AMD Vega Frontier Edition Unboxed and Benchmark

That the hell are you talking about?!

Is this improvement?


9 months gives you 0.3% improved performance if they do not make bug for your GPU that will lock your clock speeds...

The only reason why you think you are getting better performance is because they make optimizations for games which are tested by most reviewers and because up to polaris (probably vega too), every single GCN card is basically the same with some minor tweaks.
If you take a look at nV cards, you will see that with each iteration of their SM arch, they reduces number of CUDA cores per SM core which makes iterations more different. Another reason is that nV literally have real day1 ready drivers for most of the games, AMD still needs some time to figure it out how to maximize utilization of their 5+ years old arch...

On top of that, just by not fixing UVD issue, because UVD also need tweaks in order to meet new standards, using term " fine-wine technology" is just wrong

 

This is actually not a bad score given the fact that this is a early driver, a much weaker system than the one used on the rest of the benchmarks. As well as this is not a consumer gaming product and thus is clocked lower. So maybe we will see the watercooled Vega kicking the 1080ti.

 

Ahh, the internet never ceases to amaze me. How much more obvious do I have to make it? :bang:

 

There is always the case of "With a new driver performance will fix, with a new firmware performance will fly". But let's ignore this fantasy.

AMD started working on Vega when there was only a GTX 1080 from the competition, so that was their target. Time passed, probably because HBM memory turn back on them and hit them like a boomerang, probably because of AMD's financials, probably because everyone was focusing on Ryzen, or maybe all of the above reasons and a few more. 1080 Ti came out and people's needs changed from a 1080 competitor to a 1080 ti competitor. AMD's plans can't change as fast as people's wishes. Also AMD had to consider who would address. Fans, or huge corporations like Google and Allibaba?

Vega is to Fiji what Pascal was to Maxwell, with a little Volta in it. A die shrink at a higher frequency, with much better compute capabilities in the AI and deep learning areas. People might call it a disappointment, but for me it is what I was expecting. Maybe GF's 7nm will be better to offer some kind of advantage to AMD. Maybe in a couple of years AMD will have much healthier financials to become again a true competitor to Nvidia. Until then people shouldn't demand miracles.

 

Drivers won't do any miracle. If AMD wants to release this as a gaming card they better sell it cheaper.

 

I dont think they would have said "oh ok the 1080 is out so lets compete with that since Volta and GP100 dont exist in our minds", despite people having known about them for the last four years.

Vega competes quite well with Pascal in many workloads, but it isnt up to Voltas performance, but it also costs 1/12 as much as V100.

 

Indeed, you got that one right. I only believe in things that I can see with my own eyes. I am really sorry for not being able to be fanboy to agree with you.
And reason why is there 1100 mhz for newer driver used and not for 16.7.2, with 16.7.2 GPU never works at 1100MHz, and with newer driver, you will actually degrade your general performance in case of 1st gen GCN if you forget to restart your driver every time you use UVD (Unified Video Decoder). That is a fact unless you can't can see results on graph.

I cant say that I am sorry for choosing to believe in my personal tests over tests done by reviewers/youtubers or AMD/nV/intel.

 

Anyone else notice Hilbert added this last line to the text :
"I want to make note that the guy uses an early driver, Radeon Vega Frontier Edition 17.6 with base driver version 17.20.1035 and that is not the gaming optimized driver set."

Also, if you look at the GPU-Z screenshot, it shows Crimson Edition Driver 17.1.1

Waiting for a full, proper review before I accept any results.

 

I think that was a display error showing the wrong driver version. According to WhyCry:

WhyCry:
The driver from January would not even recognize Frontier.
He used the latest driver from AMD website. I checked. And it's called 17.6 or 17.20* not 17.1.1
* "Radeon™ Vega Frontier Edition 17.6 with base driver version 17.20.1035."

 

They did however say that it was built by a different team and that it would provide higher efficiency. From the looks of it they failed, I have no desire to have 300w power monsters in any systems.

 

Ah, cheers for clearing that up.

 

Has money for Vega FE
Streams low quality with shaky cam
I try to watch to get any useful info
When I see his mining rig I'm completely triggered

Ya, miners will buy Vega. AMD has nothing to worry about :banana:

 

The thing is...

He didn't run Fire Strike Ultra, which is the "4K" benchmark. He ran the "Fire Strike" test, which is for 1080P single GPU benchmarks. The way you can tell is that his benchmarks read "Fire Strike 1.1" which is the version of that test.

The Fire Strike Ultra test will read "Fire Strike Ultra" and nothing more (see screenshot).

He bascially ran the wrong test, and actually thought is PC is 600% faster than Vega & 1080Ti's. And know one caught it, interesting....I will post links in my next post since I have fewer than five posts.

 

He ran the wrong test...

Links to the two different Fire Strike tests...

"Fire Strike" test, which is for 1080P single GPU benchmarks. The way you can tell is that his benchmarks read "Fire Strike 1.1" which is the version of that test. See below
http://www.3dmark.com/fs/12995342

The Fire Strike Ultra test will read "Fire Strike Ultra" and nothing more (see screenshot). It looks like this...
http://www.3dmark.com/fs/12995285

 

We know the difference between ultra and standard on FireStrike. He did not provide a link to a score at first so I did not want to assume he was running a different bench.

 

Sometimes people make mistake. Anyway, considering that certain compute results are within margin of error from Fiji even with boosted clocks, we can assume few things:
1) Driver still has long way to go.
2) GPU itself is very limited by memory clock or latency
3) There is internal power limiter present and active.
4) There is need for some changes in vBIOS.

 

If he bought it in the UK, he can try it out for a few days, or even more, and then send it back for a refund.
Not saying he will, but he could.

 

Simply going by this I would think something is massively wrong. 1600mhz Fury X would be closer to a 1080 ti then 1080 easy.

 

Yes, proly so.
But don't forget that THAT card up there comes with 22% more bandwidth than Vega FE.

 

Well yes. But I don't believe it would make that big of a difference.

Of course even on Fury X the bandwidth wasn't exactly as advertised.



http://techreport.com/review/28513/amd-radeon-r9-fury-x-graphics-card-reviewed/4

If they got the theoretical max out better with hbm2 Vega might have more bandwidth stock form. And of course it is still 480gb/s vs 484gb/s on a 1080 ti.

Anyway too many questions, way too few answers.