Tuesday, June 11th 2024
Possible Specs of NVIDIA GeForce "Blackwell" GPU Lineup Leaked
Possible specifications of the various NVIDIA GeForce "Blackwell" gaming GPUs were leaked to the web by Kopite7kimi, a reliable source with NVIDIA leaks. These are specs of the maxed out silicon, NVIDIA will carve out several GeForce RTX 50-series SKUs based on these chips, which could end up with lower shader counts than those shown here. We've known from older reports that there will be five chips in all, the GB202 being the largest, followed by the GB203, the GB205, the GB206, and the GB207. There is a notable absence of a successor to the AD104, GA104, and TU104, because NVIDIA is trying a slightly different way to approach the performance segment with this generation.
The GB202 is the halo segment chip that will drive the possible RTX 5090 (RTX 4090 successor). This chip is endowed with 192 streaming multiprocessors (SM), or 96 texture processing clusters (TPCs). These 96 TPCs are spread across 12 graphics processing clusters (GPCs), which each have 8 of them. Assuming that "Blackwell" has the same 256 CUDA cores per TPC that the past several generations of NVIDIA gaming GPUs have had, we end up with a total CUDA core count of 24,576. Another interesting aspect about this mega-chip is memory. The GPU implements the next-generation GDDR7 memory, and uses a mammoth 512-bit memory bus. Assuming the 28 Gbps memory speed that was being rumored for NVIDIA's "Blackwell" generation, this chip has 1,792 GB/s of memory bandwidth on tap!
The GB203 is the next chip in the series, and poised to be a successor in name to the current AD103. It generationally reduces the shader counts, counting on the architecture and clock speeds to more than come through for performance; while retaining the 256-bit bus width of the AD103. The net result could be a significantly smaller GPU than the AD103, for better performance. The GB203 is endowed with 10,752 CUDA cores, spread across 84 SM (42 TPCs). The chip has 7 GPCs, each with 6 TPCs. The memory bus, as we mentioned, is 256-bit, and at a memory speed of 28 Gbps, would yield 896 GB/s of bandwidth.
The GB205 will power the lower half of the performance segment in the GeForce "Blackwell" generation. This chip has a rather surprising CUDA core count of just 6,400, spread across 50 SM, which are arranged in 5 GPCs of 5 TPCs, each. The memory bus width is 192-bit. For 28 Gbps, this would result in 672 GB/s of memory bandwidth.
The GB206 drives the mid-range of the series. This chip gets very close to matching the CUDA core count of the GB205, with 6,144 of them. These are spread across 36 SM (18 TPCs). The 18 TPCs span 3 GPCs of 6 TPCs, each. The key differentiator between the GB205 and GB206 is memory bus width, which is narrowed to 128-bit for the GB206. With the same 28 Gbps memory speed being used here, such a chip would end up with 448 GB/s of memory bandwidth.
At the entry level, there is the GB207, a significantly smaller chip with just 2,560 CUDA cores, across 10 SM, spanning two GPCs of 5 TPCs, each. The memory bus width is unchanged at 128-bit, but the memory type used is the older generation GDDR6. Assuming NVIDIA uses 18 Gbps memory speeds, it ends up with 288 GB/s on tap.
NVIDIA is expected to double down on large on-die caches on all its chips, to cushion the memory sub-systems. We expect there to be several other innovations in the areas of ray tracing performance, AI acceleration, and certain other features exclusive to the architecture. The company is expected to debut the series some time in Q4-2024.
Source:
kopite7kimi (Twitter)
The GB202 is the halo segment chip that will drive the possible RTX 5090 (RTX 4090 successor). This chip is endowed with 192 streaming multiprocessors (SM), or 96 texture processing clusters (TPCs). These 96 TPCs are spread across 12 graphics processing clusters (GPCs), which each have 8 of them. Assuming that "Blackwell" has the same 256 CUDA cores per TPC that the past several generations of NVIDIA gaming GPUs have had, we end up with a total CUDA core count of 24,576. Another interesting aspect about this mega-chip is memory. The GPU implements the next-generation GDDR7 memory, and uses a mammoth 512-bit memory bus. Assuming the 28 Gbps memory speed that was being rumored for NVIDIA's "Blackwell" generation, this chip has 1,792 GB/s of memory bandwidth on tap!
The GB205 will power the lower half of the performance segment in the GeForce "Blackwell" generation. This chip has a rather surprising CUDA core count of just 6,400, spread across 50 SM, which are arranged in 5 GPCs of 5 TPCs, each. The memory bus width is 192-bit. For 28 Gbps, this would result in 672 GB/s of memory bandwidth.
The GB206 drives the mid-range of the series. This chip gets very close to matching the CUDA core count of the GB205, with 6,144 of them. These are spread across 36 SM (18 TPCs). The 18 TPCs span 3 GPCs of 6 TPCs, each. The key differentiator between the GB205 and GB206 is memory bus width, which is narrowed to 128-bit for the GB206. With the same 28 Gbps memory speed being used here, such a chip would end up with 448 GB/s of memory bandwidth.
At the entry level, there is the GB207, a significantly smaller chip with just 2,560 CUDA cores, across 10 SM, spanning two GPCs of 5 TPCs, each. The memory bus width is unchanged at 128-bit, but the memory type used is the older generation GDDR6. Assuming NVIDIA uses 18 Gbps memory speeds, it ends up with 288 GB/s on tap.
NVIDIA is expected to double down on large on-die caches on all its chips, to cushion the memory sub-systems. We expect there to be several other innovations in the areas of ray tracing performance, AI acceleration, and certain other features exclusive to the architecture. The company is expected to debut the series some time in Q4-2024.
141 Comments on Possible Specs of NVIDIA GeForce "Blackwell" GPU Lineup Leaked
Is AMD a story of lost and wasted potential? Damn sure. I don't disagree on that. But it is what it is, and I do believe they have a strategy now that's working for them, with little reasons included to completely lose the PC market, even a small presence won't kill their margins on GPU, its just extra.
The only trend I see is that they never get above 40% marketshare and that they have been below 20% for the last seven quarters - but even here, with ups of 5%-points.You can talk someone to death. They have certainly done better before, but theay have always come up with inovations to carve out their marketshare. RDNA3 hasn't been as successfull as hoped, but RDNA2 showed again that innovation can beat long term market domination to a certain point.
Since I remember the release of Radeon 9700 Pro and HD5870 that had NV overwhelmingly beat for months and also HD48x0 and the whole GCN1-3 series that didn't beat NV in every aspect but were so competitive that prices went down to an alltime low, I am certain AMD Radeon will rise up again.
If anything those ups show that there is potential to regain market share. But selling a quarter or a few quarters 'better' than your usual trend of 'down YoY' isn't a positive. It just means you reacted to the market proper with either pricing or timing. And in AMD's case, its always pricing. Pricing under the competition.
But again, strategy is more than just the tiny slice of PC dGPU for consumers. GPU is bigger than that. For AMD, its also APU, its console business, its laptop products, etc.
And I do think Intel is their more direct competitor, especially now that they're also doing GPU; AMD has to stay ahead of them much more so than Nvidia, who's limited in the x86 space.
I don't want to see AMD exit the desktop GPU market, regardless of how many people call me a fanboy how many times, because I'm scared of what that would mean for consumers. I'm just presenting data and saying, this looks really bad, and AMD needs to change something to make it not bad - and endless forum posts accusing NVIDIA of being greedy, or anticompetitive, or whatever are not it. Nor is buying AMD products out of a misplaced sense of brand loyalty.
It is what it is, sometimes things need to turn to absolute shit before they get better.
It's like RT "just works" (except when it doesn't)Yeah, Nvidia's bread and butter is currently datacentre compute.
AMD have made fantastic inroads to the servers market with EPYC but their enterprise/datacentre GPU solutions lack the CUDA support they need to gain any significant traction. I think Nvidia knew this and they've been playing the long game with the better part of two decades of investment in CUDA's software/API/ecosystem monopolisation.
But perhaps we are interpreting AMDs chances to much from the highend point of view. After all, their marketshare was quite strong in the times of Polaris & Vega vs Pascal.
We're actually talking about them competing with Intel. And let's be real here: their CPU product is now better, their GPU product is lightyears ahead of them too. Hasn't Nvidia ALWAYS been something they couldn't quite catch? I'm not seeing a difference here in the overall perspective. Every time, even when, rarely, Nvidia did objectively worse, Nvidia won.Thanks! Luckily I can't detect the sarcasm here :)
Alternatively, it can be called an EVUL move from an EVUL anti-competitive company that is led by a Dark Lord in a Leather Jacket of EVULNESS +7 who is bent on not allowing gamers to play the latest AAA slop at highest detail out of sheer spite for poor AMD and their paladins of virtue representing them on the forums.
You know, either/or.
They make it sound like all AMD gpus are at least half as slow to their Ngreedia counterpart.
Yes, bring the RT nonsense and as stated, only influencers aka reviewers care about that.
Only 2 games (so far) are a decent sample of RT (Cyberpunk and Control) but doesnt add anything to gameplay, neither justifies the insane performance hit.
Absolute, raw performance is the only real indicator. Because even RT performance will be based on that; you can accelerate all you want, but you can't accelerate past your raster/raw perf capability.
Also, alongside scaling their GPU up, there is the matter of die space/cost. How big is their die? How big is AMD's? And that's where chiplets come in. If AMD can improve that further, they have a technological advantage here, and the first release isn't horrible with RDNA3. Its not perfect. But it still moved the ball forward.
Scaling was AMDs problem more than once, too.
Looks like Star Wars Outlaws will have some sort of RT implementation, and I look forward to seeing how that is.
Especially if you want cross platform compatibility, which is rapidly becoming a must especially for anything that also release outside the PC camp (but even within, think handhelds!), please do explain to me how you're going to guzzle extra power to enable RT proper if you haven't even got that to show the best of raster graphics.
It ain't happening. Literally every market movement except the one Nvidia tries to convince us of, is moving in the opposite direction. RT will only work if you get your game from the cloud. So, how does that mix exactly with Nvidia selling you 1500 dollar GPUs I wonder? Where is this long term RT perspective on dGPU?
Besides, not only are we a ways off, I'd argue that games still look plenty good without path tracing, I can wait.
If you check modern games, they almost all use over 12GB ram in 4K. There is nothing to argue over here, Nvidia is limiting the VRAM to up-sale people to a higher tier and more expensive card and limit how future-proof those cards are, just like when they release a new technology and limit it to only their newest series.