cancel
Showing results for 
Search instead for 
Did you mean: 

Red Team Discussions

colesdav
MVP

Re: Time zone for Vega sales?

Hi,

Those games are good but they are not costing 100 here. Prey for example is currenty about 20 on sale and Wolenstien 2 is ~ 40 (it will likely be on sale soon).

So the games stuff is better than nothing but still over cost the cards should be. 


I would not go for an Air Cooled Vega 64. I would only go for Watercooled because I think these cards are currently pulling way too much power to rely on a blower aircooler. I run compute heavy workloads and they can make my cards pretty hot, not just games.  GTX1080/GTX1080Ti are cheaper/same price as Vega 64 Liquid here.

I do not think a 2 slot high blower aircooler is enough, so ...next choice =  AIB cards anyhow, so look at the cost of those. even higher, unlikely below 2 slots high.

The voltage regulators on all of the Vega 64 cards are of a very high quality, so I do not think that an AIB card will in reality be much better.

I may as well go for Liquid Cooled RX Vega or Vega FE. The Watercoolers on the AMD cards are impressive -  there are detailed teardowns on the Web.

I  am running the OpenCL and Luxmark tests  to get an idea of ... how does R9 Nano (Fury X) compare to Vega 64 Compare to GTX1080 compare to GTX 1080Ti.
That's what is important to me.

My R9 Nano FuryX  are actually scoring higher than a GTX1080 according to Compubench in some cases. I will not comment on where the RX Vega or Vega FE sits right now because I have not looked into what Compubench says in much detail about them yet.

I intended to keep the R9 Nano(Fury X's) , I am also running compute tasks on R9 280x. I was running compute tasks on  HD7970 OC 6GB card until it went into a "Zombie state" a while back.

I bought the R9 Nanos at very low prices before the Mining Craze kicked off. 1 new in sale, 1 on Ebay. 
There is no way I will take the risk of buying a GPU on Ebay after this mining craze.
The cards could have been running for so long they could be near failing.

Cheers.

0 Kudos
ajlueke
Grandmaster

Re: Time zone for Vega sales?

Unfortunately, I do not have those applications myself to test.  I currently have a Vega 64 running on the stock cooler, and I can say that it does get hot.  If I set the GPU to turbo mode and run the Time Spy stress test, the GPU will quickly hit over 70C.  xobeloot​ has a Vega 64 in a custom water cooled loop and is observing temps below 50C.   So I think for your application, pushing the GPU at a high level for an extended period of time, you would want to go with liquid cooling for sure.

I have noticed that the Vega 64 air reference air cooler isn't as stable either.  When the GPU heats up, it drops the boost clock to compensate.  As a result, I have scores in Time Spy stress test of 97.6%, where 97% and below is failing.  My R9 Fury X produced stress test results at 99.7%.  I imagine the liquid cooled version can maintain the boost clocks continuously.  Less shifting up and down will likely lead to a much more stable experience.

daquteness
Challenger

Re: Time zone for Vega sales?

I have a slight feeling you and I are using the card for the same reason (i.e. GPU rendering, would be curious to know if anything else though!). The sad part is that because of AMD's high computational power reputation (or lack thereof depending on point of view), most rendering applications are slightly biased towards CUDA instead of OpenCL.

Blender, for instance, is finally bringing them a lot closer together in terms of performance for Cycles and by following their tweets and progress, I'd say they're really warming up to AMD with Vega so we might see a lot more optimisations to come. Also, Radeon Pro Render is NOT to be ignored, that thing packs a serious punch, it's super flexible, software-host-agnostic if plugged-in correctly, and it's completely free, just like Cycles, but a bit easier to use. Would love to see this beast come to Revit, it would literally be what the absolute ideal case of what the FE is made for! (Pro-performance in viewport + Gaming-level GPU speeds!)

daquteness
Challenger

Re: Time zone for Vega sales?

Right, well, AMD better come up with good driver soon or this puppy will not be staying with me for too long... so far the FE has been about as stable as a fat guy on a fence...

Everything out of the box + 17.8.2 Beta driver

This is the info dump from the driver:

Graphics Card Manufacturer - Designed and built by AMD

Graphics Chipset - Radeon Vega Frontier Edition

Device ID - 6863

Vendor ID - 1002

SubSystem ID - 6B76

SubSystem Vendor ID - 1002

Revision ID - 00

Bus Type - PCI Express 3.0

Current Bus Settings - PCI Express 3.0 x16

BIOS Version - 016.001.000.004

BIOS Part Number - 113-D0501100-109

BIOS Date - 2017/06/16 00:13

High Bandwidth Cache Size - 16368 MB

High Bandwidth Cache Type - HBM2

High Bandwidth Cache Clock - 945 MHz

Core Clock - 1600 MHz

High Bandwidth Cache Bandwidth - 483 GByte/s

Memory Bit Rate - 1.89 Gbps

2D Driver File Path - /REGISTRY/MACHINE/SYSTEM/ControlSet001/Control/Class/{4d36e968-e325-11ce-bfc1-08002be10318}/0000

As for scores:

Luxmark 3.1 - 4886 Hotel / 16336 Mic / 22505 Ball

Compubench... DNF!! Yes, exactly, did not finish, the driver crashes right after the second test.

I'll try installing the RX drivers to see if there's any improvement but so far it's a bit disappointing for the price...

0 Kudos
colesdav
MVP

Re: Time zone for Vega sales?

Hi, I just saw your post. Sorry for the delay.

For Compubench, all tests completed for me apart from the Local Tone Mapping 2K and 4K tests.
I ran on the R9 Nano (Fury X's)  at stock CPU/GPU and overclocked CPU/GPU.

Those Local Tone Mapping  tests start to load, but then Windows reports that there is a problem with the application and then I close the tests.

Nothing Crashes though. I will try to investigate with Compubench. There may be a known reason why those two tests do not run with later generation AMD GPU's.

First thing I think you might try is run each test individually. See what works and what does not. You can do that on the Compubench menu.

I have also run the tests on another i7=4770K + Nvidia 780Ti system just as another reference point, at stock CPU/GPU clocks. 
Some tests will not run on the Nvidia card but it is because some features are not supported. No crashing though.

I noted that you mentioned your CPU is overclocked and I do not know If you are running your Vega FE at Stock Clocks or not.
In any case. when you apply an overclock, even if you have run extensive stress testing, it is no guarantee that your system will run every application without a crash. It might be an idea to run your system at stock speeds if you still get crashing.

It might be a bit early to blame the Vega FE or it's new drivers just yet as it might be because your system is running with an overclock that causes an issue. I would have thought the Vega FE Drivers should be stable though if the card is released and targeted towards professional applications. Maybe you should try with the original Vega FE Driver first? .

I will post my Scores as soon as I can. I have to take a break from my PC's for a  while.

Thanks again.

daquteness
Challenger

Re: Time zone for Vega sales?

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:

Level-set Simulation 128                - 11316.5 MVoxels/s

Level-set Simulation 256                - 11776.8 MVoxels/s

Local Tone Mapping 2K                  - Skipped

Local Tone Mapping 4K                  - Skipped

Ocean Surface Simulation             - 4142.0 Iterations/s

Catmull-Clark Subdivision Level 3  - 202.8 Mtriangles/s

Catmull-Clark Subdivision Level 5  - 261.6 Mtriangles/s

N-Body simulation 128K                 - 438.8 Iterations/s

N-Body simulation 1024K               - 77.3 Iterations/s

Vertex Connection                           - 11.2 MPixels/s

Subsurface Scattering                    - 7559.1 Msample/s

Subsurface Scattering MultiViews  - 6995.6 Msample/s

TV-L1 Optical Flow                          - 60.5 MPixels/s

Apparently, from what I'm reading on RadeonPro twitter, there's a driver coming up soon that brings massive bonus for Vega FE, you can install a Pro driver and 2x Radeon Drivers (I'm guessing 1 standard and 1 beta to make sense). I cannot begin to describe how anxious I am for that to come out...

One thing to bear in mind: these are all results with stock cooler and stock speeds, on essentially immature drivers, I didn't adjust a single thing. I figure this card might actually be worth watercooling. Good... been looking for an excuse for that

And one thing I've noticed... coil whine from hell... That's through the case (Corsair 600C if you're curious)

Let me know what other benches you want. whiskey-foxtrot​ these are the results I had mentioned, I hope they're useful.

Cheers!

V.

colesdav
MVP

Re: Time zone for Vega sales?

Hello,

Looking into the Compubench directory and log files further I found the following issue:

Failed
OpenCL error: [CL_BUILD_PROGRAM_FAILURE]
file: c:\jenkins\workspace\cb_BuildDesktopInstaller\compubench-source\compubench\cb\cl\Program.cpp, line: 97
So it looks like there is an error in the Compubench code for these benchmark tests.

I contacted Compubench Support with a detailed description of what we are investigating and why, along with details of the error and crashes seen.
I have asked for an explanation of why the crash occurs, and an estimate of fix date if it will be fixed.
I have suggested if Tone Mapping 2K and 4K are  no longer relevant to their benchmark suite  that they either:

(1). Remove it completely.
(2). At the very least, disable it by default.
(3). Investigate the crash, and at the very least have the test exit cleanly with an understandable (human readable) error message.


If I get a response from Compubench I will let you know what they say.

I also found this information showing Compubench test results on Vega card, it seems N Body Simulation tests did not run for the reviewer.
I have not had a problem running those tests on R9 Nanos or older cards.

'Radeon RX Vega' Release Date, News & Update: Specifications of Upcoming AMD GPU Spotted In Recent B...

For anyone out there who is not familiar with what Tone Mapping is you might find the following article useful:
What is Tone Mapping?


Bye.

0 Kudos
colesdav
MVP

Re: Time zone for Vega sales?

Here are my Benchmark results running a single R9 Nano (FuryX) on Compubench. 
R9 Nano run a default Wattman Settings. No GPU or HBM Overclock used, no Temp target change, no Fan Target change no Power Target change or Voltage changes. 
i7-4790K is running with +9% OC. In MHz thats ... 46, 46, 45, 45 MHz for 1,2,3,4 cores active.
Surprisingly the R9 Nano is beating the RX Vega in some tests versus those reported in the above article... and is not that far behind in others.
The article results may have problems of course and I do not have many details.
If the article results are representative of the Vega FE today then it looks like I made a good choice buying the R9 Nanos after all.
I purchased 2 R9 Nanos for less than 600, whereas a Vega FE is ~ 1000 MSRP here.
It is looking like my 2 R9 Nano's together are giving almost 2x the OpenCL compute performance for 1/2 the cost of a Vega FE at present.
They are OK for gaming, and I can run them together in Crossfire(DX11) or MultiGPU (DX12) and for the most part that is working well.
I cannot however run Radeon Pro software.
I look forward to your Compubench results.
I certainly hope you thrash my results.

Advanced Compute
Level Set Segmentation – 12810719.7 MVoxels/s
Level Set Segmentation – 25611816.3 MVoxels/s
Local Tone Mapping 2KFailed. - Contacted Compubench.
Local Tone Mapping 4KFailed. - Contacted Compubench.
Game Effects
Ocean Surface Simulation3485.5 Iterations/s
Catmull-Clark Subdivision Level 3 147.4 MTriangles/s
Catmull-Clark Subdivision Level 5230.7 MTriangles/s
N-Body simulation – 128k409.5 Iterations/S
N-Body simulation – 1024k85.2 Iterations/S
High Quality Computer Generated Imagery and Rendering
Vertex Connection and Merging 10.4 MPixels/s
Subsurface Scattering4823.57 MSample/s
Subsurface Scattering multiple view5160.5 MSample/s
Computer Vision
TV-L1 Optical Flow49.7 MPixels/s

Note ... edit made, I had typed in the wrong number last time around.

0 Kudos
colesdav
MVP

Re: Time zone for Vega sales?

Here are results for single R9 Nano(FuryX) with 5% GCLK OC, Power, Temp, Fan Targets set to maximum. No HBM Clock OC. No Voltages Changed.

  

Advanced Compute
Level Set Segmentation – 12810987.2 MVoxels/s
Level Set Segmentation – 25612354.9 MVoxels/s
Local Tone Mapping 2KFailed. - Contacted Compubench.
Local Tone Mapping 4KFailed. - Contacted Compubench.
Game Effects
Ocean Surface Simulation 3536.8 Iterations/s
Catmull-Clark Subdivision Level 3 154.3 MTriangles/s
Catmull-Clark Subdivision Level 5 239.0 MTriangles/s
N-Body simulation – 128k422.6 Iterations/S
N-Body simulation – 1024k87.2 Iterations/S
High Quality Computer Generated Imagery and Rendering
Vertex Connection and Merging 10.7 MPixels/s
Subsurface Scattering 5549.7 MSample/s
Subsurface Scattering multiple view 5213.1 MSample/s
Computer Vision
TV-L1 Optical Flow49.9 MPixels/s
0 Kudos
colesdav
MVP

Re: Time zone for Vega sales?

Subsurface Scattering 5549.7 MSample/s above should be 4823.57 MSample/s - will correct when I can edit the post again...

0 Kudos