Hardware woes

[Prime95]

Affinity issues are rare these days. Hwloc has made prime95's handling of affinity much better.

[kladner]

I watched HWINFO with one and two 4 core workers. Even with just one such worker the four cores rarely show 100% usage at the same time. Sometimes there are none or one. Turn on the other worker and all eight cores running look much worse, fluctuating from a couple to several percent down. I'll have to install the other pair of DIMMs tomorrow. Hope that helps.

[kladner]

It took longer to get here than I expected. The memory stick I had gotten first could not be installed. The narrow motherboard put the first slot under the edge of the CPU cooler. The DIMMs were nearly 1/4 inch too tall with heat sinks. Fortunately, Crucial Ballistix were low-rise enough to just make it.

Memtest86 did not find anything wrong with the new RAM collection. When I started benchmarking and torture testing in P95 I was startled to see that the CPU was spiking 10-15 C above where it had been: hitting 80 C when it had been hanging in the mid 60s. Core usage numbers had also gone up.

Leaving aside a bunch of messing around and poking at things, I'll offer an unusual, for me, set of screen shots of P95 running at different CPU speeds, from 3200 MHz to 4500 MHz. There are interesting similarities between them. When I noticed these, I repeated some of the test runs, with essentially the same results. The file names for the images include the clock speeds. In ascending order they are 3.2 GHz, 4.0 GHz, 4.3 GHz, and 4.5 GHz. Across the range, the throughputs in the last few lines of the workers are essentially the same.

It seems that this defines what this memory array is able to support. I will try successively lower clocks to see how far this extends before the numbers fall off.

EDIT: Here's a run at 3000 RPM MHz

[chris2be8]

Originally Posted by kladner

It took longer to get here than I expected. The memory stick I had gotten first could not be installed. The narrow motherboard put the first slot under the edge of the CPU cooler. The DIMMs were nearly 1/4 inch too tall with heat sinks. Fortunately, Crucial Ballistix were low-rise enough to just make it.

Memtest86 did not find anything wrong with the new RAM collection. When I started benchmarking and torture testing in P95 I was startled to see that the CPU was spiking 10-15 C above where it had been: hitting 80 C when it had been hanging in the mid 60s. Core usage numbers had also gone up.

Could the extra memory be restricting air flow round the cooler? Or could you have slightly unseated the cooler getting the memory in?

Chris

[kladner]

Originally Posted by chris2be8

Could the extra memory be restricting air flow round the cooler? Or could you have slightly unseated the cooler getting the memory in?

Chris

The cooler airflow is above the memory. I was more concerned about airflow for the memory. My poorly expressed intent was that with more memory bandwidth the CPU was suddenly working a lot harder.


I currently have that machine running locked at 4GHz, running P95 on 6 of the 8 cores. It's getting a bit under 2.5 ms/it on CAT 4 DCs, at 62 C. In the initial run after the RAM upgrade the system was running 'stock' with Burst taking it to 4.5 GHz with 8 cores running.


I also found out this evening that this mobo/i7 combination does not enable hyperthreading. Not a really big deal, but surprising.

[sdbardwick]

Originally Posted by kladner

I also found out this evening that this mobo/i7 combination does not enable hyperthreading. Not a really big deal, but surprising.

Not the combination, just the processor. Yes, it is a non-hyperthreaded i7.

[kladner]

Originally Posted by sdbardwick

Not the combination, just the processor. Yes, it is a non-hyperthreaded i7.

Thanks! I did not know that there were such i7s. I assumed that it would have similar features to the 6700K I am running. In this situation I chose a complete system, so the processor is what came with. Again, I don't see a lot of need for HT. The CPU is running circles around the RAM as it is, and the current setup has two cores free anyhow.



I am finding that there is minimal difference between running locked at 4GHz and 4.3GHz: 0.05 ms, max. Temp goes from 64C to 69C. Seems that 4GHz (or less) should be adequate.

[VBCurtis]

Just when I upgraded my CPU from i7-5820 to Xeon-2580v3, my Corsair water cooling system seems to have failed.

I'm getting core temps (via "watch sensors" in ubuntu) of 55-60C idle, pinned at 96C and throttling under LLR. The machine ran fine for a week or so after swapping CPUs, but slowly overheats under load now- that is, temps jump quickly to 90C with 1 thread LLR, then over an hour or two slowly keep climbing.

The corsair water pump/block for this all-in-one system (possibly H100, don't have box anymore- 140mm radiator with push/pull fans out the back of the case) has an internal LED lighting doohickey. Upon boot, it's the usual white LED, but after a few min it goes to an oscillating yellow / dark pattern. The hose to radiator is quite warm to the touch, while return hose seems room-temp.

I initially suspected a bad waterblock-to-CPU interface, so I reinstalled the block with more care taken to a thin layer of paste and tightening all 4 screws in rounds so that one corner didn't end up tilted upward. Temps did not improve.

Bad pump? Googling the yellow Corsair light didn't yield much.

[paulunderwood]

Originally Posted by VBCurtis

Just when I upgraded my CPU from i7-5820 to Xeon-2580v3, my Corsair water cooling system seems to have failed.

I'm getting core temps (via "watch sensors" in ubuntu) of 55-60C idle, pinned at 96C and throttling under LLR. The machine ran fine for a week or so after swapping CPUs, but slowly overheats under load now- that is, temps jump quickly to 90C with 1 thread LLR, then over an hour or two slowly keep climbing.

The corsair water pump/block for this all-in-one system (possibly H100, don't have box anymore- 140mm radiator with push/pull fans out the back of the case) has an internal LED lighting doohickey. Upon boot, it's the usual white LED, but after a few min it goes to an oscillating yellow / dark pattern. The hose to radiator is quite warm to the touch, while return hose seems room-temp.

I initially suspected a bad waterblock-to-CPU interface, so I reinstalled the block with more care taken to a thin layer of paste and tightening all 4 screws in rounds so that one corner didn't end up tilted upward. Temps did not improve.

Bad pump? Googling the yellow Corsair light didn't yield much.

https://help.corsair.com/hc/en-us/ar...AIO-Cooler-FAQ

Check the pump and fans are on the right headers.

Also there might be a BIOS setting -- turn off Cool'n'quiet or whatever it is called.

[kladner]

This is not a woe, but an update. I got a Kill a Watt today. I found that the i7-9700K machine idles at 40 W. P95 pushes it to 160 W. This with the CPU running at 75%, six cores allocated to one worker at 4400 MHz. Using the integrated graphics. No GPU.

I'll move the K-a-W to the 6700K machine when I get a chance tomorrow. I expect a much higher draw, what with a GPU running mfaktc, a bunch of spinning rust devices, and more fans. The newer machine has only an NVMe SSD so far. Dang! That thing boots Win 10 in a few seconds. Right now there's no added drivers to load, which helps, I'm sure.