Homeserver advice: i9-14900KS vs. i9-10940X
-
Need max AVX instructions. Anything with P/E cores is junk. Only enterprise P cores have the max AVX instructions. When P/E are mixed the advanced AVX is disabled in microcode because the CPU scheduler is unable to determine if a process thread contains an AVX instruction and there is no asymmetrical scheduler that handles this. Prior to early 12k series Intel, the microcode for P enterprise could allegedly run if swapped manually. This was "fused off" to prevent it, probably because Linux could easily be adapted to asymmetrical scheduling but Windows would probably not. The whole reason W11 had to be made was because of the E-cores and the way the scheduler and spin up of idol cores works, at least according to someone on Linux Plumbers for the CPU scheduler ~2020. There are already asymmetric schedulers in Android ARM.
Anyways I think it was on Gamer's Nexus in the last week or two that Intel was doing some all P core consumer stuff. I'd look at that. According to chips and cheese, the primary CPU bottleneck for tensors is the bus width and clock management of the L2 to L1 cache.
I do alright with my laptop, but haven't tried R1 stuff yet. The 70B llama2 stuff that I ran was untenable for CPU only with a 12700 with just CPU. It is a little slower than my reading pace when split with a 16 GB GPU, and that was running a 4 bit quantization version.
-
If you'll be running Linux and trying to use steam to run games, at all, avoid the 14th gen is.
If not, the 14th gen i9 is your bet.
Something with Proton, the layer that makes steam work with Linux, has been causing tons of people a lot of grief myself included. Any games that rely heavily on vulkan shaders will cause my whole system to crash under heavy load. It's a known thing and Intel still seems clueless as to what to do to resolve it, afaik.
-
Thank you! The AMD-Route sounds also promising, but I'm not sure about there idle power consumption. they say that intle is bettle in idle mode.
But i'm not sure if proxmox can handle the E-cores properly. -
Thank you! I will look into Gamer's Nexus.
-
In my experience, idle power consumption mainly depends on the mainboard used. The processors all(?) clock down to some more or less energy-efficient level. But the specific design of the mainboard and the components on it could double or half energy consumption.
-
The 14th gen didn't only have problems with Linux and I still don't trust Intel when they say they "fixed" the CPUs disassembling themselves. Given the money involved I'd definitely advise against 13th and 14th gen Intel in any case, just not worth the potential headache down the road. Either go with older Intel or AMD.
-
https://www.cpubenchmark.net/compare/3639vs5957/Intel-i9-10940X-vs-Intel-i9-14900KS
The new one is significant more powerful, and almost certainly has better efficiency.
I would just verify that your OS properly supports P/E cores.
-
You're not going to run deepseek r1 without GPUs (plural).
-
At which altar should we pray to the idol cores?
-
10940X
"They say", but they're right. Ryzen chips do have worse idle power usage, but you're talking about 10w or so, at most.
And uh, if you were looking at an X-series CPU, I can't see how that 10w is a dealbreaker, because you were already looking at a shockingly inefficient chip.
-
you can, but it'll be the distilled versions that are significantly less impressive.
https://apxml.com/posts/gpu-requirements-deepseek-r1
Newer versions also introduce memory mapping so they're technically only bound by your storage capacity (but they're way slower in practice).
-
Might be a bit late on this, but ProxMox doesn't really handle assigning threads to the e/p cores. That's handled by the kernel and as long you're running kernel version 6.1 or greater you should be good on that front.
If you really need to, you can also pin specific VMs to specific cores. So that if you've got something that always needs the performance it can always run on the p-cores and things that aren't as demanding can always run on e-cores.
That said, especially if you're over provisioning, it's probably better to let the scheduler in the kernel handle thread assignments.
