Nvidia thinks "Pascal is just unbeatable" and already decided not to move Volta ahead

[Misanthrope]

Nvidia's Volta Architecture will not be releasing to consumers anytime soon

 

So Jen here apparently woke up in a feisty mood recently and decided to (perhaps preemptively, perhaps adequately) throw some shade @ AMD with some statements about the upcoming holiday season

 

Quote

Volta for gaming, we haven't announced anything. And all I can say is that our pipeline is filled with some exciting new toys for the gamers, and we have some really exciting new technology to offer them in the pipeline. 

But for the holiday season for the foreseeable future, I think Pascal is just unbeatable. It's just the best thing out there. And everybody who's looking forward to playing Call of Duty or Destiny 2, if they don't already have one, should run out and get themselves a Pascal.

 

People at Overclockers seem to think this is akin to a delay but according to me Volta was always announced as early 2018 so it might technically not be more than sticking to plan but regardless:

 

Quote

Volta for consumers could be facing delays for many reasons, due to delays in manufacturing, yield problems or perhaps some other technical issue. The Volta V100 is releasing into a high margin market, which means that Nvidia can afford to push through some yield issues to deliver products early, but that is not the case for consumers. 

 

On the other hand, Nvidia could just be delaying Volta due to the lack of competition within the GPU market, as their "Pascal is just unbeatable" line showcases exactly how they see their competitors at this time. Nvidia is not in a position where they need to release new products, so it is possible that Nvidia plans on holding Volta in reserve until a later date. 

 

Keeping in mind that they basically already have working Volta chips out for the enterprise anyway (afaik) they could probably have afforded to move up the release date easily but decided that they basically have no competition. This might be just posturing a false sense of confidence for investors though: Vega 56 is show to consistently beat the 1070 in many games already despite AMD's apparently immature drivers, dodgy overclocking abilities and inadequate reference cooling. However if his confidence isn't just for show, it is indeed say that he thinks effectively stiffing progress is a good thing to show off about.

 

Source 1: https://www.overclock3d.net/news/gpu_displays/nvidia_s_volta_architecture_will_not_be_releasing_to_consumers_anytime_soon/1

Source 2 (transcript for the conference): https://seekingalpha.com/article/4097782-nvidia-nvda-q2-2018-results-earnings-call-transcript?part=single

[Taf the Ghost]

The 100 CPU always comes first for Nvidia. Their system, right now, is Computer Cards for Business first. It's only in the last 9 months that the strategy has been really profitable for them. (It took a while. It was a gamble that worked, but it was a gamble.)

 

Though Volta wasn't coming this year unless Vega smoked Pascal.

 

But, that's not actually the space that Nvidia should be worried about. The Vega SSG is, far & away, the most interesting part of the Vega stack. There's a big market in Media production, and AMD might be able to wipe Nvidia out of that market by next year. It's a product that'll be very interesting for LMG and any other production house.

[Terryv]

Are there any comparisons with the current iteration of Volta and it's equivalent in Pascal?

 

P100 vs V100

[Tiwaz]

their deciusion makes sense, i mean the vega cards are just RX cards with more stream processors and not a new architecture. the Rx 580 has the same TDP as the gtx 1080, that says a lot

[Tiwaz]

also with nvidia gtx 1080 ti launch in earlier this year releasing the new architecture any time soon would not be very profitable

[tom_w141]

Yay my 1080Ti remains relevant for longer  

[tom_w141]

1 minute ago, Tiwaz said:

their deciusion makes sense, i mean the vega cards are just RX cards with more stream processors and not a new architecture. the Rx 580 has the same TDP as the gtx 1080, that says a lot

Vega is a new arch...

[TrigrH]

Did anyone honestly expect Volta to come out soon anyway?

[xentropa]

TSMCs 7nm EUV platform must have been delayed.

[Tiwaz]

7 minutes ago, tom_w141 said:

Vega is a new arch...

then why does it have such a shitty efficiency? 

[Misanthrope]

4 minutes ago, TrigrH said:

Did anyone honestly expect Volta to come out soon anyway?

Well they never indicated 2017 afaik on any of their roadmaps. I did think however, that if Vega turned to be really competitive they at least had moving up Volta on the table but alas...

[AlTech]

1 minute ago, xentropa said:

TSMCs 7nm EUV platform must have been delayed.

Volta isn't made on 7nm. It's made on a fake 12nm process. I say fake because it's not really 12nm, it's 16nm.

[Sakkura]

4 minutes ago, Tiwaz said:

then why does it have such a shitty efficiency? 

RX Vega 56 has much better efficiency than the RX 580, so...

[Taf the Ghost]

9 minutes ago, xentropa said:

TSMCs 7nm EUV platform must have been delayed.

EUV isn't supposed to reach full production until 2018, I believe. But consumer Volta was never coming in 2017.

[captain cactus]

9 minutes ago, Tiwaz said:

then why does it have such a shitty efficiency? 

It's perf/watt is way better than Polaris, I mean Vega 56 draws the same power as an OC'd RX580 while being way faster.

 

Vega 64 is past the efficiency curve of the whole architecture. For the performance AMD wants out of it they have to push it way past that curve with high clocks and thus high power draw (also doesn't help AMD pushes a crapload more voltage than is really necessary). Vega 56 seems to be the sweetspot of the entire architecture.

[ravenshrike]

12 minutes ago, Tiwaz said:

then why does it have such a shitty efficiency? 

They're planning on shrinking Vega to a process which has a theoretical power drop of 60%. So figure 40% on the low end. At which point those power figures are just fine.

[snortingfrogs]

He is not wrong, RX Vega is pretty damn shit so why would people buy it?

[Taf the Ghost]

10 minutes ago, lots of unexplainable lag said:

It's perf/watt is way better than Polaris, I mean Vega 56 draws the same power as an OC'd RX580 while being way faster.

 

Vega 64 is past the efficiency curve of the whole architecture. For the performance AMD wants out of it they have to push it way past that curve with high clocks and thus high power draw (also doesn't help AMD pushes a crapload more voltage than is really necessary). Vega 56 seems to be the sweetspot of the entire architecture.

I think Vega 56 is more just within the far outer edge of efficiency of the process + uArch, where it sort of falls off once you get to the 64 CUs. It'll be interesting to see the detailed testing over the next 3-4 weeks.

[Taf the Ghost]

1 minute ago, snortingfrogs said:

He is not wrong, RX Vega is pretty damn shit so why would people buy it?

FreeSync monitors exist, people have them, and it's a viable product for that? Considering 56 will sell over 64 anyway, and it's a good card, it's just not as good of a value proposition.

 

Or you're just trolling.

[leadeater]

2 minutes ago, Taf the Ghost said:

FreeSync monitors exist, people have them, and it's a viable product for that? Considering 56 will sell over 64 anyway, and it's a good card, it's just not as good of a value proposition.

 

Or you're just trolling.

It's the halo effect or anti-halo effect, if the top card is shit then the ones below it must be even if they mostly aren't.

[snortingfrogs]

1 minute ago, Taf the Ghost said:

FreeSync monitors exist, people have them, and it's a viable product for that? Considering 56 will sell over 64 anyway, and it's a good card, it's just not as good of a value proposition.

 

Or you're just trolling.

Perhaps.

But the RX lineup performs bad, are expensive and is almost 1½ year late to the party, also you could use them to heat up a Siberian Mansion during winter.

[Trixanity]

7 minutes ago, ravenshrike said:

They're planning on shrinking Vega to a process which has a theoretical power drop of 60%. So figure 40% on the low end. At which point those power figures are just fine.

Source?

10 minutes ago, lots of unexplainable lag said:

It's perf/watt is way better than Polaris, I mean Vega 56 draws the same power as an OC'd RX580 while being way faster.

 

Vega 64 is past the efficiency curve of the whole architecture. For the performance AMD wants out of it they have to push it way past that curve with high clocks and thus high power draw (also doesn't help AMD pushes a crapload more voltage than is really necessary). Vega 56 seems to be the sweetspot of the entire architecture.

Very true but AMD shouldn't have to clock the GPU past its efficiency curve to keep pace. That's the problem. They even did work to add pipeline stages to make it clock higher but it almost seems like it created new problems so it wasn't worth the trouble.

Speaking of too much voltage. It seems they have a problem with validating their cards to run with optimal voltage. I think it would be wise (if possible) to have the Radeon Software run a 'calibration' of the card which will find the sweet spot for voltage while maintaining stability instead of running extra high voltage through it to make sure even the worst chip can hit the desired clock speed.

[PCGuy_5960]

Makes sense to me, Vega 64 performs like a 1080 while drawing as much power as a 1080 SLI. Vega 56 is more impressive, as it beats the 1070 while costing the same, but all Nvidia has to do to remain competitive is drop the 1070's price to $350

[ravenshrike]

1 minute ago, Trixanity said:

Source?

Navi is going to be Vega with improvements and maybe a MCM design shrunk to 7nm. According to GloFo their 7nm design can either boost performance by 40% or reduce power consumption by 60%. It's possible that they could go for the 40% performance boost, but I would think they would go for power efficiency.

[porina]

55 minutes ago, Misanthrope said:

Vega 56 is show to consistently beat the 1070 in many games already despite AMD's apparently immature drivers, dodgy overclocking abilities and inadequate reference cooling. 

AMD's main weapon has never been technical, it is in pricing. They knew they had to aim competitively in pricing. 

2 minutes ago, leadeater said:

It's the halo effect or anti-halo effect, if the top card is shit then the ones below it must be even if they mostly aren't.

At least it is working out better for them on the CPU side...