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

Loks like my Fury X

 

IF Vega is using tiling, the performance is even more embarrassing...

 

Because they may be hoping some of the performance is fixed with a newer driver. Reviews using that driver will present a somewhat improved snapshot of the card.. And initial reviews are very important, as they would still be read six or more months after release, when making buying decisions.

The tests done now can be much more easily dissmissed when the real reviews are up, showing say 10% better overall FPS. (Of course, that 10% could end up being 1%, but they're taking this gamble). So this can make economical sense.
Amd may further seem to thing that target audience of the FE are satisfied withn a few powerpoint sllides and 2-3 graphs to make a buy decision. That may be cause the volume is so low. However I could see how game developers won't care about performance; they just know this is a new architecture that they need to target and whatever AMD launches in the next 3 years will surely be similar. For the 2-3 people using professional cards from AMD , this may also be a no brainer upgrade.

This is just a delayed launch without technically being one.

 

I don't think any of those statements conflict with a (near?) future where major performance improvements are possible.

I've used these same graphs when I wanted to point out that there's no reason for a Quadro to be worse than non-Quadro for gaming workloads.

But let's be honest: there are not many reviews that test exactly that, but of the ones that you can find, older ones typically show the opposite, with Quadro underperforming.

Whether or not it will apply to Vega between FE and RX is in the hands of AMD. They could still be sitting on improvements that didn't make the FE deadline in time. Or they could even deliberately have delayed the release of those improvements (it's AMD after all. And, no, that doesn't conflict with Rys' statement that FE is gimped on purpose. There's a difference between not releasing something and releasing something that's deliberately worse that it should be.)

 

Vega distribution across europe seems awkward at best. Apparently, an unknown number has been sold by scan.co.uk, one is in Helsinki but for our domestic market here in germany for example, every single E-Tailer we've contacted, has not even a definite date when their ordered shipments will arrive. Weird.

Regarding the drivers: Over the years I've lost faith in an alround-magical driver. Sure, when there's something broken and a new driver fixes this, performance might jump drastically. Experience, though, tells me, that this increasingly unrealistic to happen the broader your sample size of tested applications get. IOW: Yes, i some cases, the performance is likely to be up in the two-digit percent, but overall, I find this to be unlikely.

 

It appears to be not active, currently. The pattern in that tool is identical to Fiji. It appears to process triangles sequentially, no binning can be observed. So either that tool can't catch it properly (but it can in case of nV's GPUs) or AMD still needs to switch it on in a later driver.

 

Yeah it doesn't appear to be working / "turned on".

 

No matter how the final Vega GPUs end up this launch stinks. It's almost like the Vega FE is an afterthought for AMD. It launched what, 3 days ago? And all we got is two random people getting one and posting some benchmarks and the PCPer livestream. To top it all off AMD seems to be on a complete radio silence. Some form communication never hurts, the truth is usually better than complete silence.

"X feature doesn't work yet, we are working on it" is better than "..." even if it's stating the obvious.

 

Or this:
http://techreport.com/review/31224/the-curtain-comes-up-on-amd-vega-architecture/3

 

What should we expect of the localized tiled deferred mode for Vega?
It should try to defer for as long as it has binning capacity, perform hidden surface removal, and then raster.
However, if that did happen, how could the binning test work? It's a bunch of back to front triangles, it should just cull out all but the last triangle all the time.

However, the test works with a counter increment to the structured buffer. Each pixel is supposed to update a common counter, which might be a side effect or dependency between triangles. Even if the new rasterizer was binning, it or the driver might serialize knowing that it would be losing counter updates.

Some of the claims for a hybrid rasterizer indicate there are conditions where it falls back, including dependences or certain side effects like depth modification.

 

Has anyone tested Vega FE in applications it is meant for? Such as Maya, 3ds Max, Solidworks... How it works with Radeon Pro Renderer? How good is it for VR?

 

Here is the Link to the origanl article from David Kanter:
http://www.realworldtech.com/tile-based-rasterization-nvidia-gpus/
Also there is a Link to Github where you can download the Trianglebin

 

Did any pro site do a pro software review on Vega FE launch date? My quick googling didn't find any.

 

To my knowledge, there has been no comprehensive review of Vega FE yet. PCPer being in the process of whipping up one as the exception.

 

No reviews yet. Just random benchmarks.

 

It doesn't, but it's quite clear AMD is keen on hiding something. In PCWorld's preview they compared Vega to TitanXP in 3 games, and claimed there was no discernible difference in performance, obviously this was a lie. Couple that with their overall radio silence, and it doesn't sound good.

 

Is it? Did PC World state whether or not VSync was off? I mean, I couldn't tell 60 fps from 60fps in a single-GPU setup most of the time.

 

But this make no sens. The draw-stream binning rasterizer have always an Advantage above a common rasteriezer. Why avoid them?

 

The driver is actually from January, it was just packaged on 6/27/2017. This is why all the software tools are saying "17.1.1"

 

Always is a strong word.

„The driver“ being which components specifically?
The file quakechamps.ico is dated 4th of june, the VulkanRT installer was signed on april 6th. Some of the OpenCL files were signed on june 25th. So, I think "the driver is from january" is a bold statement.

The OpenGL driver for example is version 6.14.10.13486, the one from the June Catalyst Package 17.6.2 is 6.14.10.13476, the DirectX Universal driver (atiumdag.dll) has version 9.14.10.01280, the 17.6.2-version is 9.14.10.01261. Just for reference.