Is there any way to get further debug information out of the O2MDF application? Are the developers monitoring this bug-reports forum?
shift wrote:
Hi Justin, I think the moderators are as close as you'll get...but what would I know?
Hi shift, you probably know quite a bit :-).
The Devs are perennially busy - University style research facilities are always trying to do twice the work with half the staff - but the overworked technical staff do try to find time to deal with genuine issues. Essentially, they are forced to rely on experienced volunteers to handle general issues that don't really need their immediate attention. They always step in on the big issues.
Apart from the moderator tag, I regard myself as part of that experienced volunteer group. If I think I know even just a little bit about an issue, I will always try to respond if I think I can help.
I know absolutely nothing about Macs or the innards of MacOS. Sure, it's originally based on BSD Unix (a giant plus in my opinion) but Apple have really buried a lot of stuff in there that is hidden from the ordinary user and I would feel totally inadequate to offer any sensible advice about the questions Justin has asked.
My opinion (for what it's worth) aligns pretty much with the comment you made. Apple probably are showing video utilization rather than compute utilization but I'm not comfortable making that claim because I have no evidence to justify it.
The original issue of extremely slow GW task crunching is probably an issue for Apple to explain. The O2MDF app (at the time of the original complaint) was working fine on other architectures so it's hard to know what the real cause was - other than to suspect that Apple's implementation of OpenCL may have something to do with it. Once again, I don't know so I'm reluctant to speculate about it.
With Justin getting better results using the GRP search, that's probably what he should continue to run. I have lots of RX 570 GPUs and they perform very well on both GW and GRP tasks. For the former S2 variant of O2MDF, an 8GB RX570 was running the tasks at a multiplicity of 4 (4x) and producing 4 results about every 36 mins (9 mins per result). For the current GRP tasks, I'm running 2x and producing 2 results every 13-14 mins (less than 7 mins per result)
Justin's latest GRP results are taking close to 20 mins but I don't know if he is running multiple concurrent tasks or not. If he is running single tasks per GPU, it would be further evidence that perhaps Apple has done something to throttle GPU compute in some way. Apple make slim/elegant enclosures that must be hard to keep from overheating if compute was allowed to run full bore. Once again, simply speculation since I don't know how good the cooling system might be. With twin RX 580 GPUs, maybe it's not a standard Apple enclosure and maybe cooling is not the problem I'm imagining.
Justin Gregg wrote:Is there
)
Hi shift, you probably know quite a bit :-).
The Devs are perennially busy - University style research facilities are always trying to do twice the work with half the staff - but the overworked technical staff do try to find time to deal with genuine issues. Essentially, they are forced to rely on experienced volunteers to handle general issues that don't really need their immediate attention. They always step in on the big issues.
Apart from the moderator tag, I regard myself as part of that experienced volunteer group. If I think I know even just a little bit about an issue, I will always try to respond if I think I can help.
I know absolutely nothing about Macs or the innards of MacOS. Sure, it's originally based on BSD Unix (a giant plus in my opinion) but Apple have really buried a lot of stuff in there that is hidden from the ordinary user and I would feel totally inadequate to offer any sensible advice about the questions Justin has asked.
My opinion (for what it's worth) aligns pretty much with the comment you made. Apple probably are showing video utilization rather than compute utilization but I'm not comfortable making that claim because I have no evidence to justify it.
The original issue of extremely slow GW task crunching is probably an issue for Apple to explain. The O2MDF app (at the time of the original complaint) was working fine on other architectures so it's hard to know what the real cause was - other than to suspect that Apple's implementation of OpenCL may have something to do with it. Once again, I don't know so I'm reluctant to speculate about it.
With Justin getting better results using the GRP search, that's probably what he should continue to run. I have lots of RX 570 GPUs and they perform very well on both GW and GRP tasks. For the former S2 variant of O2MDF, an 8GB RX570 was running the tasks at a multiplicity of 4 (4x) and producing 4 results about every 36 mins (9 mins per result). For the current GRP tasks, I'm running 2x and producing 2 results every 13-14 mins (less than 7 mins per result)
Justin's latest GRP results are taking close to 20 mins but I don't know if he is running multiple concurrent tasks or not. If he is running single tasks per GPU, it would be further evidence that perhaps Apple has done something to throttle GPU compute in some way. Apple make slim/elegant enclosures that must be hard to keep from overheating if compute was allowed to run full bore. Once again, simply speculation since I don't know how good the cooling system might be. With twin RX 580 GPUs, maybe it's not a standard Apple enclosure and maybe cooling is not the problem I'm imagining.
Cheers,
Gary.