Place your bets on Vega

How good will fastest Vega be on "Game Play" as per [H] Review.

  • Better than 1080Ti

    Votes: 30 8.2%
  • Same as 1080Ti

    Votes: 28 7.7%
  • Better than 1080

    Votes: 126 34.6%
  • Same as 1080

    Votes: 97 26.6%
  • Better than 1070

    Votes: 42 11.5%
  • Same as 1070

    Votes: 12 3.3%
  • Complete piece of shit!

    Votes: 29 8.0%

  • Total voters
    364
Status
Not open for further replies.

Archaea

[H]F Junkie
Joined
Oct 19, 2004
Messages
11,551
The 1070 competitor thing doesn't make sense. It's 50% faster by the long provided tflops number than the Fury X.

1070 is like 20% faster than the two year old Fury X.

You can't compare tflops numbers between AMD and nvidia, but you should be able to compare tflops numbers between AMD and AMD.

RX480 is 5.7
RX580 is 6.2
Fury X is 8.7
Big Vega is supposedly 12.5

If that 12.5 leak (that has been out for months and revoiced innumerable times) is correct - then there isn't much shadow of doubt. It's a tad bit better performance over a 1080 and falls behind a 1080ti by about 25-30%.
 

N4CR

Supreme [H]ardness
Joined
Oct 17, 2011
Messages
4,921
I think average 1080-Ti range and in 4k and similar high res/DX12/Vulkan will take the lead in quite a few titles.
Lets face it, a 471mm² vs 500mm²+ Vega already evens the uarch differences a little.
 

Shintai

Supreme [H]ardness
Joined
Jul 1, 2016
Messages
5,678
The 1070 competitor thing doesn't make sense. It's 50% faster by the long provided tflops number than the Fury X.

1070 is like 20% faster than the two year old Fury X.

You can't compare tflops numbers between AMD and nvidia, but you should be able to compare tflops numbers between AMD and AMD.

RX480 is 5.7
RX580 is 6.2
Fury X is 8.7
Big Vega is supposedly 12.5

If that 12.5 leak (that has been out for months and revoiced innumerable times) is correct - then there isn't much shadow of doubt. It's a tad bit better performance over a 1080 and falls behind a 1080ti by about 25-30%.

Memory bandwidth doesn't change. It may actually be slower as in 409GB/sec for the top bin and 358GB/sec for a lower clocked.

12.5tflops already got changed to 12tflops(1440Mhz) once before it got removed from slides completely. While 225W increased to 300W.
 

Shintai

Supreme [H]ardness
Joined
Jul 1, 2016
Messages
5,678
I think average 1080-Ti range and in 4k and similar high res/DX12/Vulkan will take the lead in quite a few titles.
Lets face it, a 471mm² vs 500mm²+ Vega already evens the uarch differences a little.

The DX12 myth is busted. For high res the advantage Fiji had is gone and its now opposite.

A 1080ti is close to twice as fast as Fiji in 4K. And Vega wont have any faster memory, it may be slower.

You may dream of something like 1.7-2Ghz Vega with 1TB/sec memory from 4 2Ghz HBM2 stacks ;)
upload_2017-5-6_16-23-57.png

perfrel_3840_2160.png
 
Last edited:

tviceman

Limp Gawd
Joined
Oct 11, 2007
Messages
398
I think average 1080-Ti range and in 4k and similar high res/DX12/Vulkan will take the lead in quite a few titles.
Lets face it, a 471mm² vs 500mm²+ Vega already evens the uarch differences a little.

There are three fallacies to your argument. 1 - Isn't big Vega going to be a hybrid DP / SP die like GK110 and GF110 were? If so, it will need to be significantly larger (25-30%) than GP102 to be in the same league in real-world game performance. When comparing GP102 to GP100, GP100 is 25% larger in and of itself but has the same core count and cache size. 2 - Polaris 10 is 16% larger in die size and uses 30% more transistors to equal the same performance. 3 - Power draw. The GTX 1060 is 40-70% more efficient than Polaris 10. If AMD halves the efficiency gap with Vega vs. GP102, then at equal performance it's power draw will be right at the 300 watt wall single card, single GPU's are officially limited by the PCIe spec.

Vega can and may turn out good, but it's not as simple as die size = performance. There are some serious technical hurdles AMD will have to overcome if it is to be truly competitive AND sustaining (profitable).
 

Comixbooks

Fully [H]
Joined
Jun 7, 2008
Messages
18,358
They are going to Fry everyone's motherboards and SSDs then start the case on fire due to a faulty fan.
 

whateverer

[H]ard|Gawd
Joined
Nov 2, 2016
Messages
1,580
Is Vega supposed to have higher IPC than Polaris? Otherwise I just can't see these promised numbers happening.

2x the fp throughput with only 70% more units = where the fuck are they going to get the clock bump from?
 
Last edited:

spine

2[H]4U
Joined
Feb 4, 2003
Messages
2,679
We saw Vega running Doom at 1080 levels months ago, so if they want to release a part at this performance level, where the fuck is it already?

I'm guessing the answer is they're having serious yield issues, so don't expect high clocks! I get the feeling AMD is about to have their 'nvidia 480' dog... :sour:
 

Simplyfun

Gawd
Joined
Dec 17, 2016
Messages
1,012
We saw Vega running Doom at 1080 levels months ago, so if they want to release a part at this performance level, where the fuck is it already?

I'm guessing the answer is they're having serious yield issues, so don't expect high clocks! I get the feeling AMD is about to have their 'nvidia 480' dog... :sour:

Ha, that's funny because Hardware Canucks swore up and down we shouldn't buy a 5870 and wait for that POS instead.... You can tell they were paid off by Nvidia. Buy what works at the time you need to spend money, don't buy into this cycle of paid reviews.
 

257Tony

Gawd
Joined
May 18, 2011
Messages
723
Better than 1080, $449 MSRP. Anything else performance wise, or any more expensive and it's a failure. I'll buy if it hits my expected mark.
 

noko

Supreme [H]ardness
Joined
Apr 14, 2010
Messages
6,720
The DX12 myth is busted. For high res the advantage Fiji had is gone and its now opposite.

A 1080ti is close to twice as fast as Fiji in 4K. And Vega wont have any faster memory, it may be slower.

You may dream of something like 1.7-2Ghz Vega with 1TB/sec memory from 4 2Ghz HBM2 stacks ;)
View attachment 24294
perfrel_3840_2160.png
Thanks for posting that. FuryX and 1070 are 3% apart - It would be a very far stretch to think Vega will even be at 1070 level. More likely between 1080 and 1080Ti at stock speeds. No one knows yet if there will be any headroom after that allowing much greater performance even over a 1080Ti. Always a chance for that. Vega is not Polaris and Polaris is already hitting 1500mhz. The size of Vega makes it a less dense chip (made for speed?), potential is there. AMD really can't release a 350w stock card but they can release a 250w stock card and make sure it can be OC, cooled etc. for 400w+ for the true Enthusiast that wants to max it out, damn the torpedoes, full speed ahead. Why did AMD put all of the OCing software into their drivers? Wattman. Not to mention RyZen Master. We just have to see what has been cooking.

I think you're right though tying the memory bandwidth and what will come with it - If Vega is not substantially different in how it uses memory, memory bandwidth saving processes then it may well indeed be limited by memory bandwidth even with HBM2.

Now Vega 20 (correction Vega 10 x2) is suppose to appear 2nd half of this year, biggest difference I see with Vega 20 is single interposer with two GPU's, 4 stacks of HBM2 memory, memory cache will treat all 16gb as one space. The real question is will Vega 20 appear to be one gpu to the api vice two? If so then even Volta will not catch it.
 
Last edited:

N4CR

Supreme [H]ardness
Joined
Oct 17, 2011
Messages
4,921
There are three fallacies to your argument. 1 - Isn't big Vega going to be a hybrid DP / SP die like GK110 and GF110 were? If so, it will need to be significantly larger (25-30%) than GP102 to be in the same league in real-world game performance. When comparing GP102 to GP100, GP100 is 25% larger in and of itself but has the same core count and cache size. 2 - Polaris 10 is 16% larger in die size and uses 30% more transistors to equal the same performance. 3 - Power draw. The GTX 1060 is 40-70% more efficient than Polaris 10. If AMD halves the efficiency gap with Vega vs. GP102, then at equal performance it's power draw will be right at the 300 watt wall single card, single GPU's are officially limited by the PCIe spec.

Vega can and may turn out good, but it's not as simple as die size = performance. There are some serious technical hurdles AMD will have to overcome if it is to be truly competitive AND sustaining (profitable).

Of course die size isn't everything, but it at the very least helps from performance comparison POV. Who knows how they will build Vega. Likely will be a hybrid though unless they have some cheap way to do so. If they pull out mid 500s (which isn't unheard of) they can do it already. It'll cost a bit more though, however you won't see all those mining farm sales offsetting that on steam surveys, so AMD sucks right?
GP100 can do more if taken advantage of because of that, the reason it's not there in GP102 is profits and cost saving. Polaris 10 is faster than 1060 in some titles, so that must mean nvidya sucks then too? That would make it incredibly inefficient in those scenarios to be beaten by 'such a pos' like Polaris 10. Ball swings both ways but Nvidia fanboys try paint it as if Nvidia shits rainbows.

Shintai, I find it funny how an old, wheezing, inefficient 28nm Fury X manages to hang up there with 16nm 1080-1070 already, a generation behind all those much newer cards, with just a little shitty 4gb memory pool, no magic gddr10+ and yet you think Vega is going to be slower than a 1080 or incapable of touching a Ti in some titles?
Do you think they really dragged it all this time to do 1070-1080 speed, the same as a Fury?
Sure they got some catching up to do in efficiency on average, but Nvidia also suffers this issue in some cases too.

Thanks for posting that. FuryX and 1070 are 3% apart - It would be a very far stretch to think Vega will even be at 1070 level. More likely between 1080 and 1080Ti at stock speeds.
This exactly.

Now Vega 20 (correction Vega 10 x2) is suppose to appear 2nd half of this year, biggest difference I see with Vega 20 is single interposer with two GPU's, 4 stacks of HBM2 memory, memory cache will treat all 16gb as one space. The real question is will Vega 20 appear to be one gpu to the api vice two? If so then even Volta will not catch it.
Hopefully they pull this off eventually, they've dropped hints for a while. This plus a potential SSG in Navi could blow the doors open for some crazy new game/software development paths.
That said I'd be very surprised if they made Vega 20 in this manner... who knows. It is much quicker than any company making a dual gpu card in past that I can remember, in terms of release cycles - so there is precedence for something unusual afoot. It would be a stunning ball play however I don't see how they are getting the chips to do it, unless they're already binning them now.......
 

TaintedSquirrel

[H]F Junkie
Joined
Aug 5, 2013
Messages
11,155
The DX12 myth is busted. For high res the advantage Fiji had is gone and its now opposite.

A 1080ti is close to twice as fast as Fiji in 4K. And Vega wont have any faster memory, it may be slower.

You may dream of something like 1.7-2Ghz Vega with 1TB/sec memory from 4 2Ghz HBM2 stacks ;)
But the Fury X is still faster than the 980 Ti at 4K. It actually BEAT the 1070 in Prey.
It stands to reason whatever improvements Nvidia made to get more gains at 4K, AMD could also do the same and Vega would show equivalent or larger margins at higher res.

Although it seems the 1080 Ti is the only Pascal GPU that increases its performance gap over AMD when going from 1080p to 4K. May be the first time Nvidia has pulled that off since the 7970 / GTX 680 days.
 
Last edited:

Shintai

Supreme [H]ardness
Joined
Jul 1, 2016
Messages
5,678
There are three fallacies to your argument. 1 - Isn't big Vega going to be a hybrid DP / SP die like GK110 and GF110 were? If so, it will need to be significantly larger (25-30%) than GP102 to be in the same league in real-world game performance. When comparing GP102 to GP100, GP100 is 25% larger in and of itself but has the same core count and cache size. 2 - Polaris 10 is 16% larger in die size and uses 30% more transistors to equal the same performance. 3 - Power draw. The GTX 1060 is 40-70% more efficient than Polaris 10. If AMD halves the efficiency gap with Vega vs. GP102, then at equal performance it's power draw will be right at the 300 watt wall single card, single GPU's are officially limited by the PCIe spec.

Vega can and may turn out good, but it's not as simple as die size = performance. There are some serious technical hurdles AMD will have to overcome if it is to be truly competitive AND sustaining (profitable).

Vega 10 isn't a DP chip, that's for the Vega 20 in 2019 or so.
 

Shintai

Supreme [H]ardness
Joined
Jul 1, 2016
Messages
5,678
Thanks for posting that. FuryX and 1070 are 3% apart - It would be a very far stretch to think Vega will even be at 1070 level. More likely between 1080 and 1080Ti at stock speeds. No one knows yet if there will be any headroom after that allowing much greater performance even over a 1080Ti. Always a chance for that. Vega is not Polaris and Polaris is already hitting 1500mhz. The size of Vega makes it a less dense chip (made for speed?), potential is there. AMD really can't release a 350w stock card but they can release a 250w stock card and make sure it can be OC, cooled etc. for 400w+ for the true Enthusiast that wants to max it out, damn the torpedoes, full speed ahead. Why did AMD put all of the OCing software into their drivers? Wattman. Not to mention RyZen Master. We just have to see what has been cooking.

I think you're right though tying the memory bandwidth and what will come with it - If Vega is not substantially different in how it uses memory, memory bandwidth saving processes then it may well indeed be limited by memory bandwidth even with HBM2.

Now Vega 20 (correction Vega 10 x2) is suppose to appear 2nd half of this year, biggest difference I see with Vega 20 is single interposer with two GPU's, 4 stacks of HBM2 memory, memory cache will treat all 16gb as one space. The real question is will Vega 20 appear to be one gpu to the api vice two? If so then even Volta will not catch it.

Timespy and Firestrike. Vega is nowhere near. Its currently barely 10% faster than Fury X at 1200Mhz and 1400Mhz HBM2.
 

Shintai

Supreme [H]ardness
Joined
Jul 1, 2016
Messages
5,678
But the Fury X is still faster than the 980 Ti at 4K. It actually BEAT the 1070 in Prey.
It stands to reason whatever improvements Nvidia made to get more gains at 4K, AMD could also do the same and Vega would show equivalent or larger margins at higher res.

Although it seems the 1080 Ti is the only Pascal GPU that increases its performance gap over AMD when going from 1080p to 4K. May be the first time Nvidia has pulled that off since the 7970 / GTX 680 days.

Vega 10 will in the absolute most utopian case have the same bandwidth as Fiji. But currently we have only seen 1.4Ghz HBM2, aka 358GB/sec. And Hynix or any other doesn't sell above 1.6Ghz or 409GB/sec if you wish.

So unlike Fiji, there is no large bandwidth advantage anymore.

And even a 256bit GV104 Volta later this year may sport 448GB/sec with 14Ghz GDDR5X/GDDR6.
 
Last edited:

TaintedSquirrel

[H]F Junkie
Joined
Aug 5, 2013
Messages
11,155
There are AMD GPUs with lower bandwidth that still show larger gains at high res compared to Nvidia GPUs with more bandwidth. So I don't think it's entirely memory related.
 

Shintai

Supreme [H]ardness
Joined
Jul 1, 2016
Messages
5,678
There are AMD GPUs with lower bandwidth that still show larger gains at high res compared to Nvidia GPUs with more bandwidth. So I don't think it's entirely memory related.

Like what? Its not the 480 that scales worse than the 1060 despite having more memory bandwidth. A GTX970 also scales better than R9 290. Dont tell me its a Kepler GPU with half assed developer support.

And its an AMD title on top.
http://www.amd.com/en-us/markets/game/featured/prey#
 

Shintai

Supreme [H]ardness
Joined
Jul 1, 2016
Messages
5,678
So the Chinese on the NCU.

NCU and the general structure of CU different, but different places, mainly engaged in different FP16 obvious efficiency and depth of learning IN8 above, But the FP32 performance did not change much. Even with the same single-precision FP32 graphics performance almost no improvement''
 

whateverer

[H]ard|Gawd
Joined
Nov 2, 2016
Messages
1,580
Well then, with zero IPC improvement it comes down to clock to hit those stratospheric numbers. 1.7 GHz core clock, here we come?

RX 580 at 1340MHz boost and 2304 processors is rated for 6.1 TFLOPs

Leak from page three:

  • 4,096 stream processors
  • 12.5 TFLOPS / 25 (FP16) TFLOPS
VEGA;s 12.5 TFLOPS / Polaris 6.1 TFLOPS = 105% fp performance increase over Polaris 20, according to leaks.

stream processors increase = 4096 / 2304 = 77% raw units increase.

So the necessary clock speed increase to get us up to 105% more performance over Polaris 20 is:

105% / 77% = 36% CLOCK SPEED IMPROVEMENT REQUIRED.

Polaris 20's 1340 MHz boost clock * 1.36 = 1820 MHz. Let's give them a pretty impressive 10% efficiency improvement, and we still still need 1.675 GHz to deliver those numbers.

It's either that, or we have a performance efficiency increase on the same order as Maxwell was. Yeah, not buying that one.
 

Shintai

Supreme [H]ardness
Joined
Jul 1, 2016
Messages
5,678
12.5/25tflops died long ago ;)

It was then 750Gflops FP64, 12tflop FP32 and 24tflop FP16 before it vanished (1440Mhz). And now we are down to 1200Mhz(225W) and 1346Mhz(300W).
 

whateverer

[H]ard|Gawd
Joined
Nov 2, 2016
Messages
1,580
Then what are the rumors now? Because without stupid rumors to shoot down, there's no point to me posting in this thread.
 

Shintai

Supreme [H]ardness
Joined
Jul 1, 2016
Messages
5,678
Then what are the rumors now? Because without stupid rumors to shoot down, there's no point to me posting in this thread.

You have the firestrike and timespy for 1200Mhz 1.4Ghz HBM2 that's the supposed 225W part.

There should be a 1346Mhz part too, perhaps with 1.6Ghz HBM2 at 300W. However some say this may not even be released as a gaming card, but HPC only for now.
 

Shintai

Supreme [H]ardness
Joined
Jul 1, 2016
Messages
5,678
It was clear with the Polaris release that there didn't seem to be much going for it despite the shrink.

Looking at the 470 and 570 as "half" cards you see 926/1206Mhz for 470 at 120W and 1168/1244Mhz for 570 at 150W.

And that costed big time for the perf/watt.
perfwatt_1920_1080.png


Same for 580.
perfwatt_1920_1080.png
 
Last edited:

GoldenTiger

Fully [H]
Joined
Dec 2, 2004
Messages
24,074
Ehhh, the HD 7970 was a good card and so was Hawaii(especially the AIB models). That was it, though...
7970 was outmoded by the gtx 680 pretty quickly. The gtx 680 was fifty bucks cheaper and around ten pct faster, while running cooler and quieter.
 

capnstabn

Gawd
Joined
Jan 6, 2006
Messages
534
I bet Vega will be basically a 1080 with +5% perf on DX12, -5% on DX11 and a TDP of ~300 watt

Nail on the head. This is how things always turn out when AMD releases new competitive GPUs. They usually get close to NVIDIA's high end, with a good price, and some driver issues.
 

SpeedyVV

Supreme [H]ardness
Joined
Sep 14, 2007
Messages
4,211
Nail on the head. This is how things always turn out when AMD releases new competitive GPUs. They usually get close to NVIDIA's high end, with a good price, and some driver issues.

Yes, but wait 4 years and you'll find out that it will become faster than your old 1080! :sneaky:
 

Rvenger

2[H]4U
Joined
Sep 12, 2012
Messages
2,719
What? Not sure if you're being sarcastic.


Not sure if you are being sarcastic. Anyways off topic now.

I voted Vega as being a piece of shit but I for one can admit that AMD has made better cards in the past. Especially the 7970.
 

tungt88

2[H]4U
Joined
Jan 14, 2008
Messages
2,099
Not sure if you are being sarcastic. Anyways off topic now.

I voted Vega as being a piece of shit but I for one can admit that AMD has made better cards in the past. Especially the 7970.

If anything, AMD was too conservative on the 7970's clocks -- they should've been clocked at 1GHZ from the get-go, which would've really taken the wind out of the GTX 680's sails.

To their credit, they clocked the FuryX very highly on release, but with much lesser results (for obvious reasons) -- AMD had to, otherwise it wouldn't have been even a decent improvement over the R9 290x/390x.

Now the 5850/5870 lineup was nice and competitive with NVIDIA's offerings, aided by the sadness that was the GTX 480.

In this time? I'm still holding out until Vega hits, but I am thinking of a nice custom GTX 1080 Ti -- possibly a top-end Zotac model, or Gigabyte. MSI and EVGA are contenders, too.
 

Simplyfun

Gawd
Joined
Dec 17, 2016
Messages
1,012
If anything, AMD was too conservative on the 7970's clocks -- they should've been clocked at 1GHZ from the get-go, which would've really taken the wind out of the GTX 680's sails.

To their credit, they clocked the FuryX very highly on release, but with much lesser results (for obvious reasons) -- AMD had to, otherwise it wouldn't have been even a decent improvement over the R9 290x/390x.

Now the 5850/5870 lineup was nice and competitive with NVIDIA's offerings, aided by the sadness that was the GTX 480.

In this time? I'm still holding out until Vega hits, but I am thinking of a nice custom GTX 1080 Ti -- possibly a top-end Zotac model, or Gigabyte. MSI and EVGA are contenders, too.


You're probably right about the past interpretation. I'd like to point out that some enthusiast sites told people to hold off buying the 5870 and wait for the 480 because ti would be so great. It wasn't great. The issue here is the shoe could be very well on the other foot... Vega could suck. No one really knows, and these folks pretending they do might as well stick to their Bronie parties. But of the course the percentage that are correct in their future forecast will crow to the rooftops about their massive intellect and how right they were, with no mention of how lucky their guess was, win or lose. I suspect we'll see a massive amount of edited posts on launch day.

All I know is; Vega will impact pricing if there's decent launch quantity. That's all that interests me personally is the trickle down pricing impact new launches generally cause. Money talks, the rest of this guesswork, walks. Hopefully any pricing adjustments among the only two players in the game benefit our wallets.
 
Status
Not open for further replies.
Top