• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

Poll: ** The AMD VEGA Thread **

On or off the hype train?

  • (off) Train has derailed

    Votes: 207 39.2%
  • (on) Overcrowding, standing room only

    Votes: 100 18.9%
  • (never ever got on) Chinese escalator

    Votes: 221 41.9%

  • Total voters
    528
Status
Not open for further replies.
Some interesting tidbits coming out.

Firstly the FE Vega is not drawing with Tiled rasterisation: https://www.reddit.com/r/Amd/comments/6kdwea/vega_fe_doesnt_seem_to_be_doing_tiled/

And an AMD employee told Beyond3D the driver is an older one, not bang up-to-date: https://www.reddit.com/r/Amd/comments/6k9t1o/amd_employee_vega_fe_gaming_driver_is_not_gimped/

This is rather curious, makes it look like the FE edition is running on their Fiji-enhanced drivers they ran before when showing it off. It doesn't appear to be using any of its new features, which would explain why it's running like an overclocked FuryX.

And/or this could also lead more credence to the bugged stepping theory.

Something is certainly amiss here, it's not as simple as the arch is a fail. Something is not turned on, or bugged, and/or not finished.
 
And an AMD employee told Beyond3D the driver is an older one, not bang up-to-date: https://www.reddit.com/r/Amd/comments/6k9t1o/amd_employee_vega_fe_gaming_driver_is_not_gimped/

This is rather curious, makes it look like the FE edition is running on their Fiji-enhanced drivers they ran before when showing it off. It doesn't appear to be using any of its new features, which would explain why it's running like an overclocked FuryX.

But it's now running worse in doom than before when it was shown running on Fiji drivers, I guess that could come down to them using a WC card and maybe a bit oc'd.
 
Even adding more features isn't going to magically reduce temperatures and power consumption though. Even if they squeeze more out of it, it's going to be a beast to tame.
 
Even adding more features isn't going to magically reduce temperatures and power consumption though. Even if they squeeze more out of it, it's going to be a beast to tame.

Not true actually.

Tiled based rasterisation will save power/temperature because it's part of the discard process (ignoring, not drawing, overlapping objects) and simultaneously increases memory efficiency. So literally less computation needs to be done if it's enabled.

This will either lower power consumption, or increase performance at the same power consumption. And also give an effect of overclocking memory speed.

And there are a bunch of other Vega-NCU features, though I don't know the technical details of how they'd effect performance and/or power consumption.

Worth noting TBR is where Maxwell got a lot of its MASSIVE perf/w increase from, despite not being on a smaller node than Kepler.

Also another thing to remember is AMD's official slides explicitly stated an increase in IPC with the NCU, and the current state of benchmarks puts it at a slight IPC-regression over Fiji and Polaris. There's marketing hype, and there's straight up lying in official presentations (which I believe they can be prosecuted by their investors over).

The numbers just don't add up, it seems crazy for all those extra transistors to be doing nothing or even detrimenting the processor (remember a FuryX shrunk to 14LPP would only be about 290mm2, and Vega is supposed to be between 480-500mm2). I'm sitting in the "something's wrong somewhere, and I want to see the proper RX Vega launch now" camp.
 
Some interesting tidbits coming out.

Firstly the FE Vega is not drawing with Tiled rasterisation: https://www.reddit.com/r/Amd/comments/6kdwea/vega_fe_doesnt_seem_to_be_doing_tiled/

And an AMD employee told Beyond3D the driver is an older one, not bang up-to-date: https://www.reddit.com/r/Amd/comments/6k9t1o/amd_employee_vega_fe_gaming_driver_is_not_gimped/

This is rather curious, makes it look like the FE edition is running on their Fiji-enhanced drivers they ran before when showing it off. It doesn't appear to be using any of its new features, which would explain why it's running like an overclocked FuryX.

And/or this could also lead more credence to the bugged stepping theory.

Something is certainly amiss here, it's not as simple as the arch is a fail. Something is not turned on, or bugged, and/or not finished.

Does it need games to be coded in some special way for that to work?
 
Some interesting tidbits coming out.

Firstly the FE Vega is not drawing with Tiled rasterisation: https://www.reddit.com/r/Amd/comments/6kdwea/vega_fe_doesnt_seem_to_be_doing_tiled/


Gif of the pcper test. Frontier is not doing Tiled rasterization

1vUrQ2K.gif
 
I'm betting that AMD have released this card with basically current but incomplete drivers. I may be wrong, but let's see. It would explain lots of things that do not make sense.

They needed to meet a release time frame so they did so with a teaser card that isn't one thing or another, with drivers that don't fully exploit it.

Or it could just be a dog. We will not know until Sigg rolls around.

Eat that popcorn.
 
Not true actually.

Tiled based rasterisation will save power/temperature because it's part of the discard process (ignoring, not drawing, overlapping objects) and simultaneously increases memory efficiency. So literally less computation needs to be done if it's enabled.

This will either lower power consumption, or increase performance at the same power consumption. And also give an effect of overclocking memory speed.

And there are a bunch of other Vega-NCU features, though I don't know the technical details of how they'd effect performance and/or power consumption.

Worth noting TBR is where Maxwell got a lot of its MASSIVE perf/w increase from, despite not being on a smaller node than Kepler.

Also another thing to remember is AMD's official slides explicitly stated an increase in IPC with the NCU, and the current state of benchmarks puts it at a slight IPC-regression over Fiji and Polaris. There's marketing hype, and there's straight up lying in official presentations (which I believe they can be prosecuted by their investors over).

The numbers just don't add up, it seems crazy for all those extra transistors to be doing nothing or even detrimenting the processor (remember a FuryX shrunk to 14LPP would only be about 290mm2, and Vega is supposed to be between 480-500mm2). I'm sitting in the "something's wrong somewhere, and I want to see the proper RX Vega launch now" camp.

Good points, especially the IPC statement and the fact that a 14nm Fury would be 290mm2, it sure isn't adding up.
 
So the consensus seems to be that FE drivers are not currently enabling many of the new features it needs to get the expected performance. Another poor release from AMD as they once again snatch defeat from the jaws of victory. It does however hold out hope for RX and FE in the future that performance will be there with drivers that actually use the touted new features. C'mon AMD, get your act together!
 
I'm betting that AMD have released this card with basically current but incomplete drivers. I may be wrong, but let's see. It would explain lots of things that do not make sense.

They needed to meet a release time frame so they did so with a teaser card that isn't one thing or another, with drivers that don't fully exploit it.

Or it could just be a dog. We will not know until Sigg rolls around.

Eat that popcorn.

It's pretty much confirmed with the revelation that it's not utilising TBR. There's no way a 300W TDP 14nm card with arch improvements should perform just like or worse than an overclocked Fury X which was 275W TDP at 28nm.

If TBR isn't the only feature that is currently disabled, RX Vega should be a great deal faster especially for gaming and so should the Frontier edition once those features have been unlocked.
 
Not true actually.

Tiled based rasterisation will save power/temperature because it's part of the discard process (ignoring, not drawing, overlapping objects) and simultaneously increases memory efficiency. So literally less computation needs to be done if it's enabled.

This will either lower power consumption, or increase performance at the same power consumption. And also give an effect of overclocking memory speed.

And there are a bunch of other Vega-NCU features, though I don't know the technical details of how they'd effect performance and/or power consumption.

Worth noting TBR is where Maxwell got a lot of its MASSIVE perf/w increase from, despite not being on a smaller node than Kepler.

Also another thing to remember is AMD's official slides explicitly stated an increase in IPC with the NCU, and the current state of benchmarks puts it at a slight IPC-regression over Fiji and Polaris. There's marketing hype, and there's straight up lying in official presentations (which I believe they can be prosecuted by their investors over).

The numbers just don't add up, it seems crazy for all those extra transistors to be doing nothing or even detrimenting the processor (remember a FuryX shrunk to 14LPP would only be about 290mm2, and Vega is supposed to be between 480-500mm2). I'm sitting in the "something's wrong somewhere, and I want to see the proper RX Vega launch now" camp.

Question for the knowledgeable people out there. Could it be possible that this GPU can't do Tiled based rasterization because of a bug ? Or would it only come down to drivers ? Because if it's only a driver issue, I can't imagine that something that could affect performance (this much) was not sorted out, because as we know they haven't been working on the drivers for just 1 week.
 
So the consensus seems to be that FE drivers are not currently enabling many of the new features it needs to get the expected performance. Another poor release from AMD as they once again snatch defeat from the jaws of victory. It does however hold out hope for RX and FE in the future that performance will be there with drivers that actually use the touted new features. C'mon AMD, get your act together!

I think they knew Vega would be delayed due to the drivers but had to release something since they committed to a 1H release. By releasing the FE, they did maximum damage control as they managed to release some hardware "on time" whilst ensuring that it would not be bought by the masses and the consumer perception of Vega wouldn't change that much due to it being a prosumer card. Plus anyone who would actually need the card could probably make good use out of it, as you do not need features such as TBR for things like pure compute which this card can do well.
 
Question for the knowledgeable people out there. Could it be possible that this GPU can't do Tiled based rasterization because of a bug ? Or would it only come down to drivers ? Because if it's only a driver issue, I can't imagine that something that could affect performance (this much) was not sorted out, because as we know they haven't been working on the drivers for just 1 week.

I don't think there's any chance of it being coincidental. TBR should bring a massive increase in performance, so it would be quite noticeable during testing. I'm guessing that they have it working with their development drivers, but they cannot release it atm because it's not stable enough or whatever.
 
Does it need games to be coded in some special way for that to work?

No, it's architecture and driver based. The Nvidia transition from Kepler -> Maxwell is an example.


Question for the knowledgeable people out there. Could it be possible that this GPU can't do Tiled based rasterization because of a bug ? Or would it only come down to drivers ? Because if it's only a driver issue, I can't imagine that something that could affect performance (this much) was not sorted out, because as we know they haven't been working on the drivers for just 1 week.

It could be caused by a hardware bug, or simply by drivers not being finished yet.

TBR requires hardware/architecture compatibility to work. As far as I know the 2 fundamental hardware parts needed are having a large L2 cache, plus allowing the L2 cache and ROPs to communicate directly.

Vega claims to have both of these features. And also that video of the PCper dude testing the rasterisation proves the large L2 cache part, as the sections of triangle it's drawing are physically larger than Polaris, Fiji, etc. draw. And this is because the size of triangle-section is limited by L2 cache size.
 
I think they knew Vega would be delayed due to the drivers but had to release something since they committed to a 1H release. By releasing the FE, they did maximum damage control as they managed to release some hardware "on time" whilst ensuring that it would not be bought by the masses and the consumer perception of Vega wouldn't change that much due to it being a prosumer card. Plus anyone who would actually need the card could probably make good use out of it, as you do not need features such as TBR for things like pure compute which this card can do well.
They probably also thought that a tiny number of people would buy the FE so there is much less risk if it's not certified for professional use or got proper gaming drivers on release. It was the lowest risk release they could do whilst still sticking to the H1 timetable. Are there really that many people out there that do enough professional workloads to justify the £1k cost, don't need great gaming performance, but also don't need certified drivers? Surely most amateurs dabbling with professional applications would just use "normal" consumer grade GPUs, and proper professionals would be buying the higher end stuff?
 
Status
Not open for further replies.
Back
Top Bottom