x99 platform: still worth in (almost) 2025?

[faktorqm]

Hi there, I'm looking forward to mount a server-grade hardware for virtualization purposes. currently an esxi 8 user with consumer-grade hardware. As this is for my home, I usually go cheaper. I want to have some servers like gitlab, petalinux (for fpga), next cloud, openoffice, torrent box, etc.
What about those Supermicro x10dri? with two e5 26XX v4 and 512gb of ram they are still a good catch looking forward to serve at least 5 years (2025-2030)?
thank you for your advice!

 

[Chriggel]

While it's similar to the C-series server chipsets of the time, the X99 was aimed at the HEDT customers. X10DRi uses C612. Obviously this is still usable, if you have it already or can get it cheap. It's not worth paying much money for it, unless you need lots of memory, for example. Do you really need 512 GB RAM for that?

Otherwise the AM5 platform is a strong competitor, much more modern, powerful CPUs and ECC support.

 

[CyklonDX]

I'd ask what kind of virtualization you are looking for;

if you plan on gaming with gpu passthrough - you will be loosing decent chunk of fps vs zen2-4/5 platforms. As games always like having strong single threaded performance; But its not that big of an issue to run even most modern titles with most modern gpu's
(just don't expect to be able to fully utilize top-end gpu's for games.)

If you don't need gaming, v4's will serve you well into 2040's - just need to have enough memory, and cores.
AI and stuff with high-end gpu's might become a problem at some point before 2030.

 

[kapone]

I usually go cheaper

Cheaper than...what? X99 is...what it is. It's not meant to be a server platform. Buy a proper server platform, assuming that generation, like Cisco 240 M5.

https://forums.servethehome.com/index.php?threads/cisco-c240-m5-2x-6132-128gb-sas3508-240.46081/

 

[acquacow]

I sold 3 of my X99 machines, still have two as backups. Got real tired of trying to get USB 3 to work right on x99 and went AM5 for my main desktop.

 

[faktorqm]

thanks for your answers. I think I'm confused about platform naming. c612 is not x99?

for example: Supermicro x10dri + 2x e5 2680 v4 (14c 28t) and 512gb of ram (16x 32gb ddr4 ecc).

this results in 56 threads of cpu. Almost 8gb of ram per each cpu.

I don't want to run any gaming. all virtual machines are doing backend stuff. I will use fpga simulation, one gitlab instance, one openoffice, a nextcloud, all the software needing intensive cpu/memory tasks.

regarding the price, I said go cheap because the e5 2680 v4 is 20usd on Aliexpress each. 512gb ddr4 2400 ecc is about 350usd and x10dri 150usd. 550usd for the kit, I think it's not bad.

sounds reasonable for my use case? this platform worth the money for the next 5 years? thank you all for your answers. I'm getting a better picture

 

[CyklonDX]

c612 is server chipset, x99 is consumer chipset.
x99 might be better in certain areas like overclocking and overall 'hacked' chinese features with dual x99.
But if you plan on using it for server loads, i recommend proper server motherboard like supermicro's.

 

[Chriggel]

Indeed, C612 and X99 are related, but they're different SKUs.

If you need 512 GB RAM, it could be reasonable. You could also consider a slightly more modern 3647 platform. Or go AMD with an Epyc system of that time.

If you don't, then it depends on how much you actually need. The CPUs aren't really a compelling argument for this platform, because yes, 28 cores across two CPUs, but the individual cores aren't that fast by todays standards. PCIe lanes could an argument for such an older server platform though.

 

[larrysb]

I've been a hardcore Xeon E5 on X99 workstation user for years now. Honestly, they've been absolutely great. We did a lot of AI development on Asus WS--E X99 SAGE/10G boards with various Xeon E5-V4 processors. The WS-E boards will handle as many as 4 GPU cards, all on by 16x PCIE lanes. The Broadwell Xeons have been incredibly stable. Some of our WS boards have 10G networking, and we wound up networking them with Mellanox cards at 25mb x2 ports, and enabled RDMA to move data around really fast between systems and GPU's. These things would run machine-learning training jobs for days on end with multiple GPU's.

The only downside to X99 is that it only supports 1 cpu. Not much of a big deal for workstations but maybe you want multiprocessors, which would point at the C612. If you're happy with a uniprocessor server, then X99 will do the trick. Nvidia built their $60k 4-Tesla GPU workstations out of Asus WS-E SAGE 10/G boards. I figured it was good enough for me back then too. X99 will happily run Xeon E5-26xx (2S multi-capable) as well as Xeon E5-16xx (1S only) as well as Core I7 and Core I5 Broadwell and Haswell processors.

I'm kind of leery of the China-ecosystem X99 boards. I think it would be better to find quality examples of used or new-old-stock X99 boards from known manufacturers. If you're gaming, with older or Xeons those china-ecosystem X99 would be fine, but if you care at all about the work you're doing, then I'd hunt down quality components.

I'm about to clear out my X99/Xeon E5 stock. I sold my company about 4 years ago, and have a few high quality X99 boards left in the closet and some Xeon E5 CPU's and a few consumer GPU's and a few Mellanox cards and cables. I sold off a few complete systems recently for CAD/CAM use and I'm still using one as a personal PC, mostly Linux for development work and Windows 10. Microsoft isn't supporting Xeon E5 on Windows 11, though you can edit registry to allow the upgrade and it seems to work just fine.

Broadwell in my opinion was a rock-solid CPU, especially in the Xeon range and the X99 did a great job as a chipset.

 

[faktorqm]

thanks a lot for your valuable insight!

 

[Vit K]

X99, especially chinese mobos are dirt cheap, but very power inefficient. And brands / cases are even more power hungry, with their redundant fans, old hot NICs and IPMI chips.
If you want ECC, you can go with 5700x3D or 5900 and 128 Gb of DDR4 ECC in compact mATX case. It will give you raw performance of two 14 cores V4 at worst case in less that 100w. If you not planing to scale to thousand of users right away, 128 Gb is more than enough for a single node and few clusters of k8s and postgres. I bet you won't be able to consume even half of ram for what you are intending.