AMD Radeon RX Vega 56 leaked benchmarks: GTX 1070 killer

[laminutederire]

11 minutes ago, zMeul said:

 

Seems like they took minimums and not averages, which would be in accordance with the numbers you gave

 

[GatioH]

18 minutes ago, zMeul said:

the the GTX1070 numbers are pulled out of AMD's ass

some peeps on reddit got the more realistic GTX1070 numbers:

• BF1: http://www.pcgamer.com/battlefield-1-performance-analysis/
  Source: 105.3 FPS
  Leaks: 72.2 FPS
• Civilization 6: https://www.techpowerup.com/reviews/Performance_Analysis/Civilization_6/4.html
  Source: 83.8 FPS with 8x MSAA
  Leaks: 72.2 FPS with 4x MSAA
• Doom 2016: https://www.techspot.com/review/1182-nvidia-geforce-gtx-1070/page5.html
  Source: 91 FPS
  Leaks: 84.6 FPS

yeah ...

Those are 1080p benchmarks that you shared, this is 1440p...

 

edit: only bf1 was

Edited August 3, 2017 by GatioH

[cj09beira]

16 minutes ago, Lord Nicoll said:

No, that isn't, they're tested at the foundry and any less than 64 quality chips would be taken, tested, and, even though maybe one unit is dysfunctional, have the required 8 lasered off the die, as they can cause serious issues if they're present and defective. 

they do laser some off, but most of the time you can still enable other ones, for example the fury cards could go up to 60 compute units 

[Misanthrope]

If Vega was this strong AMD wouldn't be hiding their performance with blind tests and monitor rebates instead of just showing this. 

 

They would be spamming the fuck out of the numbers everywhere like they're doing with Threadripper. 

[zMeul]

18 minutes ago, laminutederire said:

Seems like they took minimums and not averages, which would be in accordance with the numbers you gave

sais who!? nowhere in the article is specifically mentioned min

[laminutederire]

8 minutes ago, zMeul said:

sais who!? nowhere in the article is specifically mentioned min

Says nobody, I looked through your numbers in your sources, and if it's minimums it's the same numbers approximately than the tweak town article, hence why I said that

 

[mr moose]

will kitguru write an article how unfair it is that tweaktown got an exclusive inside leak?  Maybe we should gather a few tweets about how unfair the world is.

[AHaskin14]

Well what happens when mining ends, 1070's go back down to $325-$350, and these have a real cost of around $425 for most aftermarket cards. That's what i imagine will happen

[zMeul]

8 minutes ago, laminutederire said:

Says nobody, I looked through your numbers in your sources, and if it's minimums it's the same numbers approximately than the tweak town article, hence why I said that

or maybe it's 

[Bananasplit_00]

i WANT this to be true, becuse that would make Vega 64 a decent buy instead of a 1080 for me, but i doubt it

[Dylanc1500]

18 minutes ago, AHaskin14 said:

Well what happens when mining ends, 1070's go back down to $325-$350, and these have a real cost of around $425 for most aftermarket cards. That's what i imagine will happen

Heck, I regularly saw a couple of them go for 295-300 before then.

[Morgan MLGman]

1 hour ago, zMeul said:

the the GTX1070 numbers are pulled out of AMD's ass

some peeps on reddit got the more realistic GTX1070 numbers:

• BF1: http://www.pcgamer.com/battlefield-1-performance-analysis/
  Source: 105.3 FPS
  Leaks: 72.2 FPS
• Civilization 6: https://www.techpowerup.com/reviews/Performance_Analysis/Civilization_6/4.html
  Source: 83.8 FPS with 8x MSAA
  Leaks: 72.2 FPS with 4x MSAA
• Doom 2016: https://www.techspot.com/review/1182-nvidia-geforce-gtx-1070/page5.html
  Source: 91 FPS
  Leaks: 84.6 FPS

yeah ...

That entirely depends on the testing environment, besides your BF1 bench is 105.3 FPS? It was tested at 1440p as said in OP so it should be Source: 83.3 FPS (As I suspect the test in OP was compared in DX12 to favor Vega even more)

As for Doom & Civilization VI, the benchmarks you linked don't specify what API it ran so it's hard to conclude on that.

[LAwLz]

1 hour ago, Lord Nicoll said:

It's not the same chip, the Frontier Edition had some gaming centric units disabled in order to ship Vega in the 2nd Quarter.

[Citation Needed]

 

1 hour ago, Lord Nicoll said:

edit: I should note that TBR can be sort of enabled by driver updates but it won't be entirely the same, and when I say different silicon chip, I mean the stepping process and yields, the actual design is probably unchanged

Can you please explain to me what the difference between TBR being "sort of enabled by driver updates" would be, compared to what AMD has done with TBR on Vega RX?

 

1 hour ago, Lord Nicoll said:

Note the edit I made, the stepping has changed, as in yields are better and it's made better, it IS the same design, but a few tiny revisions where made, probably relating to native TBR instead of the driver workout Frontier will be using when they work it out. The native one is still better, as its got some more support components. 

[Citation Needed]

Can you please explain the difference between "native TBR" and the "driver workout Frontier will be using"?

[Misanthrope]

3 minutes ago, LAwLz said:

[Citation Needed]

 

Can you please explain to me what the difference between TBR being "sort of enabled by driver updates" would be, compared to what AMD has done with TBR on Vega RX?

 

[Citation Needed]

Can you please explain the difference between "native TBR" and the "driver workout Frontier will be using"?

I think I can explain all of these for AMD enthusiasts with just 2 words: Wishful Thinking.

[AHaskin14]

17 minutes ago, Dylanc1500 said:

Heck, I regularly saw a couple of them go for 295-300 before then.

For sure! Meanwhile I was mid upgrade when they were coming out and had already sold my old 770 so I dropped $375 on my FE as soon as I could find one in stock. But that was still when the FE's were a premium card. (I wanted a blower card and couldn't wait for the cheaper blowers to come out)

[tom_w141]

Until the 1070's price comes back down to earth... Then it will be brand preference and whether you have a free/g sync monitor. The only good thing is that HBM2 is shit for mining so vega shouldn't get hiked up.

[Drak3]

1 hour ago, Misanthrope said:

If Vega was this strong AMD wouldn't be hiding their performance with blind tests and monitor rebates instead of just showing this. 

 

They would be spamming the fuck out of the numbers everywhere like they're doing with Threadripper. 

Alternatively, because VEGA is capable of going toe to toe with Pascal in gaming, pulling ahead in productivity, and unattractive to mining, Nvidia might recognize it as a viable threat. Difference between Intel and Nvidia, there is still bounds of GPU performance to gain between generations, and they have working Volta.

If AMD cued Nvidia off, Nvidia would push up the launch of x70 and x80 Volta.

[Misanthrope]

Just now, Drak3 said:

Alternatively, because VEGA is capable of going toe to toe with Pascal in gaming, pulling ahead in productivity, and unattractive to mining, Nvidia might recognize it as a viable threat. Difference between Intel and Nvidia, there is still bounds of GPU performance to gain between generations, and they have working Volta.

If AMD cued Nvidia off, Nvidia would push up the launch of x70 and x80 Volta.

Where did those leaps and bounds went during Fiji then? 

 

Don't get me wrong I would like them to push Nvidia to 1 up them, it's just so unlikely to happen we've seen nothing officially from them and at this point is ridiculous.

[Drak3]

Just now, Misanthrope said:

Where did those leaps and bounds went during Fiji then? 

I said it can be, not necessarily that it will happen.

That's also part of the aftermath of "gpus dead, apus 5eva"

1 minute ago, Misanthrope said:

Don't get me wrong I would like them to push Nvidia to 1 up them, it's just so unlikely to happen we've seen nothing officially from them and at this point is ridiculous.

We want Nvidia to push new product, but not wipe AMD from the consumer GPU space.

[leadeater]

1 hour ago, Humbug said:

Yes it will be enabled by driver updates, cause Vega FE already has the required hardware.

So then why won't TBR be entirely the same? What difference do you expect between TBR on Vega Rx vs TBR on Vega FE

 

1 hour ago, Lord Nicoll said:

immediate mode rasterization is currently in use, it's unclear if the chips have bad TBR or if the drivers weren't ready and it can be fully implemented in time, if it isn't I'd be worried, but by no means surprised, after all, the FE was clearly a rushed launch. 

AMD said at the launch of RX Vega all currently disabled features in Vega FE will be enabled, they are the same Vega 64 chip. My guess is the released driver pack will also firmware update Vega FE enabling the features that are not currently enabled. They could also be disabled in the drivers but I'd expect AMD would do it at the firmware level to stop anyone trying to force it on etc.

 

AMD also did not directly give any performance increase expectations for Vega FE either, other than 'some improvement'. I do expect a performance improvement but most of this will likely come from the lowered power usage allowing the GPU to boost to it's proper full power state which currently almost never happens.

 

The only actual difference I have heard from AMD/tech publications (from official AMD info) is that RX Vega uses different voltage stepping and power states than Vega FE.

[XenosTech]

Well I'm just here waiting to sell my 1070 since about 6 people been nagging me to buy it since the day I got it. Just waiting for actual reviews and AIB cards before I decide to sell  this.

[Droidbot]

Vega die size is huge as fuck, Nvidia have been spending their money optimising their architectures to be smaller, cheaper to manufacture and more power efficient while AMD have been trying to catch up with half the R&D budget. 

 

This shit better be good or AMD will barely break even 

[LinusDropTechTips]

Im just waiting to see what the bigger vega gpu's can do. Unless it can beat my titan X's on water I'll be waiting to upgrade then.

[Paragon_X]

2 hours ago, zMeul said:

the the GTX1070 numbers are pulled out of AMD's ass

some peeps on reddit got the more realistic GTX1070 numbers:

• BF1: http://www.pcgamer.com/battlefield-1-performance-analysis/
  Source: 105.3 FPS
  Leaks: 72.2 FPS
• Civilization 6: https://www.techpowerup.com/reviews/Performance_Analysis/Civilization_6/4.html
  Source: 83.8 FPS with 8x MSAA
  Leaks: 72.2 FPS with 4x MSAA
• Doom 2016: https://www.techspot.com/review/1182-nvidia-geforce-gtx-1070/page5.html
  Source: 91 FPS
  Leaks: 84.6 FPS

yeah ...

With my i5 at 4.2Gh and 2400mhz on DDR3 I average around 90-95 at ultra settings DX 11 and 1080x2560 UW resolution, but what settings where used on this leaks?

[leadeater]

7 minutes ago, Droidbot said:

Vega die size is huge as fuck, Nvidia have been spending their money optimising their architectures to be smaller, cheaper to manufacture and more power efficient while AMD have been trying to catch up with half the R&D budget. 

 

This shit better be good or AMD will barely break even 

Not too much bigger than GP102, 484mm2 (est/rumor) vs 471mm2. GM200 was also 601mm2 and GK110 551mm2. If you want big though look at GV100 at 815mm2 up from GP100 610mm2, but those are compute GPUs for servers not for gaming and have full FP64 and Tensor in the case of GV100 which is where most of that size increase comes from.