Best bang for the buck is usually the 6132, just because it's so cheap, you can get another few hundred MHz of single core turbo (3.7GHz for the 6132, 4.2GHz for the 6144) but you're going to pay 3-6x for 20% more performance.
No love for Intel LGA3647?
- Thread starter AveryFreeman
- Start date
Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.
Thank you Rollo and appreciate your work helping us give new options to this platform!
Wow thank you, $20 for 14 cores
Sorry I misread your post. I thought you said nvme (generically); I didn’t see that you’d explicitly written u.2.
If there’s any chance you’re going to want to leverage optane pmem make sure to choose a cascade lake generation Xeon with support for your maximum memory capacity.
given how cheap the skylake xeons are it might be worth it to buy skylake now and throw it away/resell whenever optane enters the picture—that’ll give time for cascade lake parts to get cheaper—but I don’t like to buy things twice.
The Xeon 6132 Gold is 140w TDP per Intel Ark and Supermicro specs X11SPL-F to support 165w chips.
I believe this means I can run it without VRM mod?
I believe this means I can run it without VRM mod?
VRM mod is not a TDP thing.
e.g.
X11SPi-TF supports up to 205W
cpu 8222CL TDP=200W TDC=255A not work without VRM mod.
the BIOS of X11SPL allows TDP up to 165W, you can use any standard SKU up to 205W by removing BIOS TDP block.
Last edited:
Got it, so there is 2 separate mods:
- BIOS mod only - up to 205W TDP from 165W stock
- If considering high current chips such as cloud OEM SKUs (8222CL example), one also has to do the VRM mod
(I'll re-read your guide more closely) this evening
applies only to X11SPL and X11SPM (early BIOS 3.1 do not have TDP block)
I see, X11SPL is what I'm picking up, current BIOS seems to be up to 4.4. I wonder why they changed the TDP limit to be lower after 3.1
Last edited:
Got my X11SPL, it came with a Xeon Bronze 3106, 8 cores 1.7ghz max no HT, I think literally the lowest Xeon-SP one can get haha. I do have to say I'm impressed with how low power the (relatively for me vs. X9 X10 gen) platform is, idling in Windows at around 60watts with 2 SSDs and 4 fans.
Excited that I've joined the club
Excited that I've joined the club
I power tested 4114, 3204, 5122, and 5119T, the 4114 and 3204 were the same at idle, 5122 a couple watts higher, and 5119T a couple watts lower. I don't think those bronze Xeons were binned for efficiency.
Hence I'm very impressed so far, it's running at about same idle power as my small backup NAS AMD AM1 platform with a 35W TDP CPU. This will be me new homelab/server base
Checks out.
Asrock Rack EPC621D8A, 6x 32 GB DDR4-2666 Samsung, 2x 512 GB Optane PMEM, Intel 8259CL CPU, Noctua NH-U12S DX-3647 CPU cooler, two extra Noctua NF-A12x25 PWM to cool VRM/DIMMs/PCIe cards, LSI SAS 2008 SAS Controller, Mellanox ConnectX3, NVidia RTX A4000, two 120 GB boot SSDs, Micron 5100 MAX 960 GB, Samsung SM883 3.84 GB, Atheros ath9k_htc USB stick for quick wifi networking, Seasonic Prime Titanium Fanless 600W PSU.
Idle in Arch Linux, 88 Watts at the wall. I can get it to around 450 Watts total when compiling something on all cores, while simultaneously putting load on the network, the optanes, SSD I/O and on the GPU with gpu_burn.
The NVidia card would not idle below 10 W in P8 at first, needed persistence mode enabled. Otherwise it will idle at around 40 W in P2 once CUDA drivers are loaded. I check for the card in /etc/rc.local and if present, I use nvidia-smi to turn it on. With the 1 slot cooler previously this led to the slot bezel getting hot to the touch while "idle".
I tried to build an Epyc with the H11SSL-F, and I couldn't get the damn thing to boot. Thought it was faulty, ended up selling it at a massive loss because of assumed broken state. There was no error beeps, and it wasn't posting, so I had no real way to troubleshoot it. Threads sent me on a wild goose chase swapping out DDR4, trying it with different numbers of DIMMs, etc. The CPU socket with the pressure sensitivity, and exacting requirements for radius of turn, etc. was a big factor in issues troubleshooting. The next guy who got it ended up getting it working somehow, I can't remember exactly what he said, but I think in the end it was related to the BIOS. The whole experience was a complete nightmare, but I am glad someone else was able to use it (finally)
EPYC had its issues in the first generation, especially with the H11 boards. The H12 boards are much more polished and stable. It is also very important to use the proper screwdriver for the CPU socket.
This thread is convincing me to update my X11DAi-N from dual 6134 to 2nd Gen.
Installed and stress tested last night. Have to admit first time installing a server CPU with carrier was a bit daunting even after building machines for 20+ years. Especially mounting carrier to heatsink then dropping entire assembly into the socket.
Very happy with performance so far, and kudos to the Supermicro 4u heatsink, still whisper quiet with this chip running full avx2 stress test and cores at 70c.
Time to get Proxmox running, I was running esxi before.