Search results

Milan is about 20% faster than Rome across similar SKUs. Milan idles worse (the IOD is more sophisticated), but you might be able to replace a 64c Rome with a 48c Milan at similar performance and save some power under load.

It's not obvious to me what the concern is - 18TB drives have only existed for two years so even if the labels are completely fake they are not that old. On the other hand, going and buying a drive in a bag from a random guy on eBay and expecting it to actually be new is kind of silly, they are...

Because they only perform well for select benchmarks (e.g. SPECrate) that require zero interprocess communication. You can get crazy high SPECINTrate, Passmark, or Geekbench scores out of them, but otherwise performance is a crapshoot and you end up with anomalous behavior where your application...

This board more or less maxes out what AM5 can do. The x16 root complex is bifurcated into x8 + x8, then 4 of the remaining 8 CPU lanes are routed to a third slot. The other 4 go to a pretty useless PCIe 5.0 M.2 slot. The 4 chipset lanes route to the X670E, which muxes them ~16 lanes. 4 of them...

A P2000 has dual NVENCs with unlimited sessions for $200 so for strictly Plex work it's likely better, a P4000 is about the same price. Plus they have features like fans and display outputs.

I've had nothing but mediocre experiences with them, unlike most of the large scale eBay scrappers they have a huge attitude towards returns. Usually when you get something DOA from one of the big sellers you send it back, no hassle, and they'll send out a new one. UnixSurplus tried fairly hard...

Looks like GP100 has 3 NVENCs but for some reason they are capped at 4K x 4K frames?! Seems silly to use a double-precision-optimized datacenter card for Plex transcoding unless you have some other reason to need dp performance though.

Works fine with 2400, I've never tested 2133. Rome is sensitive to memory speeds though, since FCLK scales with memory speed; I think this may be the only time in...recorded history where memory speed and channel population can make a tangible (>a few %) impact on performance. Xeons behave well...

Smallest QS is the 32-core which is around $800, but proceed with caution: we're pretty data deficient for running the 32-core QS at high clocks and voltages for extended periods of time and its unclear if they'll degrade.

Hah, if I had a CPU for it I'd put up a firmware dump. Anyone familiar with Lenovo systems know what exactly has to be dumped? Presumably it would be good to preserve an UEFI image plus whatever they use for management.

I have a somewhat obscure engineering sample of a SR655 2U system - this is a 1P Epyc 7002/7003 server. It has the special feature of *not PSB locking* your CPU despite being a Lenovo. Unfortunately the configuration it came with (8LFF front + LFF rear cage) isn't what I'm looking for, and parts...

That's a square ILM board, so *anything* that supports 2011/2066 will work. Any single-tower 120mm cooler should work fine - I'd recommend the PA120 but I don't think two of them will fit.

i've never heard of any issues with HBM, aren't there tons of HBM2 based cards deployed in datacenters?

Neat. There's an AOM-SXMV on eBay right now p Supermicro GPU Board PN: AOM-SXMV /p | eBay in case anyone wants to give it a stab.

Does this...actually work?! The OcuLink ports on the AOM-SXMV are indeed routed to the PLX switches, but the other end goes into the bottom slot of the riser and is designed to interface with an Infiniband card plugged into that slot. "Officially" each of the card edge connectors is a PCIe x16...

The thing is, the 7950x is so good that you have to really, really want those extra PCIe lanes to move to Threadripper. The CPU generates 24 PCIe lanes, all of which are at least PCIe 4.0 (actually, 20 of them are PCIe 5.0, but you can't get any peripherals yet), which is 48 PCIe 3.0 lanes worth...

I wouldn't even bother in 2022 - if you really want to reuse the CPUs, you can pick up a T5600 (or just the board out of one, if you're feeling stingy) for basically nothing. The s6500 is a nice cluster-in-a-box and still pretty relevant if you want tons of RAM for VMs, but gutting the nodes and...

I've found that AMD and Nvidia both work, but AMD cards have trouble showing the POST screen. Do note the H11SSL is PCI-e 3.0 only which is kind of a bummer if you plan on using a bunch of PCIe drives.

I am not convinced they actually consume more power than a modern server (1U's aren't particularly efficient regardless) but you can get an Optiplex or something similar for about the same price if all you are after is an educational experience. The real problem is OP is trying to *sell* them...

Seems like a solid case for an Asrock X570D4U, a 5700X, and 64GB of ECC UDIMMs. You can get a much cheaper board but the Asrock gives you guaranteed ECC support and IPMI, both of which are handy.