RX VEGA - Too shine will need Dev Support

[Mayh3M]

Desperate stuff.

I almost feel sorry for you guys. Almost - were it not for the fact that you over hyped Fury, over-hyped Polaris, and continue to over-hype Vega, based on nothing but fantasy.

I wonder how you'll react when Nvidia release Volta six months from now with 50 percent improved performance over Pascal. You will hype Navi to oblivion, claiming it will miraculously double Vega's performance.

 

[beany_bot]

Desperate stuff.

I almost feel sorry for you guys. Almost - were it not for the fact that you over hyped Fury, over-hyped Polaris, and continue to over-hype Vega, based on nothing but fantasy.

I wonder how you'll react when Nvidia release Volta six months from now with 50 percent improved performance over Pascal. You will hype Navi to oblivion, claiming it will miraculously double Vega's performance.

Can tell if massive fanboy or just massive troll?

P.S. Polaris is an EXCELLENT GPU btw. superb for it's segment in the market. I loved my 480. Only sold it because mining madness.

 

[opethdisciple]

Desperate stuff.

I almost feel sorry for you guys. Almost - were it not for the fact that you over hyped Fury, over-hyped Polaris, and continue to over-hype Vega, based on nothing but fantasy.

I wonder how you'll react when Nvidia release Volta six months from now with 50 percent improved performance over Pascal. You will hype Navi to oblivion, claiming it will miraculously double Vega's performance.

That's the nature of tech in general tho.

Take me for example. I own a 970. If I go and buy a 1080 at £500 what you say above will also apply to me.

 

[Molesy79]

Can tell if massive fanboy or just massive troll?

P.S. Polaris is an EXCELLENT GPU btw. superb for it's segment in the market. I loved my 480. Only sold it because mining madness.

I'm going with Troll on that one

 

[pmc25]

Desperate stuff.

I almost feel sorry for you guys. Almost - were it not for the fact that you over hyped Fury, over-hyped Polaris, and continue to over-hype Vega, based on nothing but fantasy.

I wonder how you'll react when Nvidia release Volta six months from now with 50 percent improved performance over Pascal. You will hype Navi to oblivion, claiming it will miraculously double Vega's performance.

A FuryX is broadly comparable with a 980Ti, its equivalent.

Polaris is directly comparable with the 1060, its equivalent.

Overhype? You mean like Volta at the beginning of February, and 50% improved performance over Pascal?

Soonest it's likely to be is May or June '18. Trickle of Founders Editions could possibly just about leak out in April.

GDDR6 production only begins in Q1 '18, and Chinese New Year is in early February ... figure that one out vis a vis Volta.

50% ... right. TSMC 12nmFF is effectively their 16nmFF, just higher power. You'll likely see higher clocks, higher power consumption, and almost no improvement in density. Hence why V100 is going to be so massive. It wouldn't be 900mm2 if as the nomenclature implies it was a 25% shrink over their 16nmFF. Add in the fact that they're going to have to add FP16 capability and other things back in, as AMD are beginning to hit them there, and a ton of transistors to cope with higher clocks, and you're looking at a struggle to improve IPC, as with Maxwell to Pascal. 10% improvement at equal clocks is likely totally impossible. -5% to +5% is a realistic range. They'll be looking to hit nosebleed clocks to make it work, since there's no way they're going to increase die sizes significantly in the consumer space. I guess you're hoping for 3Ghz overclocks to hit your 50% forecast? Regardless of AMD's difficulties or lack of, NVIDIA are not in a particularly good position with Volta ... TSMC 10nm and 7nm in the form they'd need are absolutely nowhere near ready. H2 2019 might be optimistic. If AMD do get NAVI out of the door in H2 '18, NVIDIA won't have any response for a long time.

 

[Bazzards]

Nvidia will just issue a refresh of Pascal to bump performance if they are having issues with consumer Volta.

 

[GoogalyMoogaly]

Raja in one of the recent AMD videos go's onto say Vega is here for next 20 years what does he mean by that? surely I understand 20 years on from now RX vega isn't going to be playing them titles very well so what does he mean?
https://youtu.be/-DpkVnvZ8hc?t=48

I think what he means is, based on what AMD have been doing recently, that they'll just be re-badging RX Vega each generation for the next 20 years. stick a new number on the end and up the price a little.

On a more serious note, the games AMD used on those slides (not sure which thread they're in now, but I'm sure it's one of the Vega ones), are they all games that typically favour AMD (either through optimisation or DX12/Vulkan support)?
My reason for asking is they don't exactly make RX Vega look great against the 1080 (considering the 1080 has been out over a year) and my concern is that AMD picked the games/scenarios that make RX Vega look as good as possible (like when they only release Cinebench benchmark with Ryzen). If so, what is RX Vega going to perform like in other games?
I hope they don't do like they did with Ryzen and have the NDA last until release, it'd be nice to have some independent reviews running a broad range of games and doing the normal analysis in advance. The worry now is that they'll have very limited stock so you'll need to pre-order but won't actually know what you're buying.

 

[shankly1985]

Desperate stuff.

I almost feel sorry for you guys. Almost - were it not for the fact that you over hyped Fury, over-hyped Polaris, and continue to over-hype Vega, based on nothing but fantasy.

I wonder how you'll react when Nvidia release Volta six months from now with 50 percent improved performance over Pascal. You will hype Navi to oblivion, claiming it will miraculously double Vega's performance.

I almost feel sorry for you for thinking i care about what nvidia is doing/brings All i care about is my own upgrade path.

 

[LoadsaMoney]

It'll never shine, no matter how much its polished.

 

[shankly1985]

It'll never shine, no matter how much its polished.

Lets wait for FarCry 5 and Wolfenstein 2 Two games that we know that will use the RX Vega features

 

[Martini1991]

I almost feel sorry for you for thinking i care about what nvidia is doing/brings All i care about is my own upgrade path.

Why wouldn't you care what Nvidia is doing or bringing to the table? To act like they don't exist for a pro AMD approach is pure fanboyism.

I have no interest in buying Nvidia. I'm very well aware of what they're offering however which is why I find vega to be such a failure. And would rather not purchase any card.

 

[Killer7xx]

Desperate stuff.

I almost feel sorry for you guys. Almost - were it not for the fact that you over hyped Fury, over-hyped Polaris, and continue to over-hype Vega, based on nothing but fantasy.

I wonder how you'll react when Nvidia release Volta six months from now with 50 percent improved performance over Pascal. You will hype Navi to oblivion, claiming it will miraculously double Vega's performance.

And what if Navi is double Vega's performance? It wasn't so long ago that people were saying that Zen would be Sandy Bridge performance as AMD had too many years of Intel dominance to catch up with. The fight in the GPU sector has been a lot closer between AMD and Nvidia than in the CPU sector between Intel and AMD.

 

[Molesy79]

And what if Navi is double Vega's performance? It wasn't so long ago that people were saying that Zen would be Sandy Bridge performance as AMD had too many years of Intel dominance to catch up with. The fight in the GPU sector has been a lot closer between AMD and Nvidia than in the CPU sector between Intel and AMD.

The way I see it is that it take AMD 3 tries to catch up, Bulldozer, Pile-driver now Ryzen, so if the GPU's are the same then Fiji, Vega then Navi. Let all hope Navi will be the jump that AMD need!

 

[pmc25]

Nvidia will just issue a refresh of Pascal to bump performance if they are having issues with consumer Volta.

1) They're not going to get significantly higher clocks out of Pascal.

2) For what purpose? They need Volta for features and performance, and as soon as possible. A slight stock clock bump on Pascal won't do that. Volta is only partially to compete with Vega (Machine Learning and Workstation stuff), consumer (gaming) & HPC it's absolutely not. Volta wasn't designed to compete with NAVI, but that's what it will be competing with for most of its life. Since they are going to be 2 whole nodes behind AMD, they need Volta to be on the market some time before NAVI, because it probably won't be able to compete. A slight clock bump on Pascal would achieve absolutely nothing.

The whole 'Poor Volta' thing is predicated on them being stuck 2 whole nodes behind AMD ... TSMC 12nm is basically a higher power version of their 16nm. TSMC are currently moving to 10nm, and thereafter 7nm. TSMC 7nm is the node that is comparable in size with GF/IBM 7nm, not 16/12, and not 10nm. NAVI will be on 7nm.

 

[Rroff]

50% ... right. TSMC 12nmFF is effectively their 16nmFF, just higher power. You'll likely see higher clocks, higher power consumption, and almost no improvement in density. Hence why V100 is going to be so massive. It wouldn't be 900mm2 if as the nomenclature implies it was a 25% shrink over their 16nmFF. Add in the fact that they're going to have to add FP16 capability and other things back in, as AMD are beginning to hit them there, and a ton of transistors to cope with higher clocks, and you're looking at a struggle to improve IPC, as with Maxwell to Pascal. 10% improvement at equal clocks is likely totally impossible. -5% to +5% is a realistic range. They'll be looking to hit nosebleed clocks to make it work, since there's no way they're going to increase die sizes significantly in the consumer space. I guess you're hoping for 3Ghz overclocks to hit your 50% forecast? Regardless of AMD's difficulties or lack of, NVIDIA are not in a particularly good position with Volta ... TSMC 10nm and 7nm in the form they'd need are absolutely nowhere near ready. H2 2019 might be optimistic. If AMD do get NAVI out of the door in H2 '18, NVIDIA won't have any response for a long time.

Um what? both TSMC and GF are (currently) in about equal position on 7nm with tapeouts commencing, 12FF has some reasonable gains over 16FF the version nVidia is using can do straight up 10% frequency increase or 25% power reduction or with certain techniques and using different libraries a ~33% decrease in area or potentially upto almost 33% increase in performance (there are other compromises to get that ~33% however i.e. power scaling - I believe it also has implications for yields due to more complicated masking).

EDIT: I have no idea where you get your information from for instance:

TSMC’s roadmap currently points to a mid 2018 commercial availability for 7nm processors
...
TSMC has already showcased a 7nm SRAM chip, a key milestone on the road to more complicated SoC circuits, and states that it is seeing “healthy” yields from its process. The company has also recently been reportedly testing a 7nm, 12 CPU core processor developed in conjunction with MediaTek.

Small volume risk production of a more advanced 7nm+ process is expected to appear by June 2018

GlobalFounderies is looking towards the second half of 2018 as its launch window for its first commercial 7nm products. GlobalFounderies is expected to finalize its production facilities in the second half of 2017. The foundry has already begun producing test wafers at its Fab 8 in Malta, New York.

10nm was nowhere near ready for what they wanted with Volta but 7nm is shaping up fine.

 

[pmc25]

Um what? both TSMC and GF are (currently) in about equal position on 7nm with tapeouts commencing, 12FF has some reasonable gains over 16FF the version nVidia is using can do straight up 10% frequency increase or 25% power reduction or with certain techniques and using different libraries a ~33% decrease in area or potentially upto almost 33% increase in performance (there are other compromises to get that ~33% however i.e. power scaling).

EDIT: I have no idea where you get your information from for instance:

It is not a shrink. It's just SHP 16nmFF.

The only two foundries that will have higher power processes smaller than 14/16nm are Intel with their 10nm and GF/IBM with 7nm in the near future. Intel 10nm and GF 7nm are similar density. GF/Samsung 14nm & TSMC 16nm are very similar density. Samsung 10nm & TSMC 10nm are very similar density. Samsung 7nm / GF 7nm / TSMC 7nm / Intel 10nm are very similar density.

Ask yourself why NVIDIA will be using TSMC '12nm' instead of 10nm or 7nm.

The reason is that TSMC will only have SLP 10nm this year, and 7nm SLP at some point in H2 next year. Their 10nm and 7nm processes suitable for GPUs, and particularly the monster frequencies that NVIDIA require are nowhere remotely close.

Why do you think Volta will be a full 2 nodes behind NAVI, and Volta on a process that solely NVIDIA are going to be using? They're prepared for probably 5-10% yields on V100 yet they won't go for a new node out of cost savings? Pull the other one. They wouldn't be on the old node if they had a choice. Their only option is 12nm, or SHP 16nm as it really is.

 

[Rroff]

nVidia had commercial obligations with regard to Volta (see Summit) - that they couldn't match the required thermal/power characteristics with straight up 16FF and 10nm wasn't ready in time in a shape they needed but 12FF let them hit that - don't mistake that for any other implications - it won't be the case with 7nm which is progressing rapidly and they don't have the same time constraints to deal with.

EDIT: At this point I would bet on TSMC before GF to have a fully working 7nm high performance node for GPUs - GF don't exactly have the greatest historical form for getting the best results in a timely fashion even when supposedly cloning 1:1 from another company and IIRC their 7nm is basically IBM's.

 

[pmc25]

nVidia had commercial obligations with regard to Volta (see Summit) - that they couldn't match the required thermal/power characteristics with straight up 16FF and 10nm wasn't ready in time in a shape they needed but 12FF let them hit that - don't mistake that for any other implications - it won't be the case with 7nm which is progressing rapidly and they don't have the same time constraints to deal with.

????

Volta is 2 nodes behind NAVI. It is not magically going to switch from 16/12 to 10nm or 7nm.

TSMC have not even announced 10 or 7nm products suitable for NVIDIA's GPUs yet. Nor have Samsung. The soonest you will see a 10nm or 7nm consumer NVIDIA GPU is late 2019.

As I said, for the near future, anything below 14/16nm and higher power will be on either Intel 10nm or GF 7nm. You can't make products on nodes that do not exist.

 

[Cyber-Mav]

I'm finding it seriously hard to get excited for Vega... 1080 performance but for the price of a ... 1080? Err what?

With added 100w more power consumption vega will flop hard

 

[Rroff]

It is not a shrink. It's just SHP 16nmFF.

It is an optical shrink of 16FFC - same design rules but scaled down - also I believe nVidia have access to 9 track libraries over the normal 6? which is allowing them to squeeze a bit more from it.

????

Volta is 2 nodes behind NAVI. It is not magically going to switch from 16/12 to 10nm or 7nm.

TSMC have not even announced 10 or 7nm products suitable for NVIDIA's GPUs yet. Nor have Samsung. The soonest you will see a 10nm or 7nm consumer NVIDIA GPU is late 2019.

As I said, for the near future, anything below 14/16nm and higher power will be on either Intel 10nm or GF 7nm. You can't make products on nodes that do not exist.

While the more suited 7nm+ w/ EUV won't be available until about another year TSMC is catering for all market segments with their standard and HPC, etc. implementations of 7nm including GPUs like Volta - people already have products of this kind of requirements in testing with TSMC.