• Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.

NVIDIA GeForce RTX 50 “Blackwell” GPU Configs Revealed

Bernoulli

M2 slut
  • GB202 = 24,576 Cores (512-bit G7) / +33% More Cores Vs AD102
  • GB203 = 10,752 Cores (256-bit G7) / +5% More Cores Vs AD103
  • GB205 = 6,400 Cores (192-bit G7) / -16.6% Cores Vs AD104
  • GB206 = 4,608 Cores (128-bit G7) / Same Cores Vs AD106
  • GB207 = 2,560 Cores (128-bit G6) / -16.6% Cores Vs AD107




NVIDIA GB202 GPU - Flagship Blackwell Gaming Die For GeForce RTX 50 Series

Starting with the top, we have the NVIDIA GB202, the flagship GPU die for the GeForce RTX 50 "Blackwell" lineup. This GPU is what's going to power the ultra-enthusiast GeForce RTX 5090 & higher-end SKUs. The GB202 GPU will feature a total of 12 SMs, each with 8 TPCs for a total of 96 TPCs and each TPC will be housing 2 SMs for a total of 192 SMs. The GPU will support a memory bus interface of up to 512-bit which can be cut into different configurations depending on the SKU such as 448-bit, 384-bit, 320-bit, and more unique designs as NVIDIA desires for its next-gen line.

NVIDIA GB203 GPU - Built For High-End Enthusiast Gamers

Next we have the NVIDIA GB203 GPU which will be another high-end SKU for the GeForce RTX 50 "Blackwell" graphics family. This chip will feature 7 GPCs, each with 6 TPCs for a total of 42 TPCs & that rounds up to 84 SMs in total. This GPU will feature a 256-bit bus interface and will be used by the GeForce RTX 5080 graphics card.

NVIDIA GB205 GPU - To Be Featured On The GeForce RTX 5070

Coming to the mid-tier segment which is still in the high-end performance range, we have the NVIDIA GB205 which will likely target the GeForce RTX 5070 and RTX 5070 Ti graphics card. This chip will come with 5 GPCs, each with 5 TPCs for a total of 25 TPCs and 50 SMs in total. This chip comes with a 192-bit memory bus configuration.


NVIDIA GB206 GPU - Designed For The Mainstream Gamers

On the mainstream side of things, we have the NVIDIA GB206 GPU which will carry 3 GPCs with 6 TPCs per GPC and 18 TPCs in total. The chip will feature 36 SMs and will come with a 128-bit bus interface which is quite narrow but with memory compression algorithms and larger L2 caches like the ones we saw on the Ada lineup, we can expect decent performance on the next-gen GeForce RTX 5060 (Ti) family.

NVIDIA GB207 GPU - Entry-Level Sticks With GDDR6 Memory

Lastly, there's the NVIDIA GB207 GPU which is the entry-level SKU and is likely going to target the GeForce RTX 5060 or the RTX 5050. The chip comes with 2 GPCs with 5 TPCs each for a total of 10 TPCs and 20 SMs. The chip will also feature a 128-bit memory interface but will likely come with both GDDR7 & GDDR6 memory support while the rest of the lineup leverages the newer GDDR7 memory standard.

 
Last edited:

nemiroff

Gold Member
sold out GIF by Hannah Nance
 

M1chl

Currently Gif and Meme Champion
2.5k USD, 2.3k EUR for 5090 here in Europe I am guessing and that is if there is not going to be shortage. Still as always I am in.
 
Last edited:

//DEVIL//

Member
if the 5090 is above 2k and above i am out. my 4090 is just fine.

if it was 2k for most of the board partners its fine. but if the FE is 2k, that means other cards will be 2.2 or 2.3k. for that they can fuck off.

it's not like there are games that will take advantages of its resta power. everything I have seen on the xbox show for the next year will be played just fine with a 4090 at max settings.
 

Bernoulli

M2 slut
GB205 better not be the fucking die for the 5070.
it is
NVIDIA GB205 GPU - To Be Featured On The GeForce RTX 5070

Coming to the mid-tier segment which is still in the high-end performance range, we have the NVIDIA GB205 which will likely target the GeForce RTX 5070 and RTX 5070 Ti graphics card. This chip will come with 5 GPCs, each with 5 TPCs for a total of 25 TPCs and 50 SMs in total. This chip comes with a 192-bit memory bus configuration.
 

SmokSmog

Member
GB205 better not be the fucking die for the 5070.
GB205 looks to anemic for 5070, I think it's for 5060Ti so they can finally move from 8GB Vram to 12GB.

5070 will get cut down GB203 and 5080 full GB203, both with 16GB Vram.
 
Last edited:
  • Like
Reactions: Gp1

SHA

Member
I bet the 4090 and the 3090 will cripple when real next games enter the market, the thing that hasn't changed about gpus is that they're always reacting not proacting.
 

analog_future

Resident Crybaby
If I were you I would wait for 6090. I doubt we will see differences on the level of 3090ti-4090 this time. Probably no exclusive tech like dlss3 too.

Yeah I'll wait to see how the 5090 performs. I just want something that can max out GTA VI @ 4k/60. It's that game of if I wait until the 6090, the 4090 loses a ton of resale value as well. We'll see!
 
Last edited:

JohnnyFootball

GerAlt-Right. Ciriously.
I am really hoping that I am not tempted to upgrade my 4090 to a 5090.

I could see myself sidegrade to a smaller more power efficient GPU.

But...I won't get the 24GB of RAM, that I don't need but do enjoy having.
 

nemiroff

Gold Member
Based on this 5080 should be weaker than 4090.

Massive fucking scam again, 5090 will be fucking big and powerful and then we have mid tier 5080, massive gap in between.

This is Ada part 2 - we will scam you again!
Nah man.. We all know what we buy and for how much, and how these fits in our personal situation of value. Nvidia's practices are well-known and unapologetic, arguably shameful, but it certainly doesn't constitute a scam in any sense of the term.
 

JohnnyFootball

GerAlt-Right. Ciriously.
I would guess 5080 30% faster than 4080, 10% from 4090 with tdp lower than 300w
If that were the case, I'd honestly consider swapping out.

The question would be how much VRAM would a 5080 have. I feel a 5080 would have 16GB and my feeling is that 16GB will become constraining far sooner than we expect.
 
I'm hodling for the 5080 Ti to upgrade my 3080 Ti. Should give me enough time to save up for $2k or whatever ridiculous cost it'll be.
 

Bojji

Gold Member
Nah man.. We all know what we buy and for how much, and how these fits in our personal situation of value. Nvidia's practices are well-known and unapologetic, arguably shameful, but it certainly doesn't constitute a scam in any sense of the term.

It's a scam based on how much you pay for small and cut down GPUs they offer. Not enough VRAM and super small memory bandwidth - BAM - 400 dollars!

Slightly better memory BW and "enough" VRAM - BAM - 600 dollars!

Still, you pay 600$ for super small GPU.

They are transparent with their prices so it's not a scam in literal sense, they are just a shitty and greedy company like Apple.
 
Last edited:

Kenpachii

Member
So basically the cards will be a minimal upgrade from 4000 series cards besides the 5090. Whats going to be interesting is what DLSS 4.0 is going to offer.
 

nemiroff

Gold Member
If I were you I would wait for 6090. I doubt we will see differences on the level of 3090ti-4090 this time. Probably no exclusive tech like dlss3 too.

Value is, after all, a dynamic concept, so your points are great. But at the same time, some of us game on UHD+ 240Hz+ panels and 4K+ VR headsets, so there's no problem finding places to utilize the performance. ..But if there was like a ~10% difference, then of course even a die-hard buy-everything guy like me would have a hard time justifying it.

The disparity between the 5080 and 90 is ridiculous.
Yeah I noticed that.. I agree it looks odd on paper. Context depends on price and clock speeds as well of course.

a shitty and greedy company like Apple.

A shitty and greedy company which "unfortunately" makes the best GPU technology (and with self-esteem boosting soaring stock prices).. Imagine if there was actual competition in this space. I wish AMD was able to keep up. I had hope for Intel, but it takes so much time for them to ramp up..


Edit: Removed quote from Kenpachii, I misread it.
 
Last edited:

T4keD0wN

Member
Hope the naming is not indicative of performance relative to the highest model and this ends up compltely false.
GA102 was used for as low as 3070ti, GA103 was used for a 3060ti, now GB203 is used on a 5080

So 5090 is supposed to be 50% faster and 5080 is expected to be a bit weaker than a 4090. Zero performance increase expected for 60, 60ti and 70 cards.
3060ti was faster and cheaper than a 2080 super, now 5060ti is supposed to be on par with a 4060ti (which was the same and in some cases weaker than a 3060ti), bet itll cost more too.

This is way worse than Intels infamous stagnation
bad-deal-this-deal-is-getting-worse.gif
 
Last edited:

Gaiff

SBI’s Resident Gaslighter
The 5090 should be about 50% faster than the 4090 and the 5080 a tad slower than the 4090.

With the 5070/5060 Ti/5060 being roughly the same performance as Ada equivelent.

That's in games. Add 50-100% for AI performance (TOPS).
Without clock speeds and other aspects? Don’t think anyone can tell with any degree of certainty.
 

hinch7

Member
5090 won't be anywhere near the full die. That's for AI cards costing 6,7 grand.
Perhaps. Even if cut down a little at least it will be a meaningful upgrade. They could save full die for the 5090Ti, if they want to go against top RNDA 5 - assuming AMD/RTG gets anywhere close with their next gen Radeon.

What I don't understand is the downgrade to the GB205. I'm guessing the architectural improvements and having faster memory bandwidth helps, but still. Looks to be another mid generation for everyone else.
 
Last edited:
Top Bottom