Cores vs threads, which get used first when setting up slots

[Lockheed_Tvr]

As I add cores in a slot, in what order does either the client or Windows apply them? In a 4 core, hyperthreaded CPU, are 1, 2, 3, and 4 cores and 5, 6, 7, and 8 threads? Or do they alternate?

I have a box that just runs hot. I've cleaned up the heatsink, applied new thermal compound, but it maxes out temps with 6 out of 8 cores/threads working. It does okay if I drop it down to 1 core, but if I add a second core in the slot, will it add a core or a hyperthread? If it adds a thread on the same core, the temps might stay down, if it adds a core, I might overheat.

I've tried reducing the folding power and adding cores but the temps spike again.

It's an older Dell server so I doubt that an aftermarket cooler will fit. I'm guessing I just have a hot cpu. I'm mainly using it to serve the two GPUs in it so it won't break my heart if I don't get a ton of cpu points out of it, it just sucks that there's an 8 thread Xeon that isn't doing what it can.

[toTOW]

Define "hot" ...

My old CPUs (1st generation i7s) have been running for years at 80°C and they are still doing perfectly fine ...

[JimboPalmer]

The applications just gets threads from the OS, it is not allowed to pick which ones it wants.

[scerbera]

Try 2 free programs, throttle stop and t throttle. T throttle you can set a max temp, and once it reaches that it throttles down to let it cool. throttle stop lets you adjust the multiplier very easily which will also reduce temps.

[Lockheed_Tvr]

Define "hot" ...

My old CPUs (1st generation i7s) have been running for years at 80°C and they are still doing perfectly fine ...

95 to 98 C.

[Lockheed_Tvr]

Try 2 free programs, throttle stop and t throttle. T throttle you can set a max temp, and once it reaches that it throttles down to let it cool. throttle stop lets you adjust the multiplier very easily which will also reduce temps.

Good to know. I will check those out.

[Lockheed_Tvr]

The applications just gets threads from the OS, it is not allowed to pick which ones it wants.

Fair enough. Thanks.

[Neil-B]

Might be worth checking intel site for the Tjmax temp for the CPUs in your server … For enterprise grade servers, under reasonably consistent high CPU load Xeons will be "safe" (or as safe as anything ever is) even just a few degrees below Tjmax is where most FRU/SDR/BMC setups will balance the fans when setup for aggressive CPU performance - though the fans may be screaming quite well at this point.

All depends on your kit/setup … But, for instance, my Xeons run 80C-85C dependant upon the WUs in play (some appear harder on the CPU than others) with the Tjmax for them being 93C … Monitoring max temps for each core occasionally shows one peak up to 90C but the system balances the CPU freqs and fans to keep as high a clock as it can on max cooling.

As an aside, if you are worried you may find you can "neuter" the Xeons and turn off Threading in the BIOS - It halves the core count (will impact overall processing throughput) but can help manage the CPU temps better - though not sure if this is for all Xeons or just the mobile ones.

"Most" servers should be able to handle cooling their CPUs under max load as this is one of their standards functions - though they may get a bit noisy … If any tweaking has been done (not perhaps as usual on a server as a gaming machine) then it might struggle.

However, ignoring all this … Only run your kit as warm as you feel comfortable !!

[Lockheed_Tvr]

Might be worth checking intel site for the Tjmax temp for the CPUs in your server … For enterprise grade servers, under reasonably consistent high CPU load Xeons will be "safe" (or as safe as anything ever is) even just a few degrees below Tjmax is where most FRU/SDR/BMC setups will balance the fans when setup for aggressive CPU performance - though the fans may be screaming quite well at this point.

All depends on your kit/setup … But, for instance, my Xeons run 80C-85C dependant upon the WUs in play (some appear harder on the CPU than others) with the Tjmax for them being 93C … Monitoring max temps for each core occasionally shows one peak up to 90C but the system balances the CPU freqs and fans to keep as high a clock as it can on max cooling.

As an aside, if you are worried you may find you can "neuter" the Xeons and turn off Threading in the BIOS - It halves the core count (will impact overall processing throughput) but can help manage the CPU temps better - though not sure if this is for all Xeons or just the mobile ones.

"Most" servers should be able to handle cooling their CPUs under max load as this is one of their standards functions - though they may get a bit noisy … If any tweaking has been done (not perhaps as usual on a server as a gaming machine) then it might struggle.

However, ignoring all this … Only run your kit as warm as you feel comfortable !!

Thank you. It looks like Tjmax is 98C. That's good information regarding the robustness of the Xeons. I'm not that up on server hardware and temps over 80 were looking scary to me!

[Neil-B]

Part of it is how servers tend to be loaded (fairly constantly is preferable) compared to workstations/gaming rigs/laptops which tend to have lower/spikier CPU loads … As an HPC contact explained to me years back - running hot isn't much of an issue (actually it makes the cooling more efficient) - it is the heat cycling that induces the most wear/weakness so running cold systems that spike hot occasionally may well stress CPUs more than a constantly hot system … having said that if you can run constantly pegging the CPU to it's limits and keep it cold all the better - but most servers not to be set up like that - they just screams fans hard and keep the CPUs below anything "bad".

If anything above 80C looks scary to you then try to find a setup below that as it all boils down to being within ones comfort zone (apologies for the pun) … better happy and safe than stressed and sorry … whatever setup you end up with a good monitoring software is always nice to have as you will get used to what is you normal level - different WUs can load up CPUs in different ways - an average 5C swing across all cores is possible - I use HWMonitor but there are others that other people prefer.