... I had expected greater differences in run times. I'll continue to run this host with SpeedStep ON. It will run cooler and that's what I prefer most of the time.
Like your test host, many of mine date from the 2008 - 2010 era and are not UEFI, but have the same sort of BIOS options you mention. I had intended to retire many of them but decided to try Polaris series GPUs (RX 460 to RX 580) just to see what would happen. I compared the same GPU in a modern host (Haswell refresh and Kaby Lake) against the old clunker. I was surprised to find the GPU performed reasonably similarly despite the speed and architecture difference of the CPU/motherboard. So the oldies got a bunch of GPUs instead of retirement :-).
SpeedStep had always been off because they were initially CPU crunchers only. Elapsed time for crunching GPU tasks seems relatively independent of CPU speed so I guess I should enable SpeedStep on them now.
As far as your results are concerned, the change in CPU time is quite noticeable - 218 -> 175 is a fair percentage. However, just as I saw with my old vs new comparison, the elapsed time changes by a much lower percentage. So for anybody wanting to get more out of an existing, post-2010, CPU only cruncher, the cheapest way to do so is to put a modern low power GPU - one that doesn't need a PCIe power connector if you don't want the electricity bill to rise too much. You just need to check the suitability of the existing PSU, particularly if it is also 'original' and therefore possibly unsuitable not only because of the extra load from the PCIe slot but also because of its age and probable internal condition. If it's 5V/3.3V heavy you should replace it.
As an example, I run some RX 460s and 560s with no extra connector using 300W PSUs that are rated at 270W @12V. That is cutting it a bit fine because a machine does pull around 170-180W from the wall. I originally bought these back in 2006 and virtually all of them have had some capacitor replacements and fan re-lubes but they still soldier on :-).
I want to add that I tried turning CPU tasks off on my PC with and also got similar results to the original post. I have an i5-4590 and a RX 570 running 2 tasks at a time.
1. Running 2 CPU tasks and 2 GPU + CPU tasks at a time, my GPU tasks are around 17-17.5 minutes each.
2. When I turned off CPU tasks in Einstein account settings, I was running only 2 GPU + CPU tasks (2 tasks at a time on my GPU), leaving 2 CPU cores free to do nothing, I saw my task completing times increase to 20-20.5 minutes each.
It's interesting, as my i5 is designed to increase core frequency more when there are less cores being used. It boosts to 3499 mhz on all 4 cores when they are all in use, but when a single core is in use, it will go to 3699 mhz.
It's really surprising to see a slowdown in GPU tasks, when not running CPU tasks.
Richie wrote:... I had
)
Like your test host, many of mine date from the 2008 - 2010 era and are not UEFI, but have the same sort of BIOS options you mention. I had intended to retire many of them but decided to try Polaris series GPUs (RX 460 to RX 580) just to see what would happen. I compared the same GPU in a modern host (Haswell refresh and Kaby Lake) against the old clunker. I was surprised to find the GPU performed reasonably similarly despite the speed and architecture difference of the CPU/motherboard. So the oldies got a bunch of GPUs instead of retirement :-).
SpeedStep had always been off because they were initially CPU crunchers only. Elapsed time for crunching GPU tasks seems relatively independent of CPU speed so I guess I should enable SpeedStep on them now.
As far as your results are concerned, the change in CPU time is quite noticeable - 218 -> 175 is a fair percentage. However, just as I saw with my old vs new comparison, the elapsed time changes by a much lower percentage. So for anybody wanting to get more out of an existing, post-2010, CPU only cruncher, the cheapest way to do so is to put a modern low power GPU - one that doesn't need a PCIe power connector if you don't want the electricity bill to rise too much. You just need to check the suitability of the existing PSU, particularly if it is also 'original' and therefore possibly unsuitable not only because of the extra load from the PCIe slot but also because of its age and probable internal condition. If it's 5V/3.3V heavy you should replace it.
As an example, I run some RX 460s and 560s with no extra connector using 300W PSUs that are rated at 270W @12V. That is cutting it a bit fine because a machine does pull around 170-180W from the wall. I originally bought these back in 2006 and virtually all of them have had some capacitor replacements and fan re-lubes but they still soldier on :-).
Cheers,
Gary.
I want to add that I tried
)
I want to add that I tried turning CPU tasks off on my PC with and also got similar results to the original post. I have an i5-4590 and a RX 570 running 2 tasks at a time.
1. Running 2 CPU tasks and 2 GPU + CPU tasks at a time, my GPU tasks are around 17-17.5 minutes each.
2. When I turned off CPU tasks in Einstein account settings, I was running only 2 GPU + CPU tasks (2 tasks at a time on my GPU), leaving 2 CPU cores free to do nothing, I saw my task completing times increase to 20-20.5 minutes each.
It's interesting, as my i5 is designed to increase core frequency more when there are less cores being used. It boosts to 3499 mhz on all 4 cores when they are all in use, but when a single core is in use, it will go to 3699 mhz.
It's really surprising to see a slowdown in GPU tasks, when not running CPU tasks.