Show Posts
I might as well report on my findings here for the Radeon™ Pro WX 5100 card since I seriously doubt anyone else will. It is a 4th gen GCN polaris card with 1792 cores and 8GB memory with a 75 watt TDP (no external power). Essentially an underclocked RX 565 if there ever was one.
Stock clocks were fairly horrid, as seen in the image. The only redeeming factor being Samsung memory. I had to heavily OC the memory to get any decent result. Anything past 1900 mhz resulted in decreased performance and it crashed at 1990 mhz.
When I attempted to dual mine (Eth + SIA) after getting the clocks in the sweet spot the card just wasn't having it.
Best I could do was 16.4 mhs Eth and 460 mhs for Sia, it was forcing the card to draw too much power and the clocks were throttling hard down to like 750-850 mhz. It was also at 89 C dual mining.
Changed over to Eth only and got a much better 21 mhs with a stable 79 C and clocks around 1173 mhz. Power was hovering around 75-82 watts , the other card is an RX560 with a maxed out memory OC.
It is absolutely not a card built for sustained heavy load but it does manage and gets the job done decently. For my $250 I dont think I did bad, will mine with it and resell for more at a later time.
If you somehow get these for super cheap(less than $200) then its an instant buy
Stock clocks were fairly horrid, as seen in the image. The only redeeming factor being Samsung memory. I had to heavily OC the memory to get any decent result. Anything past 1900 mhz resulted in decreased performance and it crashed at 1990 mhz.
When I attempted to dual mine (Eth + SIA) after getting the clocks in the sweet spot the card just wasn't having it.
Best I could do was 16.4 mhs Eth and 460 mhs for Sia, it was forcing the card to draw too much power and the clocks were throttling hard down to like 750-850 mhz. It was also at 89 C dual mining.
Changed over to Eth only and got a much better 21 mhs with a stable 79 C and clocks around 1173 mhz. Power was hovering around 75-82 watts , the other card is an RX560 with a maxed out memory OC.
It is absolutely not a card built for sustained heavy load but it does manage and gets the job done decently. For my $250 I dont think I did bad, will mine with it and resell for more at a later time.
If you somehow get these for super cheap(less than $200) then its an instant buy
On paper it doesn't seem like a bad card. 1792 stream processors is like china's Rx 470D, a slightly cut down 470 and it has 8gb ram too. I think the problem is 75W tdp limit, it is limited by power draw, not core or memory. Have you tried severe undervolt or raising power limit? Eth only mining should increase hash also.
1000mhz clock, 1850 memory, 800mv. Use 1425 or 1500 straps.
Ill look into it later today but ill have to try and undervolt, raising the power limit is useless as its already at that 75 watt limit at default. Eth only mining does increase hash from 16 to 21.
undervolting and memory straps may yield excellent results. Still, you can tell its really struggling in that 75 watt power envelope.
Yeah undervolting would definitely be nice. The nice thing about these cards is you can cram them into a fullsize PC case and potentially end up with a 6 GPU rig. The same with the WX7100. This includes the use of the PCIE 1x port, which you can get adapters to adapt it over to x16 and it will not impact the performance for mining in the least...unlike video games. This is because the GPU doesn't need to constantly speak and potentially bottleneck the CPU. All the calculations happen elusively in the card. It would be the only reason to look at these cards, or the Quadro/ NVS single slot famly of cards. Of course, their cost would probably be harder to make up/ break even on. It's too bad AMD and Nvidia's most expensive cards are the low profile/ low TDP ones. You would definitely save on power, however.
Not sure why you'd have to OC the memory too much. The nice thing about this card is it's on a 256bit bus, which automatically grants you all the memory performance you could wish to get out of this card. It's when you are using >256bits that you really benefit from OC. It's the GPU I would maybe look at a little more --make up for some of those missing stream processors, in addition to undervolting. You'll crash trying to find that sweet spot, but you can find a nice compromise...even if it results in lowering voltage by -.050 and OC'ing +50MHZ, or lowering -.015 and OC'ing 100MHZ etc.
However, if you haven't already, Radeon Pro Adrenaline software has a Compute Mode. I have this mode with my Radeon Pro Duo(Polaris). It earns me an extra 1-2 Mega Hash/ sec when I do that. Not sure which drivers you are using, but it should be in Radeon Settings > Profiles > Global. The Radeon Enterprise Software doesn't have this mode. Also make sure you are using 17.7, or earlier. I seem to get worse hash performance with the latest. Considerably...like up to -5MH/S Oh...and the odd thing about Compute mode...I gain an extra 1-2 frames even in graphics benchmarks. Likely because benchmarks do a lot of physics/ compute scenes. Especially in DX12.
After flipping it over to Compute mode, lower the voltage -.025 at a time and record you average frames. This is going to tighten up the timing/ IPCs(instructions per cycle) in Compute and graphics scenarios. So if one side tells a better story, the other will as well. It doesn't always work that way, particularly with overclocking, but undervolting is like the universal language lol. You may fair better at either frames, vs, hashes, or vice-versa, but it's a positive and healthy and positive change all around...that is, of course, unless the card came a little on the undervolted side in the first place(not unheard of). You may as well further capitalize on that low TDP and drop the temps a couple degrees. Lower voltage theoretically means less resistance, means less heat, means better IPC performance(less latency). Regarding the throttling I recall seeing a tweak article that forces your GPUs to stay in a high priority state. A month ago I followed a Windows benchmarking tweak guide that suggested this. I wonder if that would help here?
Find HKEY_LOCAL_MACHINE\SOFTWARE\Microsoft\Windows NT\CurrentVersion\Multimedia\SystemProfile\Tasks\Games
Change the following registry values:
“GPU Priority” change its value to 8
“Priority” set to 6
"Scheduling Category" set to "High"
https://forums.redacted.tv/threads/ultimate-windows-10-gaming-tweak-guide.1205/
