cancel
Showing results for 
Search instead for 
Did you mean: 

Red Team Discussions

colesdav
MVP

Re: Time zone for Vega sales?

Here are the LuxMarK Results on my System, Dual R9 Nano FuryX and single i7-4770K.
You achieved 22505 Ball (running GPU only test). Relative perfomance difference =  just over 2x versus Just running on the Vega FE.

I will run single GPU and Dual GPU tests later and post those results in a summary table.

pastedImage_1.png

0 Kudos
red5
Challenger

Re: Time zone for Vega sales?

IMG_2446[1].JPG

First run - Ryzen 1600 (lapped) @ 3.925 all cores, Sapphire Vega 64 stock clocks, Corsair Vengeance RGB 3000 15-16-16, Gigabyte GA-AB350-Gaming 3, Corsair H110i with lapped water block,

Max wattage from the wall during run was 543 ( took the pic when it showed 536) This is just the system itself. It doesn't include the monitor.

IMG_2444[1].JPG

daquteness
Challenger

Re: Time zone for Vega sales?

Update on scores.

Used Wattman to undervolt down to 1100mv, adjusted the fan slightly, just 400rpm higher, and gave +50% power limit. The increase in results is spectacular, as the clock is now rock-steady at 1600MHz, never dips, and temps never go higher than 70-ish (but bear in mind I do live in a colder region and ambient is around 18C)

Just to recap:

-----------------  17.8.2 FE Beta6 Drivers

Luxmark 3.1 - 4886 Hotel / 16336 Mic / 22505 Ball

Blenchmark - 01m:21s - horrible...

------------------  17.6 FE Drivers

Luxmark 3.1 -  4717 Hotel / 16205 Mic /  23940 Ball

Blenchmark - 32.7s

------------------  17.6 FE Drivers, Undervolted, +50% Power

Luxmark 3.1 - 5695 Hotel / 19359 Mic / 29525 Ball

Blenchmark - 32.7s default tile size / 24.2s @512x512 tile - matching my EVGA 1080ti FTW3 render times)

Undervolting and allowing 50% more power, with stock clocks alone yields 19.5-24% better performance depending on scene, do the math if you don't believe my numbers. I'll start tickling some overclock and see what that does and also redo Compuench for you later on.

0 Kudos
whiskey-foxtrot
Forerunner

Re: Time zone for Vega sales?

These are the scores on a stock RX Vega 64 (GPU only) on current 17.8.2 drivers (Vega, not FE Beta). Holy hell those Fury's still pack a punch!

Vega-Lux.png

And just for the heck of it, My RX 480's on the same test (with -d strict aliasing)

RX-VEGA-LUX.png

kxuping
Forerunner

Re: Time zone for Vega sales?

My vega arrived to my country but its still not cleared by customs. I hope to get it next Monday or Tuesday!

daquteness
Challenger

Re: Time zone for Vega sales?

Hopefully there will be some significant stability improvement with newer drivers, I still have the old, bugged Wattman, where it shifts to lower P-state but shows full speed still, the one documented by Gamers Nexus.

colesdav
MVP

Re: Time zone for Vega sales?

Thanks ...

That RX Vega number is starting to look more encouraging  it if that is a single GPU Result. (Vega, not FE Beta).
Is that with running your cards on Air or with Waterblocks Fitted?
Hopefully that performance figure might show up in the Vega FE but we will not know until it can be proven.

RE: (Vega, not FE Beta).

I wasn't meaning to compare to the FE "Beta" driver results, although they are the latest drivers for the Frontier Edition, and they have been released...


Here is the info I have for Vega FE:

"colesdav

Did a bit more digging and decided to wipe the Beta drivers, went with standard FE drivers (17.6) and the result was... jaw-dropping at least in terms of Blender performance increase:

Luxmark 3.1 - 4886 Hotel / 16336 Mic / 22505 Ball

Blenchmark - 00:01:21

------------------  17.6 FE Drivers:

Luxmark 3.1 -  4717 Hotel / 16205 Mic /  23940 Ball

Blenchmark - 32.7s!! What..?? Goes down to 26.8 if you use 512x512 tiles by the way (1080Ti FTW3 - 24s for reference)

Here's a list of Compubench 2.0 results: ... "

I took the top line to be the Vega Beta FE results.
I took the Bottom Line to be the original Vega 17.6 FE Drivers that the card was shipped with.

For my performance comparison I was going to report just against the Original Vega FE Launch Drivers, since Blender seems to be working correctly with them.
I did that for the Hotel comparison, however for the last two LuxMark numbers (Mike, Ball) I reported against the newer  Beta Drivers ... by mistake not by design, I was rushing to get  my results posted, anbd picked up the newer Vega FE "Beta Driver" numbers.

I am a bt confused about what is a Vega FE Beta Driver, a Vega FE Gaming Driver, A Vega FE production driver ...

0 Kudos
colesdav
MVP

Re: Time zone for Vega sales?

H,

I never said I don't believe your numbers ... Seeing your Vega FE result improving is what I want to see. I want you to be able to thrash the Dual R9 Nano(FuryX) setup I am running so I can justify going out and buying one ...  . Thanks again for your help.
Increasing the Power Target almost always gives an improvement in performance for me on R9 Nano's and on other AMD Cards.

I have tested Undervolting R9 Nano for 3DMark, Gaming Benchmarks etc and using LuxMark in the past but found it was not always practical to use that undervolt for real life rendering work/jobs. Undervolt work or not depends on different rendering scenes for me. I have not reported any Compubench 2.0 or LuxMark results using undervolting because I feel it is a step too far and somewhat impractical in real life based on what I know with the R9 Nano behaviour on the rendering jobs I run. 
I could also have looked at overclocking the HBM on my R9 Nano's as well ... it can be done. I do not normaly do it for render jobs.
The +5% GPU overclock and, maxed out Power, Temp, Fan  Targets is stable for the R9 Nanos when I use them for Rendering.

Cheers.

0 Kudos
daquteness
Challenger

Re: Time zone for Vega sales?

If I can be a bit painfully honest here...unless you do heavy CAD work like I do in Revit, I reaaaaaaly would NOT go for an FE as RX64 is almost completely identical, but with a different instruction set on the FE to make it work with PRO drivers and CAD better, and 8GB more memory which is probably irrelevant anyway due to HBCC. Slap a fast PCIe NVM SSD in your PC and you're off to the races, graphics memory limitation goes out the window. Also FE has slightly higher base clocks but nothing that can't be made up for, it is essentially almost the same card...

Let me clarify what the difference is between drivers: Vega FE is made to work with both PRO and RX drivers - hence the "Frontier" name. Inside the main UI you can switch between PRO drivers for when you have heavy CAD-specific work (yet to be making a difference as shown by GN), and RX for... well... gaming and apparently it influences rendering due to wattman and the fact that most GPU renderers are in fact biased towards consumer GPUs.

Undervolting is not unpractical AT ALL as it lowers temperature and keeps clock steady, and it is very stable if done correctly - at least on Vega it is, for now, and I have the old bugged to hell drivers.

0 Kudos
colesdav
MVP

Re: Time zone for Vega sales?

RE: If I can be a bit painfully honest here...unless you do heavy CAD work like I do in Revit, I reaaaaaaly would NOT go for an FE as RX64 is almost completely identical, but with a different instruction set on the FE to make it work with PRO drivers and CAD better, and 8GB more memory which is probably irrelevant anyway due to HBCC. Slap a fast PCIe NVM SSD in your PC and you're off to the races, graphics memory limitation goes out the window.

No pain no gain .
I am interested in running Radeon Pro related software but it is not an absolute must right now.
I try to use OpenSource / Free / as low as possible cost software as much as possible to get my work done.

Performance/Price is essential for what I am doing.

Vega FE(Air) is ~ same price as RX Vega Liquid here.
Yes I understand that Vega FE 64 and RX Vega 64 are practically identical abart for 16 v 8GB HBM2 and the VBIOS and drivers.

I already have a Samsung 1T NVMe PCIe SSD which I purchased almost 1 year ago in anticipation of testing it with High Bandwidth Cache on Vega,

and for some other work I am doing.

RE: Undervolting is not unpractical AT ALL as it lowers temperature and keeps clock steady, and it is very stable if done correctly.


I was not disagreeing with lowering temps and keeping clocks steady. I have been asking people to try it out on Vega on benchmarks for ~ 1 week.
A particular undervolt and clock speed tend to work on a case by case, application by application basis though, as far as I understand it.
A number of reviewers have stated this when testing the RX Vega and Vega FE as well.
They suggest if you are going to undervolt you do so by setting it in the Profile got the Game or application you are running.
I have seen 40-80Watts power reduction for same performance in some applications reported by undervolting a little on the Vega FE.

Maybe my render working on 1 job with one scene working but failing with another scene using the same undervolt was due to another issue on the R9 Nanos.
Maybe the undervolt I used was simply too close to the edge and I might look at it again.
My point was I just want to kick off a job with settings that I know work for sure on all renders I have run in the past.
I don't want to run a job with an undervolt which might complete faster and save me power, but fail overnight because it was not stable enough or a change in the input data for rendering just kick the cards over the edge.

Thats all.

I really appreciate all of your input. I am not trying to be funny or sarcastic or cheeky in what I say next.
If you have found an Undervolt with your Vega FE that you have found to be stable and save power over a wide rage of applications and games then

please give that feedback and information to AMD because if this is the case then maybe those settings should be in the card VBIOS / Drivers for the Vega FE by default.  It would put the Vega FE cards in a much better Performance Versus Power situation, and may help the RX Vega 64 cards as well.

Maybe my last attept at undervolting my R9 Nanos was not good enough.
If I get some more time I will look again at undervolting my R9 Nanos when I run them on Windows 10 64Bit.
I might have some machines freeon Wednesday  to run some test rendering examples.

I am going to have to stop working on these performance comparisions now for the rest of today because it is another sunny day here,
Very hot, (30'C in this room already, it was only ~ 20'C yesterday.) and the temps will likely cause misleading performance numbers I would put out today anyhow.

I might try later on today once the sun goes down and the temperature drops.

Thanks again for all of your help.

Bye.

0 Kudos