Could This be the NVIDIA TITAN Volta?

[cdawall]

Far behind something that is 10-11 months away?

We aren't talking an AMD schedule rumor mill already has them coming out sometime this year. The end of the year is only 6 months away. Now an AMD timeline yes that means a year, but this is nvidia.

 

[64K]

I doubt we will see consumer Volta cards until closer to the end of the year and that's probably a bit overly optimistic. Likely they will drop Q1 2018. First will be Titan Volta for an obscene price and 2070 and 2080. Then we all get to debate about whether Nvidia will release a 2080 Ti at all. And later we get the 2080 Ti.

I may be concerned for nothing but I expect the Volta GPUs are going to be more expensive than the Pascals no matter what.

 

[cdawall]

I doubt we will see consumer Volta cards until closer to the end of the year and that's probably a bit overly optimistic. Likely they will drop Q1 2018. First will be Titan Volta for an obscene price and 2070 and 2080. Then we all get to debate about whether Nvidia will release a 2080 Ti at all. And later we get the 2080 Ti.

I may be concerned for nothing but I expect the Volta GPUs are going to be more expensive than the Pascals no matter what.

As long as the 2080 GV104 silicon performs better than the GP102 it becomes kind of mute. Also remember we still wont see a release for consumer level Vega until June/July assuming it isn't pushed back yet freaking again.

 

[Slizzo]

As long as the 2080 GV104 silicon performs better than the GP102 it becomes kind of mute. Also remember we still wont see a release for consumer level Vega until June/July assuming it isn't pushed back yet freaking again.

Supposed to be a brand new architecture on this card. I think it'll hit performance targets it should be.

All we've seen so far is a block diagram of Volta.

 

[efikkan]

Consumer versions of Volta wouldn't arrive anytime soon.

Meanwhile we are still waiting for Vega, and AMD just talked about taping out Navi at the end of this year. It looks like Vega and Volta will be competing then.

 

[cdawall]

Supposed to be a brand new architecture on this card. I think it'll hit performance targets it should be.

All we've seen so far is a block diagram of Volta.

Well I mean nvidia is probably just playing tiddly winks with themselves trying to figure out if they will ever have competition again.

Interesting part for the rumor mill is so hynix announced that they were ramping up hbm2 production for a release of a new graphics card lineup start of 2018.

Now you can take that one of two ways, nvidia will release some nice cards start of 2018 or AMD is further behind than we think.

Consumer versions of Volta wouldn't arrive anytime soon.

Meanwhile we are still waiting for Vega, and AMD just talked about taping out Navi at the end of this year. It looks like Vega and Volta will be competing then.

AMD has to release something for it to compete.

 

[ratirt]

This is really funny you guys. I remember speculation about Vega when people say that Vega might be as fast as 1080 TI. Of course this statement had an impact on all NV fanboys saying it is not possible since 1080TI is better( yeah just like that) and there's no way that Vega may not saying surpass but even come a bit closer. Leaks and other stuff prove it might be as fast or even faster. Now titan VOLTA yeah!! !!! one pick which we don't even know that it is the titan Volta and it is already faster than 1080 TI by well 25-35%. what a bullshit crap. Fell from a chair laughing really Don't you guys have any sense at all ?? Really Lame and shame

Just to add: That's the joke of the 2017 for me. Nothing is going to beat what I read in this thread ever.

 

[Nuckles56]

Vega is the joke of 2017 because at the way that AMD is taking its sweet time releasing the cards, it will be 2018 before we see them at this rate, or certainly more than like the two launch cards they had

 

[eidairaman1]

not holding breath, it will be out when it is out

 

[Prima.Vera]

It looks more and more like a V war. Volta vs. Vega....

 

[ratirt]

Vega is the joke of 2017 because at the way that AMD is taking its sweet time releasing the cards, it will be 2018 before we see them at this rate, or certainly more than like the two launch cards they had

Yeah not event out yet and a joke. Listen to yourself. Guess I was wrong. There is another a better joke thanks bro for saying it Although people heard some feedback, got rumors etc. Titan Volta You saw 1 photo that may not even be a titan Volta and it is already better for over 25-35% without facts or any feedback. It's lame to even say that. For me it is a joke how people blindly follow one camp. Maybe it is time to stop and rethink stuff bro? It's kinda sad. NV is giving titans and all fanboys follow this saying it's awesome, ripping us from money. Titans for me are worthless. Pointless like driving a Lamborghini on the Sahara desert. Sheik might buy it to keep it in his mansion and drive around. Or start it and listen to the engine running . Just for the sake of having it. That's what titans are. For me NV can shove that titan up it's ass where it came from with it's price point. Waste of tech for sky-rocking price.
Stop being delusional. Where did people get this 25-35% faster? deducted from the picture they seen which may not even be Titan Volta or I don't know listening to the aether?
I hope that AMD's Vega shows up and smoke the 1080Ti and blows fanboys pants off. Just for the sake of it

 

[medi01]

Sorry to bust bubbles, but could you remind me, how 1080 compared to 980Ti (OC vs OC). Last time I've checked the difference was about half of the 25-30% some are smoking about.

Vega is the joke of 2017 because at the way that AMD is taking its sweet time releasing the cards, it will be 2018 before we see them at this rate, or certainly more than like the two launch cards they had

Yeah, fucking underdogs, how dare they!

 

[ratirt]

Sorry to bust bubbles, but could you remind me, how 1080 compared to 980Ti (OC vs OC). Last time I've checked the difference was about half of the 25-30% some are smoking about.



Yeah, fucking underdogs, how dare they!

If titan Volta turns out to be even 50% faster I don't mind. I'd be glad that performance improved so much. Price would be enormous considering NV's greediness. But saying it without any proof or even a rumor is just stupid. making assumptions based on what nothing.
That's just silly.

 

[Caring1]

Is that pic meant to signify anything in particular, is he meant to be well known?

 

[springs113]

It looks more and more like a V war. Volta vs. Vega....

What's so ironic is that, the original road map had these 2 going head to head. According to speculation...Vega refresh has already been taped out. It would be nice to have a card in my possession but hey as soon as a new one is released 6 to 13 months later here comes another.
@Caring1 that is linus,he's got a nice following he's good just annoying lol.

 

[64K]

Sorry to bust bubbles, but could you remind me, how 1080 compared to 980Ti (OC vs OC). Last time I've checked the difference was about half of the 25-30% some are smoking about.

Not true. The reference 1080 is 37% faster than the reference 980 Ti and as far as overclocks the 1080 was able to OC for 12.8% improvement in performance in BF3 and the 980 Ti was able to to OC for 9.4% improvement in performance in BF3.

https://www.techpowerup.com/reviews/NVIDIA/GeForce_GTX_1080/30.html
https://www.techpowerup.com/reviews/NVIDIA/GeForce_GTX_980_Ti/34.html

Spoiler: Click for benches

As far as speculating on Volta Titan I would look at past performance gains for an idea.

Maxwell Titan X was 64% faster than Kepler Titan.

Pascal Titan X was 72% faster than Maxwell Titan X.

Spoiler: Click for benches

https://www.techpowerup.com/reviews/NVIDIA/GeForce_GTX_Titan_X/29.html

https://www.techpowerup.com/reviews/NVIDIA/Titan_X_Pascal/24.html

It seems reasonable to me to expect similar performance gains with Volta Titan and possibly even more due to the lower process node and new architecture.

 

[springs113]

@64K going by what you said then volta titan could possibly be 80% faster.

 

[64K]

@64K going by what you said then volta titan could possibly be 80% faster.

It could be but who knows right now. I'm expecting at the absolute minimum we should get 60% faster performance over Pascal Titan X basing that on past performance results increases. Probably we will get more. Possibly a good bit more than that but that's just my opinion. Assuming of course that this really is the Volta that we have been hearing about for years now.

 

[Slizzo]

Is that pic meant to signify anything in particular, is he meant to be well known?

That's Linus Sebastian. Runs Linus Tech Tips, formerly worked for NCIX doing NCIX Tech Tips.

@64K and @springs113 , Volta is supposed to be a new architecture, should have more gains that was was previously seen since Kepler (Maxwell was refined Kepler arch).

Remember, all we've seen of Volta so far is just a block diagram.

 

[FordGT90Concept]

...Tesla V100...

Volta (15 TFLOP), like Vega (13 TFLOP), is marketed as a deep learning chip. Both companies are bringing versions of it to consumers in (relatively) low volumes.

 

[efikkan]

Maxwell Titan X was 64% faster than Kepler Titan.

Pascal Titan X was 72% faster than Maxwell Titan X.
...
It seems reasonable to me to expect similar performance gains with Volta Titan and possibly even more due to the lower process node and new architecture.

FYI: TSMC "12nm" is not a real node shrink, as evident by GV100 having only < 3% higher transistor density vs. GP100.

 

[jabbadap]

FYI: TSMC "12nm" is not a real node shrink, as evident by GV100 having only < 3% higher transistor density vs. GP100.

Not to mention it's TSMC 12nm FFN, especially tailored for nvidia only. But yeah for the pic of the this news, I would say if it's anything real titan it would probably be pascal gp100 titan version of Quadro GP100 rather than GV100(And probably being canned by nvidia for not offering consumer/prosumer grade fp16 card).

 

[efikkan]

Not to mention it's TSMC 12nm FFN, especially tailored for nvidia only. But yeah for the pic of the this news, I would say if it's anything real titan it would probably be pascal gp100 titan version of Quadro GP100 rather than GV100(And probably being canned by nvidia for not offering consumer/prosumer grade fp16 card).

Well, I guess you can call it "tailored for Nvidia" when they is the "only" customer for the HP nodes, ever since AMD went with Samsung.

GV100 will not be a high volume chip, it's huge and the yields wouldn't be that great. GV100 is really pushing the physical limits of the node, and it was originally planned to be made on a smaller node. Not that it's that big of a problem for consumer products though, a GV102 at ~600 mm² would be fine. I still hope that GV102 and GV104 finally will bring full fp16 support though, if used, it can provide a huge boost for gaming.

 

[jabbadap]

Well, I guess you can call it "tailored for Nvidia" when they is the "only" customer for the HP nodes, ever since AMD went with Samsung.

GV100 will not be a high volume chip, it's huge and the yields wouldn't be that great. GV100 is really pushing the physical limits of the node, and it was originally planned to be made on a smaller node. Not that it's that big of a problem for consumer products though, a GV102 at ~600 mm² would be fine. I still hope that GV102 and GV104 finally will bring full fp16 support though, if used, it can provide a huge boost for gaming.

Which makes me wonder, will nvidia use that node for other volta chips too. Or will they use TSMCs vanilla 12nm FF or even 16nm FF+. For those future die size estimations it would be great if nvidia would release die shot of GV100(like they did with gp100). There's lot of things that won't be on GV102(nvlink, probably Tensor cores, 1:2 fp64 units which takes lot of space of GV100 area).

For that manufacturing process, yeah it won't be high volume chip. First thing what nvidia must deliver is Summit super computer. Summit seems to have six Tesla V100s per node. And it would have ~4600 nodes. That would make 27 600 Tesla V100 for summit super computer(Peak fp64 Petaflops from teslas only is then 207 PFlops). GV100 is huge die so they will need lot of wafers to collect that amount of working dies(TSMC has 12nm FFs in it's 12" gigafabric so it would make appr. 64 dies per wafer at max. If yield are really good that would make ~50 working die per wafer which would need 552 wafers for whole Summit. If Yields are bad like 15 working dies per wafer it would make 1840 wafers to get enough working dies).

Spoiler: pic

 

[efikkan]

Which makes me wonder, will nvidia use that node for other volta chips too. Or will they use TSMCs vanilla 12nm FF or even 16nm FF+. For those future die size estimations it would be great if nvidia would release die shot of GV100(like they did with gp100). There's lot of things that won't be on GV102(nvlink, probably Tensor cores, 1:2 fp64 units which takes lot of space of GV100 area).

Since 12nm FF is just a refinement of 16nm FF, it probably will be used for as many chips as possible.