The time you were paying a xx80 GPU 35% less for just 10% lower performance than a 90 class model and using the top die is over. Blackwell will make the performance gap between xx80 and xx90 class even bigger
It makes sense given there’s a much more lucrative market for those big consumer dies now. The 4090 is the “cheap” AD102 option.
Hopefully that ambition trickles down to the entire product stack. If AMD did roll out an MCM monster they would presumably have an equally monstrous second tier card. Nvidia would need to prepare a competitor for that card too carved from GB202.
Last edited:
(1) Exactly but Blackwell is a bit different. That's why they have GB102 and GB202
(2) nothing is more sure. It will ultimately depend on AMD competitiveness... I suspect 5090 to be the biggest performance uplift of the stack and maybe even the best value for money... Another factor to consider is that Blackwell will fight both RDN4 and RNDA5 during it's lifespan. So decisions will be different this round...
Interesting. There’s a separate GDDR datacenter chip with GB202 taking on gaming and visualization duties?
DegustatoR
Legend
My assumption is that GB102 is a smaller version of GB100 and both aim HPC / DC markets. GB20x are gaming chips.
Assuming GB202 is 512-bit bus I can see with 2GB Modules the following SKUs:
- Quadro GB202 - 64GB
- 5090 Ti (if they do that) - 32GB
- 5090 - 28GB.
- 5080 Ti - 24GB.
- 5080 - 20GB.
DegustatoR
Legend
I wouldn't expect GB202 in anything but 5090. It's memory capacity will depend on its bus and what G7 chips will be available at launch but it shouldn't be less than 24GBs which is way more than enough for everything in 4K at the moment.
5080 will likely use GB203 which means that it's either 16GBs or 24GBs if 3GB G7 chips will be available at launch. Either option is fine IMO for a card which prooooooobably won't be able to beat 4090.
5080 will likely use GB203 which means that it's either 16GBs or 24GBs if 3GB G7 chips will be available at launch. Either option is fine IMO for a card which prooooooobably won't be able to beat 4090.
That would be disappointing. The 4070 Ti is more than a match for the 3090 and it’s based on AD104.
Let’s assume these rumors are accurate for a sec. GB202 is a 192SM part and GB203 is 96SMs. The 4090 is 128SMs. So a 5080 on GB203 is starting off with quite a deficit but it’s not impossible that new arch + clocks can close the gap.
And what better way to redeem the venerable FX 5800 Ultra!
Bondrewd
Veteran
That was a big shrink.
Which we're out of.
N3e is barely faster than any relevant N4 derivatives.
All down to uarch basically.
Which tier though? Other than the 4090, the increase was one of the smallest in history.The time you were paying a xx80 GPU 35% less for just 10% lower performance than a 90 class model and using the top die is over. Blackwell will make the performance gap between xx80 and xx90 class even bigger
PS: don't forget that Blackwell was designed when NVDA predicted to compete with a 3nm MCM RDNA4 monster. So the increase in performance over Ada is not lower than Ada vs Ampere...
DegustatoR
Legend
What's this has to do with Blackwell family? These appeals to how things were some time ago aren't doing us any favors.
I mean it's not impossible I guess but I certainly don't expect it. The cards will likely be on par at best IMO.
4N is supposedly customized N5 rather than N4 though
Bondrewd
Veteran
Every node every vendor is using is 'customized' since DTCO has been name of the game since CLN10FF-ish.
In effect, it's a meaningless word.
Point was that the small gains N4 gives over N5 aren't included in 4N, so N3e should offer them some improvement
Bondrewd
Veteran
Again, those are meaningwell words.
We don't know what N5 derivative PDK Nvidia uses, it's confidential data.
Yea but it's tiny.
Which is the point.
Shrink bonanza is ~over~, we're at the end of CMOS history.
Frenetic Pony
Veteran
Could you provide any data to back up that claim? Haven't seen the game showing any VRAM issues on any cards on any preset thus far. IIRC they are allocating all VRAM available but that doesn't mean that they need that much to run without issues.
View attachment 10967
Avatar: Frontiers of Pandora: PC Performance Benchmarks for Graphics Cards and Processors in Unobtanium quality mode | Action / FPS / TPS | GPU TEST
We ran Avatar: Frontiers of Pandora tests using unlocked Unobtanium graphics settings on GEFORCE RTX and RADEON RX graphics cards. We evaluated both the performance and the graphical quality of the ggamegpu.tech
Also this one is just at Ultra but since there are no real difference between 4060Tis there I wouldn't expect that the 16GB one would suddenly start showing issues on Unobtaineoum.
Truth is 16GBs is what current gen consoles have (total VRAM yada yada) and thus it will very likely remain the "sweet spot" until we'll switch the generations of console h/w again.
The exclusions here would be titles like CP2077 which are very rare and thus won't affect the overall picture too much.
For those who think that 16GBs aren't enough well there are products with more VRAM. Prepare to pay a lot.
welcome GAA
DegustatoR
Legend
That's a video from a 4090. As I've said the game tends to allocate VRAM until it's available (which is a good thing tbh as too many games these days seem to stick to artificial limits for no reason).
Doesn't mean that it actually *needs* as much to run without issues.
Frenetic Pony
Veteran
Earlier in the same video lower settings use much lower VRAM though, just rewind
Also game generally don't become "unstable" (well, they shouldn't) when they run out of VRAM, they become slow.
Last edited: