. After all, AMD has given the Pro card a full 32 GB of GDDR6 memory instead of 16 GB (clamshell), connected to the GPU via a 256-bit interface. The operating frequency of the GPU is at a comparatively low 1855 MHz and can be clocked up to 2499 MHz if required, while the memory works at a speed of 2250 MHz (effective 18 Gbps)."



Overall Igor's benchmark tests of his simulated gaming card show small performance advantages - 4% (1080p), 8% (1440p) and 12.5% (2160p) - over the previous gen Radeon RX 6800 XT model, he also compares it to mid-range Team Green offerings: "Now, of course, everyone will ask why AMD hasn't brought the Radeon RX 7800XT yet. According to my projections, it would pretty much fit between a GeForce RTX 4070 and the RTX 4070 Ti - without any real competition from NVIDIA. Actually, something like that is an ideal case, if two things wouldn't speak against it. AMD's biggest problem is the overstock of the Radeon RX 6800 XT and RX 6900 XT. The Radeon Pro W7800 tested today is a card trimmed for efficiency, (this places it) exactly between these two cards."

He concludes that AMD could be onto a winner, but a new upper mid-range model will likely face internal competition from a heavily discounted predecessors: "The Radeon RX 7800 XT with a power limit of around 270 to 275 watts and a higher clock could well become a real blockbuster for AMD, if the price is right. This is because it fills a gap that NVIDIA has criminally left open and that definitely cannot be filled with the current generation. The (hypothetical) Radeon RX 7800 XT not only has more memory (assumed 16 GB) than the two NVIDIA cards (12 GB), but it also places itself in the field of the more powerful WQHD cards for higher FPS rates without a direct opponent. NVIDIA would have to boost the GeForce RTX 4070 quite a bit as super version, which is not really realistic."

View at TechPowerUp Main Site | Source

 

[ToTTenTranz]

This isn't a great simulation... N32 has 60 CUs, not 70 like igorslab assumes. The frontend is also reduced.

Most probably Radeon N32 is going to clock above the Pro N31, especially during boost.

 

[fancucker]

The entire lineup could be rendered obsolete by the 4000 super series. Sad overall generation.

 

[Mr. Perfect]

Is there any official word on stock levels of the older 6000 cards? Sometimes manufacturers talk about stock levels on earning calls, is there anything like that to go off of or is this all speculation?

 

[Dragokar]

This isn't a great simulation... N32 has 60 CUs, not 70 like igorslab assumes. The frontend is also reduced.

Most probably Radeon N32 is going to clock above the Pro N31, especially during boost.

Igor does not simulate N32, he does simulate a potential 7800XT. There are rumors that full N32 with 60CU is going to be 7800, 7700XT and also the 7600XT. There are also rumors that N31 salvage will be used for a 7800 XT or a 7900 non XT.

 

[LabRat 891]

All this brings to mind is that I'm sad we won't see more Radeon Instinct cards that can be turned into Radeons.

 

[Keullo-e]

The entire lineup could be rendered obsolete by the 4000 super series. Sad overall generation.

Are there super series coming? The 30 series lacked those.

 

[Crackong]

The entire lineup could be rendered obsolete by the 4000 super series. Sad overall generation.

I doubt that Ngreedia would release super series when there is no significant competition.

 

[uftfa]

I'm very confused by the benchmark data, to be honest. The 4K data suggests that a 3090 is 27% faster than a 3080 (it isn't), and a 6800XT is faster than a 3080 at 1080p and 4K, but slower at 1440p.

 

[wolf]

I doubt that Ngreedia would release super series when there is no significant competition.

Maybe, but sales for many SKU's have been super lackluster, I see them doing a mid cycle Ti refresh/adding cards to the lineup to try boost sales. Unlikely to wear the Super namesake though.

What a disappointing generation from both camps so far overall.

 

[Minus Infinity]

I mean if this is how the 7800XT performs in real life it would be an epic failure and sad indictment of how pathetic RDNA3 has turned. As I said the other day, they might as well have skipped this gen and waited for RDNA 4. The delays on N32/N33 are a joke and yes I know they claim to have good reasons. But they knew the market would suck this year, they knew they were going to have warehouses full of unsold RDNA32 cards.

 

[Crackong]

Maybe, but sales for many SKU's have been super lackluster

We all know Ngreedia purposely made the lackluster SKU prices so to upsell the 4090.
If they really care , just lower their prices back to where they should be, and subsidies the AIB partners.

The AIB partners already had many dead stock on hand.
Making more super variants won't help.

 

[thunderingroar]

I'm very confused by the benchmark data, to be honest. The 4K data suggests that a 3090 is 27% faster than a 3080 (it isn't), and a 6800XT is faster than a 3080 at 1080p and 4K, but slower at 1440p.

My guess is that 3080 10GB runs out of vram at 4k, it explains why other 2 cards with 24GB and 16GB outperform it

 

[TheoneandonlyMrK]

What is meant by colported ?!?

"Furthermore, 70 ray tracing acceleration cores are integrated. Up to this point, the data also exactly matches the colported RX 7800 XT. After all, AMD has given the Pro card a full 32 GB of GDDR6 memory instead of 16 GB"