Nvidia Announces GV100 Volta Based GPU, Built on TSMC 12nm FinFET with a Die Size of 815 mm^2

[Valentyn]

4 minutes ago, NumLock21 said:

it was also in the 6 figure price range.

Yeah, early adopters essentially paying off all the R&D costs, and costs of early production.

 

V100 cost 3 Billion in R&D alone.

[tsk]

3 hours ago, Taf the Ghost said:

Volta, consumer side, would appear to be a die-shrink, so big architecture updates would be 2019 or so.  Considering where Nvidia sits, that's not bad deal for them.  They can offer 1070 experience for 1050 to 1060 cost and lower wattage. Considering the "king" card appears to still be the 970, that's clearly the important level for "enough" gaming power.

 

For the HPC side of things, I'm very curious about the new Tensor processing units. We've been seeing this trend a lot recently, so it's cool to see some of the interesting ones that come along.  (Ryzen has a specialized AES encrypt/decrypt unit, making it upwards of 400% faster than Intel solutions for cryptological tasks. But this type of stuff isn't new.) 

GP100 at the same die size would be 20.5B transistors, so this is hardly a die shrink, just your usual TSMC nm marketing. Volta is all about improvements to compute power.   

 

12nm FFN =12nm FinFetNvidia btw. 

[VagabondWraith]

Can't wait to get my hands on the TITAN behemoth. I'm getting tired of this 1080 Ti already.  

[HelplmChoking]

4 hours ago, Andrevcn said:

Just bought an expensive 1080 Ti. Thanks Nvidia. 

 

 

You didn't think they'd be coming out with a replacement eventually, just like the have for decades?

[DocSwag]

1 hour ago, Flavio hc 16 said:

maybe i'm saying nonsense here, but if i'm remembering it right when a gpu has hbm memory, the memory controller has to be on the gpu die, as it is with vega, so it's normal to have a huge gpu

the memory controller is on the gpu die in all products, pascal, maxwell, polaris, etc. included.

[DocSwag]

4 hours ago, Andrevcn said:

Just bought an expensive 1080 Ti. Thanks Nvidia. 

 

You're not gonna be seeing consumer Volta GPUs until 2018, and most likely no xx80ti card until end of 2018 or 2019.

[Avcn42]

1 hour ago, Rangaman42 said:

You didn't think they'd be coming out with a replacement eventually, just like the have for decades?

No! 

But seriously now, I just think that they have been putting the owners of top end cards in a bad situation, leaving consumers to believe that they bought the best out there, only to have their product replaced in months. However, just like the other user who quoted me, I think (I hope) they won't release a consumer volta card soon. But I've dealt with Nvidia cards suffering with bad drivers after being replaced, and I hope they prolong this inevitability. 

[Juanitology]

5 hours ago, VanayadGaming said:

I so hate the 12nm description ... the actual density is not that much better than Pascal. 
 

Compare the transistor/ mm² ratio between the 2 nodes:

16nm: 15.3BT/ 610mm² = 25MT/ mm²

12nm: 21BT/ 815mm² = 25.7MT/ mm²

The distance that companies usually quote as their process in marketing and such is just the gate-width. There are many more transistor features on the chip (especially on GPUs) which are larger than the gate itself and make up most of the space used on the die. While the gate-width might be 12nm, all surrounding features probably add up to 70 or 80nm. Therefore a 25% (4nm) improvement on gate without corresponding improvement on the rest of the features will have very little impact on overall die size, such as in this case. 

[Jito463]

7 minutes ago, rrubberr said:

When you realize you could have 20 Ebay Titan Z's for this price.

But where would you fit them all.....?

 

[Drak3]

4 hours ago, MageTank said:

The fact that it's scheduler structure looks very similar to Fermi, tells me that it might run hot, like Fermi, lol.

Nah, it's fine.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

[Equlix]

9 hours ago, VerticalDiscussions said:

Okay, i dont want to accuse anyone and i personally did not hate whoever i state here, but can you just tell me if this was the person im actually thinking about?

 

 = 

PS: No names placed. I would appreciate if you could tell me wether im near the hot or cold zone.

 

 

wut. I haven't been on the forums in a while and I come back to find that the everyone's favorite master baiter is banned? The mods are absolute mad men. Mage must be bored af now though.

[Carclis]

9 hours ago, Trixanity said:

Dat die size tho.

 

I wonder what the yields are on that bad boy considering the size and the new node.

I believe Jhen-Hsun Huang said on the stream that you would be lucky to get a single GV100 chip out of a whole wafer.

[leadeater]

2 hours ago, Carclis said:

I believe Jhen-Hsun Huang said on the stream that you would be lucky to get a single GV100 chip out of a whole wafer.

That seems a little insane?

[Carclis]

27 minutes ago, leadeater said:

That seems a little insane?

Yeah, it seems a little desparate. I'll try and find the quote again when the full presentation gets uploaded.

[TheRandomness]

Ah, so they're taking on the brute force approach with the stream processors, I see?

 

Sarcasm aside, holy shit 5120 SPs. 

[Valentyn]

15 hours ago, VerticalDiscussions said:

Okay, i dont want to accuse anyone and i personally did not hate whoever i state here, but can you just tell me if this was the person im actually thinking about?

 

 = 

PS: No names placed. I would appreciate if you could tell me wether im near the hot or cold zone.

 

 

 

Haha! His new home is the WCCFT comment section,you can find him there on the current AMD 16 Core CPU post.

[yian88]

Imagine that Vega100 running 4K games on its own inside your PC on your 4K OLED HDR10 1ms Gsync 50 inch TV, along a Zen++ 10c/20t CPU 4ghz. 

Thats something worth donating a kidney for, you have 2 anyway dont need 2.

[Carclis]

3 hours ago, leadeater said:

That seems a little insane?

Found it.

[The Benjamins]

1 hour ago, yian88 said:

Imagine that Vega100 running 4K games on its own inside your PC on your 4K OLED HDR10 1ms Gsync 50 inch TV, along a Zen++ 10c/20t CPU 4ghz. 

Thats something worth donating a kidney for, you have 2 anyway dont need 2.

you mean naples 32 core CPU OCed to 4Ghz.

[leadeater]

1 hour ago, Carclis said:

Found it.

"If anyone would like to buy one it's 3 billion dollars" LOL!

[Eroda]

does it support full async compute tho

[Valentyn]

2 minutes ago, Eroda said:

does it support full async compute tho

 

 

 

[DocSwag]

1 hour ago, Carclis said:

Found it.

No wonder it's $15k...

 

The GV102 GPU that's gonna be for consumers is also gonna be insane... Judging by GV100 and GP100 vs GP102 it's gonna be around 600 mm^2...

[Princess Luna]

18 hours ago, DocSwag said:

What are you guy's thoughts of this LITERAL monster?

Makes me happy that I ignored the GTX1080ti in favour to keep saving money for the TITAN X (Volta).

[79wjd]

3 hours ago, TheRandomness said:

Ah, so they're taking on the brute force approach with the stream processors, I see?

 

Sarcasm aside, holy shit 5120 SPs. 

How dare you insult Nvidia like that, they're CUDA cores you peasant.