[Updated with final Review] Vega FE Benchmarking by PCPER

[Valentyn]

If we take Fiji performance as a baseline again compared to Frontier Edition.

It now matches reference 980Ti in cases.

 

That's 8.6 TFLOPs vs 6.1 TFLOPs, Fury X vs 980Ti.

 

So assuming for simplicity's sake AMD needs 2.5 extra TFLOPs to compare to NVIDIA, that means 13.1 Vega should be around 10.6 T FLOPs from NVIDIA at a worse case. 

As we're assuming NO architectural improvements over Fiji at all.

 

That puts it at around 10.6 TFLOPs ( ~NV Flops), which is between 1080 (9 TFLOPS) and 1080Ti (11.3 TFLOPS) reference.

 

Yet we're seeing it barely able to beat a GTX 1070 which is 6.5-7 TFLOPs. It's even losing compared to their old engineering sample demos, which at the time matched overclocked 1080 performance; so around what we're supposed to be seeing. 10-11 NV Flops.

 

The AMD Driver and Firmware team have a lot of work to do; that's unless there's an improved stepping also being used for Radeon RX Vega compared to Frontier Edition.

 

[cj09beira]

Just now, valdyrgramr said:

Well, I mean they did say it wold mature through drivers  Maybe that's what they meant?  

probably

[raphidy]

Guys, just wait for FineWine™. FE card is for CAD user mainly. I'm gonna wait for RX Vega to be either disappointed or surprised.

[NvidiaIntelAMDLoveTriangle]

1 hour ago, valdyrgramr said:

@Valentyn As Jason said, don't worry about gaming performance on a card not meant for gaming.  

Are you still going on with this? Once again I remind you it's the same GPU.

Sure the "gaming" version of the card is going to have drivers optimized for games, but don't think that's just going to give you 50% or more performance over what we saw. Sure AMD's drivers especially the launch ones are absolute crap, so over time they are going to improve them. But never to such a degree to make up for 50% or more.

Accept the fact that Vega, just like Ryzen isn't what AMD's marketing department made it out to be.

It's an ok product, but no where near what the competition has to offer.

Especially in the "gaming" department. 

[cj09beira]

2 minutes ago, NvidiaIntelAMDLoveTriangle said:

Are you still going on with this? Once again I remind you it's the same GPU.

Sure the "gaming" version of the card is going to have drivers optimized for games, but don't think that's just going to give you 50% or more performance over what we saw. Sure AMD's drivers especially the launch ones are absolute crap, so over time they are going to improve them. But never to such a degree to make up for 50% or more.

Accept the fact that Vega, just like Ryzen isn't what AMD's marketing department made it out to be.

It's an ok product, but no where near what the competition has to offer.

Especially in the "gaming" department. 

its perf in games is bad to the point where there must be something super wrong with it, what i think it happened  is that the gaming side of the driver isnt done at all,

people have also noted that vega isn't using its tiled based rasterizer which helps support this idea 

[cj09beira]

Just now, valdyrgramr said:

Well then all the VEGA cards will have this issue considering it's the same GPU!  /s

its all about software, its probably disabled because it might need some sort of driver level support, and as i said the gaming driver isnt ready yet

[cj09beira]

2 minutes ago, valdyrgramr said:

I changed the /s to /jk to make it a tad more obvious that I was joking.  Since I keep reading and hearing that's it's the same GPU, so that doesn't matter.

oh, i missed that :-|

[NvidiaIntelAMDLoveTriangle]

5 minutes ago, valdyrgramr said:

We've seen one card.  e.e We don't know how well the RX cards will perform.  I've already stated that could mean 1% or a lot higher...we don't know.  This card has an obvious issue  Since OCing it dropped the performance down towards 980ti level.  Also, Nvidia has had some pretty crap drivers in recent years.  I've experienced both sides for quite some time.

It doesn't matter if it's just one card, we know what the RX will perform, just like this.

Yeah, in the last couple of years Nvidia's drivers weren't anything to brag about, but even with their crappy drivers we never saw such a disappointment.

It's obvious that AMD's drivers are currently not finished, but it's not going to magically improve to such a degree. Unless they fucked them up real good.

[ARikozuM]

15 minutes ago, cj09beira said:

what i think it happened  is that the gaming side of the driver isnt done at all,

The problem is AMD's choice in cooling solutions. They couldn't keep the Fury water-cooled enough and then they stick a blower cooler on the release product for their new line-up. AMD has a history of having bad first-party cards and, unless they give the GPU to AIB's, they aren't doing Vega any service.

[ARikozuM]

Just now, valdyrgramr said:

So, you're telling me that you have this magical time machine showing us the actual performance on a product that hasn't been released, and nobody but AMD has seen the true performance of the card?  Even more so with unfinished drivers? 

This is a similar argument to those who look for overclockable chips and people point to AMD. If every CPU and GPU they make can overclock to a certain point, why wasn't it done at the factory rather than left to the consumer? There's no reason to release any product with a half-a$$ed driver. AMD and Nvidia have done this and no one seems to care much, but it's a self-defeating practice if we have to wait for more performance with updates time after time. 

[cj09beira]

2 minutes ago, ARikozuM said:

The problem is AMD's choice in cooling solutions. They couldn't keep the Fury water-cooled enough and then they stick a blower cooler on the release product for their new line-up. AMD has a history of having bad first-party cards and, unless they give the GPU to AIB's, they aren't doing Vega any service.

that helps making it worse, but its not that bad as amd already told us that aibs will be able to make their own versions of Rx vega.

if i was running the company i wouldn't make a first party cooler at least for the consumer card.

the cooler problem might what 5% of the problem probably less 

[cj09beira]

2 minutes ago, ARikozuM said:

This is a similar argument to those who look for overclockable chips and people point to AMD. If every CPU and GPU they make can overclock to a certain point, why wasn't it done at the factory rather than left to the consumer? There's no reason to release any product with a half-a$$ed driver. AMD and Nvidia have done this and no one seems to care much, but it's a self-defeating practice if we have to wait for more performance with updates time after time. 

i myself believe that having the full potential in day 1 inst too important as in most of the games you will have enough perf, but as time goes on and your card gets older, then that extra perf becomes more important as it might be the difference between 50s and 60s.

 

[Valentyn]

More frontier edition game tests, with Overclocks. Pro Mode is giving better performance overall compared to Gaming Mode at stock.

 

 

[NvidiaIntelAMDLoveTriangle]

7 minutes ago, valdyrgramr said:

I mean blue screens through six drivers is pretty disappointing, especially from Nvidia.  It was more of an expected thing from AMD.

 

Also, that's not a fact that's an assumption.  Drivers can change how a card performs compared to another, especially if it's the same company.  Others here have stated that it's mostly likely driver related too.  So, you're telling me that you have this magical time machine showing us the actual performance on a product that hasn't been released, and nobody but AMD has seen the true performance of the card?  Even more so with unfinished drivers? 

I don't need a magical time machine to tell me how the RX series of Vega is going to perform, all I need is to look at this card.

How hard is it for you to understand that it's using the same GPU?

Once again the difference between a gaming card and a workstation one, is that the second one has more features or specific features that are dedicate for that task, but ultimately both are using the same GPU, the same architecture.

The performance in games is going to be almost exactly the same. 3% at best difference, so within margin of error.

 

[cj09beira]

3 minutes ago, NvidiaIntelAMDLoveTriangle said:

I don't need a magical time machine to tell me how the RX series of Vega is going to perform, all I need is to look at this card.

How hard is it for you to understand that it's using the same GPU?

Once again the difference between a gaming card and a workstation one, is that the second one has more features or specific features that are dedicate for that task, but ultimately both are using the same GPU, the same architecture.

The performance in games is going to be almost exactly the same. 3% at best difference, so within margin of error.

 

you are underestimating what a driver can do to a card, amd is saying this is not a gaming card for a reason, the drivers aren't ready, to the point where the rasterizer isn't working as it should. 

[PCGuy_5960]

40 minutes ago, Valentyn said:

That puts it at around 10.6 TFLOPs ( ~NV Flops), which is between 1080 (9 TFLOPS) and 1080Ti (11.3 TFLOPS) reference.

Nope, a stock 1080 boosts to 1900MHz, so it has ~10TFLOPs  If you overclock the 1080, you can get 11TFLOPs

 

Using the 1060 (boosts to 1900MHz) and the 580 (at 1350MHz) as a baseline:

• Nvidia can get the same performance as AMD with 21% fewer TFLOPs
• So a Vega card at 1600MHz (13.1TFLOPs) competes with a 2000MHz 1080 

[cj09beira]

one thing i noticed is that vega is losing ground at 4k which is odd as amd cards usually gain ground at 4k, and its happening consistently.

might be related to the rasterizer maybe?

[othertomperson]

2 hours ago, Valentyn said:

The Frontier Edition outside of Tomb Raider is essentially scoring 980Ti numbers, that is just above Fury X.

 

It's losing to a reference 1080 in Ashes of the Singularity as well!!!

 

 

Old engineering sample matched 1080 in Ashes.

https://www.overclock3d.net/news/gpu_displays/alleged_amd_vega_benchmarks_appear_online/1

This idea among fanboys that Nvidia somehow does badly in Ashes of the Sinhularity needs to die. Look at the 580 and 1060, at how they perform next to each other. Identically.

[PCGuy_5960]

4 minutes ago, othertomperson said:

This idea among fanboys that Nvidia somehow does badly in Ashes of the Sinhularity needs to die. Look at the 580 and 1060, at how they perform next to each other. Identically.

The only game in which AMD cards considerably outperform their Nvidia competitors is Doom with Vulkan

[NvidiaIntelAMDLoveTriangle]

22 minutes ago, cj09beira said:

you are underestimating what a driver can do to a card, amd is saying this is not a gaming card for a reason, the drivers aren't ready, to the point where the rasterizer isn't working as it should. 

I know what a driver can do for the performance. It's obvious that they aren't ready with the drivers, but even with fully working drivers that's not going to magically give 50% or more performance in games.

The only way that would even be possible is if AMD fucked up the drivers big time, to the point in which these ones barely work.

[Valentyn]

35 minutes ago, PCGuy_5960 said:

Nope, a stock 1080 boosts to 1900MHz, so it has ~10TFLOPs  If you overclock the 1080, you can get 11TFLOPs

 

Using the 1060 (boosts to 1900MHz) and the 580 (at 1350MHz) as a baseline:

• Nvidia can get the same performance as AMD with 21% fewer TFLOPs
• So a Vega card at 1600MHz (13.1TFLOPs) competes with a 2000MHz 1080 

 

If you look at my post I do state reference, I'm comparing reference to reference here from manufacturing specs; and assuming no architectural improvements over Fiji.

Also a Vega engineering sample at 1200Mhz on Fiji Drivers with debugging layer was beating a "Reference" GTX 1080 by 10 FPS in DOOM Vulkan in January, while also running 8X TXAA.

 

Now a Frontier Edition is losing to a reference 1080 by 5% in Doom Vulkan without AA even applied.

 

Something went wrong along the way. As shown last page Vega's engineering samples were faster than Frontier Edition in Ashes, and even Sniper Elite 4 as well.

 

 

[PCGuy_5960]

Just now, Valentyn said:

Also a Vega engineering sample at 1200Mhz on Fiji Drivers with debugging layer was beating a "Reference" GTX 1080 by 10 FPS in DOOM Vulkan in January.

No it wasn't.....Vega was running Doom at 4K at a stable 60FPS and as you can see, the 1080 gets 67FPS using the same settings

Just now, Valentyn said:

 

[Valentyn]

10 minutes ago, PCGuy_5960 said:

No it wasn't.....Vega was running Doom at 4K at a stable 60FPS and as you can see, the 1080 gets 67FPS using the same settings

 

PCWorld stated it was, as did Linus back then, and other publications. As Linus confirmed, 4K full Ultra with 8X TSAA.

 

Gamers Nexus 1080 is getting 67 average with no AA.

 

Once again, and engineering sample, that's same ID was seen in other tests, at 1200Mhz Core clock, it did better than Frontier in Ashes, and Sniper Elite 4 as well.
 

 

 

 

 

[ravenshrike]

2 hours ago, Valentyn said:

So if FE has no optimization either, the fact that it's clocked 400Mhz higher should mean improved performance over the old Engineering samples from 6 months ago. Not less performance.

Except you have zero idea what workstation performance numbers were like under the Fiji drivers. It is entirely possible the new drivers significantly improved performance there and borked something heavily on the gaming side.

[JuztBe]

17 minutes ago, Valentyn said:

Something went wrong along the way. As shown last page Vega's engineering samples were faster than Frontier Edition in Ashes, and even Sniper Elite 4 as well.

 

But would AMD release under performing graphics card then without any explanations? Reviewers who ask to comment on this situation are being met with silence. IF something went wrong in drivers department I would expect AMD to say so.