Nvidia's next GTX architecture is rumoured to be called Ampere, not Volta

[WYP]

Will Nvidia's compute and gaming cards be moving further in different directions?



Read more about Nvidia's rumoured Ampere GPU architecture.

 

[AlienALX]

I would hazard a guess and say it's about the money. Volta right now is reserved for high end stuff and high end prices. Maybe they are splitting them up to keep Volta desirable at the top end.

It could also mean that Ampere could be quite cut down, given the lack of competition from AMD. I hope not, because I hate incremental drivel like Intel were doing before they got a Ryzen chop from Lisa Su.

 

[AngryGoldfish]

I would hazard a guess and say it's about the money. Volta right now is reserved for high end stuff and high end prices. Maybe they are splitting them up to keep Volta desirable at the top end.

It could also mean that Ampere could be quite cut down, given the lack of competition from AMD. I hope not, because I hate incremental drivel like Intel were doing before they got a Ryzen chop from Lisa Su.

I agree. Volta could be considered too high-end or too expensive to produce for the GeForce market, and not necessary. Maybe Ampere is a tweaked version of Volta without Tensor cores. We never did see the full fat GP100 in the GeForce sector and no one really missed it, so maybe Volta is going to remain for the compute market where it'll be in higher demand while a 'shrunken' version will become Ampere and be a successor to Pascal that's more about DX12 features and increased clock speed and efficiency.

I don't see Nvidia pulling an Intel personally. We might not see 50-100% gain with each generation, but as long as prices are kept reasonable I think it'll be a worthy upgrade for Maxwell and even some Pascal owners. The days of 100%+ improvements between GPU generations I think are gone. For whatever reason I think it's gone. Personally I don't think we need 100% growth between generations. Games just aren't demanding enough at 1080p, the most popular resolution.

 

[Korreborg]

Well of cause it's about money. As the article says. Why spend money on Tensor cores for gamers, that can't use them.

 

[NeverBackDown]

So basically no reason to upgrade. Just a refresh on a smaller node and probably with HBM 2 on high end cards.

 

[TheF34RChannel]

I won't indulge such a cryptic rumor other than stating that the site it originated from seems to have pulled it from thin air; there's no credible information nor source.

 

[AlienALX]

I don't see Nvidia pulling an Intel personally.

I really hate saying this but I do. I totally do. At some point Nvidia realised that they were doing a tortoise and a hare (they were the hare) and they have now realised that they can slow down a bit and take the pee. As I say, I hate HATE saying this but the 1070Ti was their clear message that this is what they are doing.

They released a card on the tech they have had out for ages and ages that was pretty "meh" but did a basic purpose and that was to get sales back from AMD. I really can't see them rushing anything forward now like they have been for the past few years. From Kepler* and on they have been in a quite clear lead in some shape or form and IMO there really was no need to push on and release Pascal and etc so fast.

*I know that later on the 7970 drivers matured and put it back in front of Kepler, but the 670 for example was just so much nicer to live with. I bought a 7970 stock card at launch and it was hot, loud and uncouth. Especially when you consider that the 670 used about half of the power, got half as hot and was nearly as fast.

So yeah, I think we all need to face up to the fact that at some point, even if not right now, Nvidia are going to capitalise on their lead and release incremental bolacks.

 

[Korreborg]

 

[AngryGoldfish]

I really hate saying this but I do. I totally do. At some point Nvidia realised that they were doing a tortoise and a hare (they were the hare) and they have now realised that they can slow down a bit and take the pee. As I say, I hate HATE saying this but the 1070Ti was their clear message that this is what they are doing.

They released a card on the tech they have had out for ages and ages that was pretty "meh" but did a basic purpose and that was to get sales back from AMD. I really can't see them rushing anything forward now like they have been for the past few years. From Kepler* and on they have been in a quite clear lead in some shape or form and IMO there really was no need to push on and release Pascal and etc so fast.

*I know that later on the 7970 drivers matured and put it back in front of Kepler, but the 670 for example was just so much nicer to live with. I bought a 7970 stock card at launch and it was hot, loud and uncouth. Especially when you consider that the 670 used about half of the power, got half as hot and was nearly as fast.

So yeah, I think we all need to face up to the fact that at some point, even if not right now, Nvidia are going to capitalise on their lead and release incremental bolacks.

And what happens when Nvidia take the back seat and pull an Intel, offering 5-15% uplift between each generation? Will AMD take five years to create Navi like it took to create Ryzen? This is the biggest flaw to your argument: If Nvidia stop producing worthy investments, AMD will take over. Navi is supposedly already taped out; they're likely just waiting on things like the 7nm process. Vega offers roughly 20-40% performance increase over Fiji. If Volta/Ampere is only 10-30% performance increase over Pascal for the same price as Pascal, Navi could step up and offer 20-40% or more. Nvidia then lose money and AMD gain additional income for their R&D budget, thus increasing their performance lead even further with their next generation. Game developers code more for their GPUs and the gap widens even further.

The main reasons—at least that we can see—for Intel not offering any real tangible performance gains between architectures was, firstly, because they have a bigger mind share than Nvidia do, secondly, because Bulldozer was such a resounding failure (Polaris, Fiji, Hawaii, Vega, none of them are resounding failures; they're just a little misplaced), because higher performing CPUs (read: cores) were not necessary in the gaming sector, and finally because Ryzen took so long to come to the market. Nvidia doesn't have the mind share of Intel, is up against tougher competition from RTG (Hawaii, Fiji, Polaris, Vega, all good GPUs with a lot of potential that was sadly not fully utilised because game developers were choosing to optimise for Nvidia), and because GPU performance for gaming is in higher demand than CPU performance.

It would be foolish for Nvidia to pull an Intel now, and I don't think they're foolish. They might slow down, but that's not pulling an Intel. You'd have to stoop much, much lower to become Intel.

 

[AlienALX]

Right now we have a rough idea of what Navi is. Well, we can guess. It's multiple Polaris on one core. So right now it would be about as helpful as a wet tissue. Until they launch the tech into consoles it will be useless.

That is where AMD are headed, and I am pretty certain Nvidia made the same noises too. We are headed into a world of dual/quad core GPUs and beyond.

So until that happens? Nvidia can lap it up in the sunshine with 5% BS like Intel did for all of those years.

Honestly pulling an Intel will only lead to more $. Why give things away when you don't need to? why not just rinse it?

Edit as an example to that rinsing, right. We were promised Volta. We were even given a release date for Volta. Now it's Ampere. Was that mentioned before Vega came out? nope, didn't think so.

So they have already pulled an Intel.

 

[AngryGoldfish]

Right now we have a rough idea of what Navi is. Well, we can guess. It's multiple Polaris on one core. So right now it would be about as helpful as a wet tissue. Until they launch the tech into consoles it will be useless.

Why would they make Polaris cores when they could design a new architecture and stick those together? Other people have said Navi is going to be Polaris cores stuck together as well, but I've not seen any evidence corroborating it. Ryzen isn't Bulldozer cores stuck together. Ryzen is Ryzen cores stuck together. Navi could be Navi cores stuck together, which means that by the time Navi comes around, a GPU the same size as Polaris that costs the same to produce as Polaris will actually be significantly faster than Polaris, just like Vega 11 is likely to be faster than Polaris, just like Polaris was faster than Tahiti, and so on. Why would they use an older generation when they could develop a new small GPU (292mm²) that's as fast as Vega 56 or even faster, and stick those two together via the Infinity Fabric. Why not even use Vega 11? We have no idea how efficient that is. This Polaris idea makes no sense to me. If I've missed some piece of evidence that explains it, fine, but from what I've seen, sticking Polaris cores together would be pointless. We already know that two Polaris cores isn't faster than the 1080Ti, and we also know that four Polaris cores won't likely be as cheap or efficient as two newer cores, so why do that?

That is where AMD are headed, and I am pretty certain Nvidia made the same noises too. We are headed into a world of dual/quad core GPUs and beyond.

So until that happens? Nvidia can lap it up in the sunshine with 5% BS like Intel did for all of those years.

Honestly pulling an Intel will only lead to more $. Why give things away when you don't need to? why not just rinse it?

Again, because AMD will take over. Intel being a much larger corporation, even they have seen how one smaller company can encroach on their market space, despite being in such a stable position. Skylake-X, Coffee Lake, we all theorise that these are both direct and indirect responses to architectures that are not actually faster. Ryzen and TR aren't faster than Skylake and Coffee Lake, but they can be produced and sold for cheaper, and without supply issues. So given that Ryzen and Threadripper aren't faster than Skylake-X or Coffee Lake, yet are clawing back market share for AMD, why would Intel bother? Because they rested on their laurels and stopped innovating and suddenly needed a big push. Surely Nvidia sees this mistake as an interested observer and knows that, even now after Ryzen has launched and is actually BEHIND Intel in performance, but is still competitive and relevant, and even after Bulldozer's failures, AMD are competitive and could challenge Intel. So imagine how Nvidia are looking at AMD now with Vega. They might not be scared of it, but they realise that if they never had a Pascal to get consumers wet over, they would be in an awkward position. So if Nvidia stops offering Pascal (the concept of Pascal), they stop being as relevant and AMD takes over, slowly but surely.

Edit as an example to that rinsing, right. We were promised Volta. We were even given a release date for Volta. Now it's Ampere. Was that mentioned before Vega came out? nope, didn't think so.

So they have already pulled an Intel.

You've jumped to conclusions there and made a rather huge assumption. You're saying that, because Ampere has been rumoured by one German site, Nvidia have redacted their 'promise' of Volta—which I don't consider a promise because Pascal was not initially 'promised', and neither do I consider it a negativism as Pascal delivered something quite excellent—and instead replaced it with Ampere, which apparently is going to be inferior to Volta just because Vega didn't wipe Pascal off the map? That is a huge assumption and a kneejerk reaction. I understand what you mean—and from what we've seen there is a slight possibility of it being true—but it makes far more sense that Ampere is Nvidia's realisation that higher clock speeds are more important than Tensor cores. Or maybe Nvidia have always planned Ampere and simply never told anyone. Or maybe Ampere doesn't exist and Volta will still come to GeForce. If it did, would you still say Nvidia are pulling an Intel? This is just one rumour, and your whole argument hinges on it being true. If it's not and Volta comes to the gaming market next year and is as powerful as it is for the compute market already, your points are completely invalidated.

 

[BigDaddyKong]

One reason why AMD might use Polaris cores instead of Navi cores is because they are more for gaming. Navi was built as a compute core first, and gaming second. It's more economical to use those for now.