NVIDIA skipping Volta for gamers? Ampere might be next!

[Shreyas1]

I want to see Einstein 

[Cookybiscuit]

5 hours ago, Bcat00 said:

Meh, i highly doubt it knowing how a company likes to milk everything they can out of a lineup before sending it to the bucket.

 

Let's assume someone has a product they just developed that's even slightly better than the last gen, would you want to throw it away after spending a lot of money on RnD and staff wages?

Hell no!!

You can't milk the crowd who buys the x80/Ti every year if you don't release it every year. The 2080, or whatever they are calling it, will be released by April 2018 at the latest, Pascal is well over a year old at this point.

 

1 hour ago, Phentos said:

Huang was correct in his remark that Pascal is unbeatable for gamers atm. 

He's hardly going to say "guys don't buy a 1080Ti or any of our other stuff because we will have new cards in a few months", is he?

[Master Disaster]

1 hour ago, Cookybiscuit said:

You can't milk the crowd who buys the x80/Ti every year if you don't release it every year. The 2080, or whatever they are calling it, will be released by April 2018 at the latest, Pascal is well over a year old at this point.

 

He's hardly going to say "guys don't buy a 1080Ti or any of our other stuff because we will have new cards in a few months", is he?

4 and a bit months time? Really?

 

You do realise that 4 months before Pascal launched the entire launch line up was already leaked, there was shipping manifests showing code names and even early sample benchmarks all out in the public domain. It was the same thing for Ryzen too, and Vega.

 

The writing is very much on the wall, Vega is no competition and Nvidia have known since August (when Huang commented about Pascal being a long term product because Vega wasn't a threat to it) that they have no need to release anything new for a looooong time. What exactly are consumers going to do about it if Nvidia keep Pascal for 2 years and simply keep refreshing it? If you want the absolute fastest you buy a 1080ti or Titan and that's a fact that Nvidia is all too aware of, do you really think they'd launch a 1070ti (to try and combat Vega at that tier) weeks before they knew the entire range was getting replaced anyway?

 

Face it, Nvidia are at the top by a long way and because they have no competition they have no need to try and compete anymore. Volta is raking in the big bucks in the machine learning sector and until AMD has anything even close to 1080ti (let alone Titan) Nvidia have no reason to devalue their product by doing a consumer release.

 

Pascal is here to stay, I'm guessing for the large majority of 2018 if not until early 2019. When the rumors about AMD replacing Vega start is when Nvidia will act on replacing Pascal.

[mr moose]

28 minutes ago, Master Disaster said:

4 and a bit months time? Really?

 

You do realise that 4 months before Pascal launched the entire launch line up was already leaked, there was shipping manifests showing code names and even early sample benchmarks all out in the public domain. It was the same thing for Ryzen too, and Vega.

 

The writing is very much on the wall, Vega is no competition and Nvidia have known since August (when Huang commented about Pascal being a long term product because Vega wasn't a threat to it) that they have no need to release anything new for a looooong time. What exactly are consumers going to do about it if Nvidia keep Pascal for 2 years and simply keep refreshing it? If you want the absolute fastest you buy a 1080ti or Titan and that's a fact that Nvidia is all too aware of, do you really think they'd launch a 1070ti (to try and combat Vega at that tier) weeks before they knew the entire range was getting replaced anyway?

 

Face it, Nvidia are at the top by a long way and because they have no competition they have no need to try and compete anymore. Volta is raking in the big bucks in the machine learning sector and until AMD has anything even close to 1080ti (let alone Titan) Nvidia have no reason to devalue their product by doing a consumer release.

 

Pascal is here to stay, I'm guessing for the large majority of 2018 if not until early 2019. When the rumors about AMD replacing Vega start is when Nvidia will act on replacing Pascal.

yep, if Nvidia don't release 2000's next year it will be the longest time between series releases since I don't know when. 

[Master Disaster]

18 minutes ago, mr moose said:

yep, if Nvidia don't release 2000's next year it will be the longest time between series releases since I don't know when. 

They could do the old AMD trick for the 1100/2000 series...

 

Titan XP becomes Titan X

Titan Xp becomes 11/2080ti

1080ti becomes 11/2080

1080 becomes 11/2070ti

1070ti becomes 11/2070

 

And it continues down the chain, a small boost in clock speed and/or something like GPU Boost 4.0 and suddenly Nvidia have a new lineup of cards without devaluing Volta.

[Cookybiscuit]

37 minutes ago, Master Disaster said:

-snip-

There's absolutely no way they'll wait until 2019. You say Vega isn't competitive, you seem to forget AMD has spent the last four/five years rebranding the 7000 series and it didn't stop Nvidia releasing new stuff. Yes, Pascal has a considerably greater advantage over Vega than previous Nvidia vs AMD offerings, but 1000 series launched in mid 2016, you really think they are going to make it last two and a half years until 2019? It's not like Pascal kills Vega in terms of price to performance, Vega 64 and 1080 aren't a million miles apart in either price or performance.

 

April 2018 for sure, leather jacket man won't keep us waiting too long.

[Master Disaster]

3 minutes ago, Cookybiscuit said:

There's absolutely no way they'll wait until 2019. You say Vega isn't competitive, you seem to forget AMD has spent the last four/five years rebranding the 7000 series and it didn't stop Nvidia releasing new stuff. Yes, Pascal has a considerably greater advantage over Vega than previous Nvidia vs AMD offerings, but 1000 series launched in mid 2016, you really think they are going to make it last two and a half years until 2019? It's not like Pascal kills Vega in terms of price to performance, Vega 64 and 1080 aren't a million miles apart in either price or performance.

 

April 2018 for sure, leather jacket man won't keep us waiting too long.

I'll quote myself as a response

 

12 minutes ago, Master Disaster said:

They could do the old AMD trick for the 1100/2000 series...

 

Titan XP becomes Titan X

Titan Xp becomes 11/2080ti

1080ti becomes 11/2080

1080 becomes 11/2070ti

1070ti becomes 11/2070

 

And it continues down the chain, a small boost in clock speed and/or something like GPU Boost 4.0 and suddenly Nvidia have a new lineup of cards without devaluing Volta.

I'm not saying they won't release anything in that time, only that they won't replace Pascal in that time.

[leadeater]

1 hour ago, Master Disaster said:

Volta is raking in the big bucks in the machine learning sector and until AMD has anything even close to 1080ti (let alone Titan) Nvidia have no reason to devalue their product by doing a consumer release.

Well if you compare the use case Nvidia tries to claim the Titan cards are for Vega FE is very competitive and a heck of a lot cheaper. It's really isn't going to change much though, people will still buy Titans and Quadros anyway (the Titan was actually pathetic until Vega FE re driver update making Titan not pathetic) because it's the more known and trusted brand for these buyers and some of them might be invested in CUDA.

 

Nvidia certainly has the technical lead on basically everything but that puts them in a diffcult position because they end up competing with themselves, Titan should never have existed and Quadro pricing is too high. Fixing a problem the wrong way basically.

[Cookybiscuit]

6 minutes ago, Master Disaster said:

I'll quote myself as a response

 

Maybe, but Nvidia haven't done rebrands of their high-end products in quite some time (unless you count Titan cards becoming the x80/Ti). Plus I doubt too many people are going to care about the 1080Ti getting another GB of VRAM. 1080Ti at 1080 pricing would ruin Vega though.

 

Volta in April 2018 confirmed.

[Master Disaster]

22 minutes ago, leadeater said:

Well if you compare the use case Nvidia tries to claim the Titan cards are for Vega FE is very competitive and a heck of a lot cheaper. It's really isn't going to change much though, people will still buy Titans and Quadros anyway, the Titan was actually pathetic until Vega FE re driver update making Titan not pathetic, because it's the more known and trusted brand for these buyers and some of them might be invested in CUDA.

 

Nvidia certainly has the technical lead on basically everything but that puts them in a diffcult position because they end up competing with themselves, Titan should never have existed and Quadro pricing is too high. Fixing a problem the wrong way basically.

Oh 100% agreed, look what happened to Intel when they had no competition. They sat on ass for so long that when the competition came back they weren't ready for it. 

 

I'm not exactly au fais with compute stuff but it's my understanding that CUDA pretty much has dominance over OpenCL and that Nvidia products are almost a requirement in that sector so I don't think Nvidia really care if they're competing against themselves, heck they released 2 Titan cards into the consumer market with only a few months between them. Ultimately as long as Mr Joe Customer buys Nvidia they don't care which SKU.

 

21 minutes ago, Cookybiscuit said:

Maybe, but Nvidia haven't done rebrands of their high-end products in quite some time. Plus I doubt too many people are going to care about the 1080Ti getting another GB of VRAM. 1080Ti at 1080 pricing would ruin Vega though.

 

Volta in April 2018 confirmed.

Ignoring the Titan cards (which Nvidia say are not gaming cards anyway) Nvidia haven't been so far ahead at the top in a long time either and Vega was supposed to be AMDs answer to 1080ti. It failed. A good business is able to react to current trends and base decisions on what's currently happening in the market. Is now the right time for Nvidia to release a new architecture?

 

Vega ultimately failed in its goal

RTG just lost their lead Engineer

Intel are trying to make GPUs

Intel & AMD just teamed up to try and take some market share in the mid range laptop sector

 

It honestly wouldn't surprise me if we didn't see a reversal of Nvidias usual tactics where they release Small Volta first this time around and try to hold onto the mid range market that Intel & AMD are trying to take. We know Big Volta is ready so it's not beyond the realm of possibility that they hold onto big for a while longer and release small first.

[mr moose]

6 minutes ago, Master Disaster said:

 (which Nvidia say are not gaming cards anyway)

 

Except on the Titan landing page;  "GAMING" is the most prominent word up front in the centre.

https://www.nvidia.com/en-us/design-visualization/products/titan-xp/

 

 

I don't think they'll release anything soon,  not because they want to milk the market in absence of competition, but more because a good company keeps tricks up it's sleeve so to speak.  Look what happened to Intel when AMD released Ryzen, they almost had something to release,  but instead fumbled the unfinished X299 release and brought  CL forward too quickly opening themselves up for all sorts of PR issues.  Intel were not ready and I doubt Nvidia are going to make the same mistake.  So I think at he very least they will wait until they have a completely finished product and all their ducks are lined up in a row rather than wait till AMD release something new and catch them off guard.

[Master Disaster]

4 minutes ago, mr moose said:

 

Except on the Titan landing page;  "GAMING" is the most prominent word up front in the centre.

https://www.nvidia.com/en-us/design-visualization/products/titan-xp/

 

 

I don't think they'll release anything soon,  not because they want to milk the market in absence of competition, but more because a good company keeps tricks up it's sleeve so to speak.  Look what happened to Intel when AMD released Ryzen, they almost had something to release,  but instead fumbled the unfinished X299 release and brought  CL forward too quickly opening themselves up for all sorts of PR issues.  Intel were not ready and I doubt Nvidia are going to make the same mistake.  So I think at he very least they will wait until they have a completely finished product and all their ducks are lined up in a row rather than wait till AMD release something new and catch them off guard.

By the sounds of it Nvidia have a few unreleased architectures in the pipelines already. Big Volta is out in the compute sector already and now there's rumblings of Ampere being close to ready.

 

I've said this before, unlike Intel Nvidia don't seem like the types who are happy sitting on ass doing nothing. I get the feeling they're always working on something whether they need too or not.

[mr moose]

Just now, Master Disaster said:

I've said this before, unlike Intel Nvidia don't seem like the types who are happy sitting on ass doing nothing. I get the feeling they're always working on something whether they need too or not.

 

That's good business practice.  Although I think Intel weren't sitting on their ass, I think they just got unlucky with timing/development constraints (because unlike GPU's where the sky is the limit and you can make it bigger if you have to, x86 is coming to the pointy end of silicon and power is an issue).   Had they had just a few more months to finalize everything my guess is no one would have made a big deal out of it. 

[leadeater]

25 minutes ago, Master Disaster said:

By the sounds of it Nvidia have a few unreleased architectures in the pipelines already. Big Volta is out in the compute sector already and now there's rumblings of Ampere being close to ready.

 

I've said this before, unlike Intel Nvidia don't seem like the types who are happy sitting on ass doing nothing. I get the feeling they're always working on something whether they need too or not.

 

19 minutes ago, mr moose said:

 

That's good business practice.  Although I think Intel weren't sitting on their ass, I think they just got unlucky with timing/development constraints (because unlike GPU's where the sky is the limit and you can make it bigger if you have to, x86 is coming to the pointy end of silicon and power is an issue).   Had they had just a few more months to finalize everything my guess is no one would have made a big deal out of it. 

Also unlike Intel Nvidia aren't managing two different architectures for each market segment. Nvidia might have more than one current architecture available but they don't split them up in the way Intel does, they scale them which makes it easier for them.

 

Intel has consumer CPUs and chipsets and enterprise CPUs and chipsets and those are not scaled up or down versions of each other, and even when one product line gets to the same architecture codename as the other they still aren't all that similar e.g. Skylake vs Skylake-SP.

 

Nvidia does have some differences in the biggest die they make compared to their other ones but those changes aren't a big technical challenge as it's mostly a case of taking something completely out, FP64 cores, or front end pipeline stuff that I currently forget what it is but inconsequential for gaming.

[Taf the Ghost]

13 minutes ago, leadeater said:

 

Also unlike Intel Nvidia aren't managing two different architectures for each market segment. Nvidia might have more than one current architecture available but they don't split them up in the way Intel does, they scale them which makes it easier for them.

 

Intel has consumer CPUs and chipsets and enterprise CPUs and chipsets and those are not scaled up or down versions of each other, and even when one product line gets to the same architecture codename as the other they still aren't all that similar e.g. Skylake vs Skylake-SP.

 

Nvidia does have some differences in the biggest die they make compared to their other ones but those changes aren't a big technical challenge as it's mostly a case of taking something completely out, FP64 cores, or front end pipeline stuff that I currently forget what it is but inconsequential for gaming.

The "big" GPU in Volta is pretty much Pascal + Tensor Cores, while it's shaping up that Volta is an improved node version of Pascal. I'm sure there'll be tweaks, but some of the rumbling about Volta running pretty hot would probably be at the "1180" range. To get it to up to 1080 Ti performance is going to take more clocks.

 

It's looking a lot like the core Maxwell uArch is going to span 3 full generations, which suggest Ampere might end up being bigger uArch changes. Doesn't mean there won't be some small clock for clock improvements with Volta, but it looks a lot more like more CUDA cores per level and higher clocks.

[Fetzie]

11 hours ago, techstorm970 said:

First, a Grammar Nazi will be visiting you shortly.

 

Second, I agree.  It's from WTFTech, for crying out loud!   

heise.de is a reliable source for technology news, and that's where wccftech got it from.

[leadeater]

24 minutes ago, Taf the Ghost said:

The "big" GPU in Volta is pretty much Pascal + Tensor Cores, while it's shaping up that Volta is an improved node version of Pascal. I'm sure there'll be tweaks, but some of the rumbling about Volta running pretty hot would probably be at the "1180" range. To get it to up to 1080 Ti performance is going to take more clocks.

 

It's looking a lot like the core Maxwell uArch is going to span 3 full generations, which suggest Ampere might end up being bigger uArch changes. Doesn't mean there won't be some small clock for clock improvements with Volta, but it looks a lot more like more CUDA cores per level and higher clocks.

Outside of gaming Volta even on the non Tensor stuff does actually have some big changes, have a read of the Anandtech article on it as it's rather good. But yea not much that's going to push gaming performance up other than the node shrink allowing more of everything.

[Mr.Meerkat]

12 hours ago, Himommies said:

The fact that Nvidia names their GPU's after dead pepole still bothers me.They should name it after inanimate objects like Vega

Tonga, Fiji, Hawaii, Turks, Caicos, Pitcairn, Tahiti ect. 

 

Here's a fun fact, since 2010/2011 and till Fiji (Fury X) polaris, AMD had used Islands as the codenames for their GPUs. Also in 2009/2010, AMD had used names of plants for the codenames of their GPUs  

[Notional]

10 hours ago, leadeater said:

There are Pascal cards with HBM2. Volta isn't actually a big change from Pascal for anything not Tensor, it does make sense if they have any refinements to that part of the architecture to do so for the market that primarily uses it. Volta was mainly to release a Tensor product, no real surprise it might get skipped.

Yeah, that's my point. Pascal and volta cards with HBM are only professional cards with insane price tags. And since volta doesn't provide enough of an architectural improvement, and it seems volta's memory controller is only for HBM, it might explain why they are scrapping it on the consumer side. Going by AMD's "success" with HBM, I can't blame them.

[CTR640]

nVidia indeed has some fetish for naming their products after dead people. Maybe they should name them after people that are still alive?

Neil deGrasse Tyson

Michio Kaku

 

And perhaps also Mike Rowe lol

[Taf the Ghost]

2 hours ago, leadeater said:

Outside of gaming Volta even on the non Tensor stuff does actually have some big changes, have a read of the Anandtech article on it as it's rather good. But yea not much that's going to push gaming performance up other than the node shrink allowing more of everything.

Saw a few higher level discussions about Volta changes, but it's pretty much nothing that touches gaming. Though it gets some nice uplift in Compute tasks, even without the tensor cores. Which also helps point to what's going on. To get the "1180" to 1080 Ti levels, it'll need about 15% more CUDA cores and the 15% higher clocks with the process change. Everything below the new "1170" will also probably be nothing more than a rebrand, on the assumption they launch with the rest.

59 minutes ago, Notional said:

Yeah, that's my point. Pascal and volta cards with HBM are only professional cards with insane price tags. And since volta doesn't provide enough of an architectural improvement, and it seems volta's memory controller is only for HBM, it might explain why they are scrapping it on the consumer side. Going by AMD's "success" with HBM, I can't blame them.

Nvidia's mainstream GPUs will sell in the millions, there isn't enough HBM supply to go there for a long time. Also, given the way the lower-SKU Nvidia's currently operate with memory, they bandwidth also really isn't needed. They seem to hit a saturation well before even the limits of GDDR5. 

[Agosto]

Aaaaaaaaaaand I'm still waiting for custom RX Vega 56.

Which probably won't even happen

[Nicnac]

I want my Einstein architecture already. maybe with relativity cores?

[laminutederire]

34 minutes ago, Nicnac said:

I want my Einstein architecture already. maybe with relativity cores?

You're at least certain to have relastivic prices

[Syntaxvgm]

17 hours ago, Himommies said:

The fact that Nvidia names their GPU's after dead pepole still bothers me.They should name it after inanimate objects like Vega

coming soon- the NVIDIA Edison. It'll just be a Tesla rebrand.