Intel Xeon E-2400 vs AMD EPYC 4004

Which would you go for?

  • Intel Xeon E-2400

    Votes: 7 50.0%
  • AMD EPYC 4004

    Votes: 7 50.0%

  • Total voters
    14
FreeBSD Fans,

As you all know, AMD released the EPYC 4004 series this week. This caught my eye since I've been looking to buy a little server for mixed FreeBSD and ESX use. 'til now I'd only been looking at the Xeon E-2400 family.

I've put a poll in this thread to see which you all like more.

My thoughts:

I'm partial to Supermicro. And I'm also partial to RAM made by Micron, Hynix or Samsung--none of which make 48 GB ECC UDIMMs (wiping out AMD's RAM limit advantage).

So, has anybody noticed that Supermicro only offers a workstation motherboard for the EPYC 4004 family (H13SAE-MF)?

Since the EPYC 4004 family includes a GPU this doesn't surprise me too much.

But as far as using this motherboard in a server goes, beyond the wasted USB and video outputs, its slot configuration isn't ideal; it does actually have an x4 slot, but Supermicro's own integrated offering (AS-3015A-I) obviates its use due to CPU cooler interference. This struck me as bizarre so I confirmed it with their support department--it's for-real; you're left with but two x8 slots.

AttributeIntel (SYS-531R-I)AMD (AS-3015A-I)Winner
Speed1.00~1.88 (about 88% faster, depending on the benchmark)AMD
RAM limit128 GB192 GBAMD
NVMe slots12AMD
Price1.001.13 (13% higher)Intel
FreeBSD compatibilityUnknownUnknownUnknown
ESX HCLYes (the E-2488, at least)NoIntel
ESX nested virtualizationFull supportLimited supportIntel
PCI-e slots42Intel
SATA ports84Intel
Lowest CPU power draw during benchmarks19 W37 WIntel
 
What flavor of FreeBSD do you want to run? If it is -current, -stable or if you want to build ports like Chrome or anything electron-based you definitely want those 16 cores in the EPYC.

What is HCL?
 
Thanks for replying cracauer@ and Jose.

I'll be running RELEASE and using packages. I do run a custom kernel with QLogic FC HBA target mode enabled, but the kernel compile times are already okay on my 2015-era hardware.

HCL is indeed a VMware thing (as is ESX); it's their Hardware Compatibility List. I know that's off topic for this forum but I figured I'd mention it since I'll be flip-flopping between FreeBSD and ESX.
 
You'll need at least 14.1-RELEASE and it's basically a Ryzen CPU with a few more instructions (ISA) and faster IOD. I would guess that it performs pretty much the same in most cases compared to its counterpart. To take advantage of the CPU you likely want to recompile packages to match your CPU arch (multimedia related at least and likely scientific too) there are however a few potential optimization bugs in LLVM/Clang that may reduce performance (they're being worked on upstream) using zen* targets.
 
Unless you’re buying the very fastest cpu it never makes sense to buy the lastest cpu. Last years model is 1/3 the price (less used) so for the same price you can usually get a faster cpu with more cores. And you don’t have to worry about compatibility.
 
Based on the survey question, I assumed we were talking AMD 4004 or and Intel E-2400 CPU, not a prebuilt system with specific motherboards. When I read the chart about AMD/Intel wins, I was taken off guard by the AMD only supports two PCIe slots. Well that certainly is not the case. That specific motherboard is the limiting factor, not the CPU. the CPU has 28 lanes of Gen 5 PCIe lanes, that is a lot. Depending on the motherboard you can have quite a lot of PCIe slots. And when looking at those two slots on the AMD configured board, I'm certain those are a x16 and a x8 wired (or could be x4) slots.

The motherboard matters just as much as the CPU. So based on the survey, yes I like the AMD CPU. The survey should be changed to select which system is preferred.
 
The PCIe 4.0 x4 slot is actually not obstructed itself, but only the space 2-3 cm after it is obstructed, so you can actually use AOCs, but only very short ones that are barely longer than the slot itself.

But I have another issue with this board, it doesn't boot FreeBSD 14.2-RELEASE-p1. After the line:

Code:
"Setting up (amdgpu).. Please wait.."

The display just goes black, so I assume it's a driver issue. This thread says the AMD Raphael GPU which is part of the EPYC 4004 APU is not supported yet: https://forums.freebsd.org/threads/is-the-gpu-of-the-amd-ryzen-8000g-series-phoenix-supported.94620/
Also this article says it is not yet supported: https://xyinn.org/md/freebsd/amd_radeon_6900_xt
 
Is there a way to disable the CPU or disable probing the gpu? Don’t need it for a server.

The issue between E-2400 and the AMD counterparts is that there’s a $500 difference in price for equal performance. It seems that Dell and Supermicro (who lno longer sells barebone systems) have sucked up most of the available stock.
 
What? Please don't tell me that's true :(
It’s true. They only sell configured systems directly. While you can still theoretically buy barebones from distributors, they no longer use large distribution (tech data, ingram). So you have to order them drop ship, and supermicro favors their own direct system sales over distribution barebones. For example I tried to order a Power Supply, and they were “out of stock” but you could get one with a complete system. So they basically just didn’t want to sell it.

TBF I can’t speak to how they’re operating in Europe. This is a very new situation in the US. Their system prices aren’t bad, but it’s a couple $100 more for them to slap a CPU and some memory in
 
I bought the mainboard (MBD-H13SAE-MF), a smaller chassis (CSE-731I-404B) and the matching cooler (SNK-P0093AP4) from a European distributor. Only strange thing was that the initial lead time for the mainboard was 12-16 weeks while in the end it was 2 weeks after I paid it.

Maybe they just test if the customer has a real intention to buy because private customers mean more trouble because of the warranty.
 
Mainboards != barebones.

Right. I’m referring to the superserver systems. You CAN buy the chassis, IO Shield and MB separately. Just not from Supermicro directly. They’ve never really sold those directly, but they had large stocking distributors. Now they have a bunch of so-called distributors who just take orders and supermicro will drop ship it to you when they get around to it. But it’s a black hole, so you can’t track it or get info on when you might be getting it. And their systems orders get priority for the stock. They claim to be out of stock, then you get into a chat, and “oh, we cat get you one”. Find if you’re buying one, but not something a business can rely on.
 
I bought the mainboard (MBD-H13SAE-MF), a smaller chassis (CSE-731I-404B) and the matching cooler (SNK-P0093AP4) from a European distributor. Only strange thing was that the initial lead time for the mainboard was 12-16 weeks while in the end it was 2 weeks after I paid it.

Maybe they just test if the customer has a real intention to buy because private customers mean more trouble because of the warranty.
No, it’s because they have no idea. Supermicro is 100% unpredictable as to when they’ll ship. So they quote you worst case.

So did you get this to boot? Were you able to bypass the GPU?
 
FWIW I have a B650D4U-2L2T/BCM MB (Ryzen 7700 also Raphael) boots 14.1 just fine. No Desktop. I had it lying around and was concerned by vpx23’s post.

The E-2488 is $800 and the 7700 is lower power and < $300 in the US. Not really a close call.
 
So did you get this to boot? Were you able to bypass the GPU?
I think it would work with the drm-66-kmod instead of the drm-61-kmod, I have yet to test.

FWIW I have a B650D4U-2L2T/BCM MB (Ryzen 7700 also Raphael) boots 14.1 just fine. No Desktop.
I bet, because your board is actually using the AST2600 BMC VGA output first. I deactivated the VGA output because it switches to the HDMI output and causes a hang at the "EFI framebuffer information:" table while loading the kernel. I could activate VGA output again and connect using serial console but I want to use it as a workstation, not as a server.
 
I deactivated the VGA output because it switches to the HDMI output and causes a hang at the "EFI framebuffer information:" table while loading the kernel.
That's with H13SAE-MF? I'm using it with Ryzen 9900X (previously with 7950X3D) in that board, and didn't notice any issues with booting FreeBSD and monitor connected via HDMI, that's with current snapshots from half a year ago though.
 
That's with H13SAE-MF? I'm using it with Ryzen 9900X (previously with 7950X3D) in that board, and didn't notice any issues with booting FreeBSD and monitor connected via HDMI, that's with current snapshots from half a year ago though.
Could you please check the drm-kmod version then? I suppose it's 6.6 when it works for you.
 
Could you please check the drm-kmod version then? I suppose it's 6.6 when it works for you.
Sorry, I don't use drm/xorg on that system. I thought you meant it hangs for you when going from loader to kernel, i.e. earlier than loading the drm kmods.
 
Sorry, I don't use drm/xorg on that system. I thought you meant it hangs for you when going from loader to kernel, i.e. earlier than loading the drm kmods.
I just assumed you were using the amdgpu driver because of using HDMI, then you probably are using the vt(4) driver? Yes, you are right the framebuffer issue isn't related to the other amdgpu driver issue, these are two separate issues.

But I wonder why you don't have the framebuffer issue, is your AST2600 VGA output still enabled? But maybe it was already fixed in your CURRENT snapshot.

I'm actually using GhostBSD, that's why the drm kernel module with the amdgpu driver is already built-in.
 
I activated the VGA output again and tested with the latest CURRENT (FreeBSD-15.0-CURRENT-amd64-20250522-30fd79b0c0a3-277396-memstick.img), same result.

But I also connected a serial cable to my other PC and selected "Cons: Dual (Serial primary)". FreeBSD is booting totally fine to the "Welcome to FreeBSD!" message, just the monitor on HDMI hangs at the "EFI framebuffer information" table.

So it is possibly the hardware revision or the BIOS version, the former is 2.0, the latter 2.3
 
Back
Top