AMD's Vega architecture previewed at ve.ga

[roylapoutre]

On 1/5/2017 at 11:38 PM, Citadelen said:

The Vega chip shown was Vega 10 to my knowledge, which was faster than a 1080 in that test. The name scheme goes in the order of which was developed first. Polaris 10 was developed first, then Polaris 11, and more recently Polaris 12. Vega 10 in this instance is supposed to be the smaller chip with 4,096 shaders, while Vega 11 is the bigger chip rumored to contain 6,144 shaders. 

What excites me is the sheer number of shaders built on a new architecture built for high clock speeds and an even higher IPC than GCN already has.

I totally agree with you but the problem is that they stated that the chip running Battlefront and Doom was their top of the line Vega GPU, la crème de la crème. At least that's what I understood, did I miss something here ? Because my whole argument was based on the fact that these demos were shown running on the most powerful Vega consumer chip! And IF that's the case, it's underwhelming af and kinda weird with all the hype in trailers plus de "poor Volta" tackle at Nvidia.

[Citadelen]

2 hours ago, roylapoutre said:

-snip-

I don't believe it was ever said which chip was being run. Even if it wasn't, an important bit of information to note is that all the demos were run using Fiji drivers, which means quite a few of the benefits of the massively upgraded architecture simply weren't being used, and the card was brute forcing its way through those games. It would generally be quite safe to expect gains of up to 20% when these features are optimised for and enabled.

[captain cactus]

So Linus did a video on Vega: 

 

Confirming the Doom demo is indeed running at max settings with TSSAA (for async compute presumably unless id fixed async compute not working unless TSSAA is on or no AA at all) at around a 65-70 fps average. For a engineering sample Vega card running Fiji drivers going above GTX1080 performance (which does around 60 fps) that pretty ballin'. But then again, this is Doom on Vulkan on AMD hardware. We know AMD hardware has an intimate relationship with Doom on Vulkan, so we'll have to wait for some more months to see what other games perform like. But so far, not bad.

[xAcid9]

30 minutes ago, lots of unexplainable lag said:

So Linus did a video on Vega: 

 

Confirming the Doom demo is indeed running at max settings with TSSAA (for async compute presumably unless id fixed async compute not working unless TSSAA is on or no AA at all) at around a 65-70 fps average. For a engineering sample Vega card running Fiji drivers going above GTX1080 performance (which does around 60 fps) that pretty ballin'. But then again, this is Doom on Vulkan on AMD hardware. We know AMD hardware has an intimate relationship with Doom on Vulkan, so we'll have to wait for some more months to see what other games perform like. But so far, not bad.

That is pretty concerning actually since from this benchmark shown that the Fury-X only 7 FPS behind GTX 1080 in Doom 4K Ultra while in majority of other games the Fury-X got destroyed.

 

 

 

 

But it is an engineering sample and with super early driver so many things could change.

[Vode]

From my testin on Doom 4k on Ultra it looks like that mysterious Vega card (~70FPS) is about ~30% faster than my OC R9 Fury Nitro (~50FPS).
(Very rough estimate)

If that's on an older, unoptimized driver... oh boi pretty impressive IMO.

 

Fury X was only doing 50FPS in testing during Summer (Digital Foundry Doom Fury X video). Now we're at 55 on a Fury non X.

 

A lot of potential here. Hope it lives up to it or even surpasses expectations.

[Vode]

1 hour ago, xAcid9 said:

That is pretty concerning actually since from this benchmark shown that the Fury-X only 7 FPS behind GTX 1080 in Doom 4K Ultra while in majority of other games the Fury-X got destroyed.

 

 

 

 

But it is an engineering sample and with super early driver so many things could change.

How old are those benchmarks? Looks like they are a coupple months old.

[xAcid9]

15 minutes ago, Vode said:

How old are those benchmarks? Looks like they are a coupple months old.

From late July last year. 

 

http://www.pcgamer.com/doom-benchmarks-return-vulkan-vs-opengl/2/

[Vode]

4 hours ago, xAcid9 said:

From late July last year. 

 

http://www.pcgamer.com/doom-benchmarks-return-vulkan-vs-opengl/2/

You can add 5 FPS+ to the R9 Fury cards then.

 

I was stumped when my R9 Fury seemed to perform better than the Fury X in Digital Foundry's testing from July (which were the same results as in the chart you posted.)

 

Still no working V-Sync though. Haha

[leadeater]

I would be rather weary of using Fury X performance figures in DX12/Vulkan games versus DX11 games and then using that information to draw conclusions over the shown Vega performance in Doom. Vega is not GCN, it does have GCN DNA but there is such significant architecture improvements the DX11 vs DX12 difference may not be as great or even a thing with Vega.

 

It likely is still a thing though since AMD realistically wouldn't have spent a proportionally greater development effort on DX11, as far as they are concerned DX11 is legacy and rightly so. As long as you get good performance and stability in DX11 that is enough, new games will be DX12/Vulkan focus on that.

[ivan134]

7 hours ago, xAcid9 said:

That is pretty concerning actually since from this benchmark shown that the Fury-X only 7 FPS behind GTX 1080 in Doom 4K Ultra while in majority of other games the Fury-X got destroyed.

 

 

 

 

But it is an engineering sample and with super early driver so many things could change.

Not to mention reference cooler with all the holes in the case taped shut. That shit is pretty impressive considering all that.

[Kenoxabe]

All these debate / discussions about Vega being sh*tty / better than Pascal...I really could care less.

 

I just want it to perform better than 1070 or even on par with 1070 or even the 1080 (and in due time, better.. Here's hoping) with a cheaper price tag.

 

DON'T HATE ON ME. 

 

I'm not a fan of either NVIDIA or AMD... I just wanna see some competition. Lol. 

[bcredeur97]

I'll just buy whatever works for me... recommend everyone do the same and have no bias

[Wolther]

23 minutes ago, bcredeur97 said:

I'll just buy whatever works for me... recommend everyone do the same and have no bias

Should be a given. Only care about performance, price and durability; I couldn't care what brand it was. 

[Valentyn]

Taking a gander inside the VEGA demo unit. Looks like the card might have a max TDP of 225W at the moment, using a single 8-pin.

Not bad for a very early engineering sample I think. Hopefully they can further improve performance and power usage in the next few months coming up to launch.

Fingers crossed for not another RX 480 PCIe power draw issue eh?

[leadeater]

38 minutes ago, Valentyn said:

Taking a gander inside the VEGA demo unit. Looks like the card might have a max TDP of 225W at the moment, using a single 8-pin.

Not bad for a very early engineering sample I think. Hopefully they can further improve performance and power usage in the next few months coming up to launch.

Fingers crossed for not another RX 480 PCIe power draw issue eh?

Well that power cable could have 2 PCIe 8 pin power connectors on it, all mine do. But I agree it's likely to be a single, we basically already know this from the recent announcements of the professional/server Vega GPUs.

[LoE Ferret]

I would bet it is 1 8-pin connector. Even if it is a 2x 8 pin I don't see the complaints as long as the performance can match the draw requirements (even if it is not a huge for myself).

 

Of course everyone will test the PCIE draw at release. You would scary not too even the past, but also AMD would be crazy to do that again, so I highly doubt it happen.

 

More details on release date would be nice.

 

[cj09beira]

On 09/01/2017 at 0:37 PM, Valentyn said:

Taking a gander inside the VEGA demo unit. Looks like the card might have a max TDP of 225W at the moment, using a single 8-pin.

Not bad for a very early engineering sample I think. Hopefully they can further improve performance and power usage in the next few months coming up to launch.

Fingers crossed for not another RX 480 PCIe power draw issue eh?

Amd be like, i knew we shouldn't let linus open that panel. 

[Carclis]

On 1/9/2017 at 11:45 PM, leadeater said:

Well that power cable could have 2 PCIe 8 pin power connectors on it, all mine do. But I agree it's likely to be a single, we basically already know this from the recent announcements of the professional/server Vega GPUs.

It's a Thermaltake Smart 1200 which doesn't have any 8+8 pin connectors. They're all individual 8 pins, which still doesn't tell us much since an 8 pin is probably quite sufficient still for up to 250w.

[leadeater]

2 minutes ago, Carclis said:

It's a Thermaltake Smart 1200

That seems like an odd choice of a PSU, wattage wise, if you are trying to portray Ryzen and Vega as power efficient lol.

[Carclis]

16 minutes ago, leadeater said:

That seems like an odd choice of a PSU, wattage wise, if you are trying to portray Ryzen and Vega as power efficient lol.

I think it was just chosen so that nothing would be revealed about the power consumption of the card. It's not like power consumption is super important at that end of the market anyways and they did show Vega working which I think is all they intended to do.

[Notional]

18 minutes ago, leadeater said:

That seems like an odd choice of a PSU, wattage wise, if you are trying to portray Ryzen and Vega as power efficient lol.

Probably to get very little ripple. Most PSU's run the most efficient at about 50%, which also means it will be quite silent at that draw.

[leadeater]

1 minute ago, Notional said:

Probably to get very little ripple. Most PSU's run the most efficient at about 50%, which also means it will be quite silent at that draw.

Not likely to be using 600W thought, just find it amusing how big the PSU is.

[Humbug]

1 hour ago, Carclis said:

I think it was just chosen so that nothing would be revealed about the power consumption of the card. It's not like power consumption is super important at that end of the market anyways and they did show Vega working which I think is all they intended to do.

also that card was very much part of AMD's engineering process... You can tell from all the additional diagnostic hardware and USB port at the back. The final cards specs and speeds are not locked down yet.  These things are designed for engineers to test and tweak under various conditions including including various voltages and clockspeeds. It may have been outfitted with overkill power inputs just so that the engineers are not hamstrung when they need to torture it or overvolt it.

[Humbug]

1 hour ago, Notional said:

Probably to get very little ripple. Most PSU's run the most efficient at about 50%, which also means it will be quite silent at that draw.

I bet we are over thinking this and the guy at AMD who put together the test system just used that PSU because he had it lying around.

[cj09beira]

5 minutes ago, Humbug said:

I bet we are over thinking this and the guy at AMD who put together the test system just used that PSU because he had it lying around.

you are probably right