Blackwell gpus

TNA · Sunday at 19:58

MatteRB26 said:
£700 if I can swing it by the boss

Nice price don't you think @mrk :cry:

KompuKare · Sunday at 20:19

CAT-THE-FIFTH said:
Until they lock the next big feature behind the RTX5000 or RX8000 series.

Hype aside, what would the "next big thing" be anyway?

After Fake Frames, sorry Frame Generation, well there be AIDLGPA - AI™ Deep Learning™ Game Playing Assistant? Not only plays the boring bits for you in the playing style it has learned from you but also because it is able to give temporal hints to the driver and game engine it greatly enhances what FG and DLSS can do by giving those hints before anything happens. Think of it like CPU branch prediction and speculative execution if all branches were known in advance. Laser Disc games eat your heart out!

I for one would be glad to be locked out of such a feature!

mrk · Sunday at 20:22

RTX 50 won't have any new tech exclusive to it, that's Rubin. 50 series is just a speed boost, from everything known so far.

TNA · Sunday at 20:51

mrk said:
RTX 50 won't have any new tech exclusive to it, that's Rubin. 50 series is just a speed boost, from everything known so far.

Better cross yours fingers that's the case!

AMDPower · Sunday at 20:53

mrk said:
RTX 50 won't have any new tech exclusive to it, that's Rubin. 50 series is just a speed boost, from everything known so far.

I though Blackwell is multicore type desing. Its basically two dies connected to each other through a fast interconnect but in the same package?

mrk · Sunday at 20:54

It's in the codename for the architecture, Blackwell is statistics based given who it's named after and is the first major leap in AI data crunching for the business side of things, it's why NV are a trillion dollar company now. Rubin is next generation, literally "out of this world" - It makes sense that this uses tech we've not seen before as it aligns with NV's codename theme^ Edit* obviously that's for the business side of things, might they change things for gaming GPUs? it's possible

CAT-THE-FIFTH · Sunday at 21:01

KompuKare said:
Hype aside, what would the "next big thing" be anyway?

After Fake Frames, sorry Frame Generation, well there be AIDLGPA - AI™ Deep Learning™ Game Playing Assistant? Not only plays the boring bits for you in the playing style it has learned from you but also because it is able to give temporal hints to the driver and game engine it greatly enhances what FG and DLSS can do by giving those hints before anything happens. Think of it like CPU branch prediction and speculative execution if all branches were known in advance. Laser Disc games eat your heart out!

I for one would be glad to be locked out of such a feature!

Even new iterations of the existing tech - in the end they get less money from people not upgrading.I can see AMD also heading this route too.

JediFragger · Sunday at 23:34

TNA said:
Better cross yours fingers that's the case!

It's fine, he's already selling donated review kit left and right to pay for his next upgrade

mrk · Sunday at 23:37

Yes... Some I keep, some I give away to family, some I sell, no biggie. The only instructions I am given are to "just recycle them, we're sure you can find ways". So yes, I can do what I want with stuff that is now my property

TNA · Monday at 06:41

mrk said:
Yes... Some I keep, some I give away to family, some I sell, no biggie. The only instructions I am given are to "just recycle them, we're sure you can find ways". So yes, I can do what I want with stuff that is now my property

If you ever do not know what to do with an item, let me know and I will provide you with an address to an recycle centre in London

Grim5 · Monday at 07:01

KompuKare said:
Hype aside, what would the "next big thing" be anyway?

After Fake Frames, sorry Frame Generation, well there be AIDLGPA - AI™ Deep Learning™ Game Playing Assistant? Not only plays the boring bits for you in the playing style it has learned from you but also because it is able to give temporal hints to the driver and game engine it greatly enhances what FG and DLSS can do by giving those hints before anything happens. Think of it like CPU branch prediction and speculative execution if all branches were known in advance. Laser Disc games eat your heart out!

I for one would be glad to be locked out of such a feature!

Jensen has also discussed what is next and he said future versions of dlss will generated new assets and new textures

Noxia · Monday at 10:05

TNA said:
If you ever do not know what to do with an item, let me know and I will provide you with an address to an recycle centre in London

Just saw your ad! Are you moving into the first world with 4k?

brostradamus · Monday at 10:43

rtx 5090 is supposedly 30% more dense than 4090, and i expect a bump in die size as well.. theoretically it could pack anything between 120-130 billion transistors, i think its going to be 105-110 billion transistors.. the difference between the transitor count of 4090 and 5090 is going to be more than a full 3090 chip, thats a pretty big improvement.

also blackwell based GPUs might have hardware support for new lower precision datatypes for much faster execution of dlss models, the b100 seems to have already baked in support for 4bit or lower datatypes in hardware. so i think a new dlss model dlss 4 which will be significantly faster on rtx 5090 can be expected as a huge feature upgrade

JediFragger · Monday at 10:52

brostradamus said:
dense

Reminds me of half of this forum.

mrk · Monday at 10:56

TNA said:
If you ever do not know what to do with an item, let me know and I will provide you with an address to an recycle centre in London

Fancy a 4090?

TNA · Monday at 11:36

Noxia said:
Just saw your ad! Are you moving into the first world with 4k?

Maybe. We will see

TNA · Monday at 11:39

mrk said:
Fancy a 4090?

For a three figure sum, maybe..

Especially one that has been enhanced by being in the presence of the infamous mrk

KompuKare · Monday at 14:36

brostradamus said:
rtx 5090 is supposedly 30% more dense than 4090, and i expect a bump in die size as well.. theoretically it could pack anything between 120-130 billion transistors, i think its going to be 105-110 billion transistors.. the difference between the transitor count of 4090 and 5090 is going to be more than a full 3090 chip, thats a pretty big improvement.

Yet TPU have their placeholder at x1.33 shader count and same 4N as ADA:

NVIDIA GB202 GPU Specs

24576 Cores, 768 TMUs, 192 ROPs

www.techpowerup.com

NVIDIA GB202 GPU Specs

24576 Cores, 768 TMUs, 192 ROPs

www.techpowerup.com

AD102 is 600mm², so x1.33 would be around 800mm², and seems is what is expect them to do while still on 4N. Maybe if they are separate libraries they could use, they can increase density a bit but they are hard to mix (the Zen 3D cache dies are produced separately using a tweaked process - unsure whether anything like this could be done on the same wafer as logic).

All the rumours have been pretty consistent on 4N though. I guess theoretically, Nvidia could go half chiplet like AMD by moving cache and IO to chiplets but no evidence of them considering that outside data centre.

Far more realistic is shaders 1.33 of previous, clocks back a bit since 800mm² chips are huge, but less disabled cores and faster memory for around 30% more performance.

Price? Well this would be the first time consumers get 800mm² which they usually reserve for data centre = £Not.cheap!

BigBANGtheory · Monday at 15:13

I would be very surprised if we see an 800mm2 die for consumers, sure it might happen but can you imagine the price and availability?

JediFragger · Monday at 16:58

Yeah, these ******* idiots are getting wet for a chip that will never happen (as per usual). Happens every gen.