AMD Vega 10, Vega 11, Vega 12 and Vega 20 Rumors and Discussion

It's what AMDMatt said.

 

It literally has disabled uArch feature in need of surrounding software.

 

Oh? Any reference I missed?

 

Forza 7 last I checked. That's with the latest Nvidia driver. It was out in front, but the driver release I'd consider to have made it competitive. Changing places a bit based on resolution, CPU, and average or min FPS. It's not clear any of the new uArch features have landed or to what degree.

The implementation should be different with all the added caches to facilitate the higher clocks. HBCC potentially in play there as well. The practical difference however is unclear. Seems likely it evolved a bit as they iterated on the design, but Polaris may have received those changes as well.

 

With the latest drivers, Vega is competitive with the original 1080, depending on the resolution and settings.

 

At least HBCC helps in superposition(tested it myself), >10% with 1080p extreme settings but not for higher resolutions.



Might help in some games as well,



I can run the Vega56 at 1.45Ghz consistently now with a few watts over the stock, still a bit shy of the 1471Mhz boost clock and quite a way off from the 1.6Ghz boost on p7 in wattman. With 880Mhz of HBM2 it matches the 1080FE in benchmarks.

So if AMD can improve the clocks on these chips and bring down the temperatures on the hotspot with the 14nm+ it should do well enough for a refresh. Hopefully they don't go the 480->580 route.

 

Here https://forums.overclockers.co.uk/t...owners-thread.18789713/page-116#post-31197875

And I can't find it now, but I remember an other amd guy saying the same thing on twitter.

 

And the 1080ti all at the same time! Amazing.

 

If it's in ONE game it's not amazing at all, TBH. More like oddball fluke actually.

 

It was still a good come back line! ROFL!

 

Oops. Wrong thread

 

The GTX 1080 is the same speed as Vega 64 @1440p and @2160p in Forza 7, GTX 1080 Ti is at least 20% faster than this, So NO Vega 64 is not competitive with 1080Ti now, Only the 1080.

 

That would be amazing; where are you seeing those results?

 

simply because the 1080Ti results have not been updated with the last drivers

 

AMD would require lots of improvement in the card's base performance to be competitive with 1080Ti even if it were running at its advertised boost clocks. As it is now it's barely matching Fiji in clock-for-clock.

nvidia have more granularity with their boost clocks and a nice curve to handle it. With Vega you have 7 p-states and you can't change the lower p1-p5-states, and since they've higher voltage than what you'd set for the p6-p7 states, the card gets stuck there with lower clocks and power ceiling if you exceed its operating parameters(temp, voltage, power). The workaround currently being used is to max out the power slider so that it doesn't go to lower p-states but it'd be great if the other states were unlocked.

New afterburner release has the option for voltage change that affects all p-states for a max of 200mV, unfortunately it applies to all of them at the same time and so is unstable at lower p-states if you pull down the the ludicrous 1.2V of the p7 to something more sensible like 950-1050mV.

 

Even with the drivers it would be competitive. Vega was 22% ahead of 1080ti and drivers added about that much for the 1080. That's not considering minimums either, where Vega was doing rather well.

Not necessarily, as async compute is designed to work around bottlenecks. If actual shading, or any limit, is less than half the overall work and not serially dependent without overlapping frames it will be fine. Compute tasks aren't using ROPs, TMUs, or primitive setup. Games are trending that direction. So any title favoring say 80%+ async compute, should scale by FLOPs and bandwidth.

 

Yes, it is illegal for compute tasks to use atomics or work on images previously output by the rasterizer.

 

Is this a joke? Once again the 1080 = Vega 64 averages and minimums, 1080Ti is higher than both.

 

You're right about ROP's and primitive setup but TMU's you're wrong. You can sample a texture using samplelevel from a compute shader.

 

Raven Ridge shows up for the first time in an official laptop data sheet.

https://videocardz.com/newz/hp-envy-x360-spotted-with-ryzen-5-2500u-and-vega-graphics



Anyone else a little depressed that we're just now seeing this kind of thing?

Raven Ridge's timing feels awkward:

• It's too early for 12nm (Q1'18's Pinnacle Ridge gets that).
• It's too late for 2017 "back to school" season.
• It's probably too late for 2017 holiday season.

I know that Q1 (i.e. CES) is the classic time to release new laptops, but it seems so strange that Summit Ridge arrived almost a year ago (not to mention its successor is approaching) while Raven Ridge is just now getting out of the door.

I can't help but suspect that Vega could be the culprit since it's the only substantial difference between Summit Ridge and Raven Ridge.