H11SSL-i with Threadripper

bob_dvb

Active Member
Sep 7, 2018
171
74
28
Not quite London
www.orbit.me.uk
I currently have a Supermicro H11SSL-i with an Epyc 7401P doing virtualisation and I'd like more headroom for the Windows VM that's running on it. Today I saw LTT demonstrating using a Threadripper on an ASrock Epyc motherboard successfully and I wondered if I might be able to do the same with the H11SSL-i?

I've seen all the discussion about the reverse, using Epyc's on X399 boards, but little discussion about the reverse. I know that the ThreadRipper doesn't have SATA support, but neither does the Epyc and I have that covered with an HBA anyway. I also have ECC RAM so I don't think there's any issue there. Some comments about the X399 boards comment on the CPU microcode not being in the BIOS, which I suppose could be an issue?

However, I've seen this from CPU-Z which seems to indicate that it's been done with a 3990X: AMD Ryzen Threadripper 3990X @ 2199.49 MHz - CPU-Z VALIDATOR, but what confuses me is that I don't think a 3990X should be compatible with the H11SSL-i because the MoBo is TR4 not sTRX4 which is used in the 3990X. Is it possible that CPU-Z could be wrong/deceived for some reason?

I have my eye on a 2970WX which looks suitably spec'ed, but I don't think I am prepared to go to the cost of a 3960X or 3990X. I'd rather like some opinions before I go spending on a CPU that might not be compatible, and I don't really like the idea of buying a new motherboard as well.

What do you think?
 
Last edited:

RageBone

Active Member
Jul 11, 2017
570
141
43
Link to the vid and details please.

TR sockets only use the first and last mem channel on each side of the cpu.
As in A, D, E and H.

I expect that only those slots would work on an epyc board with a TR Cpu.
Microcodes are not what keep amd cpus from working unlike with intel.
 

mirrormax

Active Member
Apr 10, 2020
190
77
28
I currently have a Supermicro H11SSL-i with an Epyc 7401P doing virtualisation and I'd like more headroom for the Windows VM that's running on it. Today I saw LTT demonstrating using a Threadripper on an ASrock Epyc motherboard successfully and I wondered if I might be able to do the same with the H11SSL-i?

I've seen all the discussion about the reverse, using Epyc's on X399 boards, but little discussion about the reverse. I know that the ThreadRipper doesn't have SATA support, but neither does the Epyc and I have that covered with an HBA anyway. I also have ECC RAM so I don't think there's any issue there. Some comments about the X399 boards comment on the CPU microcode not being in the BIOS, which I suppose could be an issue?

However, I've seen this from CPU-Z which seems to indicate that it's been done with a 3990X: AMD Ryzen Threadripper 3990X @ 2199.49 MHz - CPU-Z VALIDATOR, but what confuses me is that I don't think a 3990X should be compatible with the H11SSL-i because the MoBo is TR4 not sTRX4 which is used in the 3990X. Is it possible that CPU-Z could be wrong/deceived for some reason?

I have my eye on a 2970WX which looks suitably spec'ed, but I don't think I am prepared to go to the cost of a 3960X or 3990X. I'd rather like some opinions before I go spending on a CPU that might not be compatible, and I don't really like the idea of buying a new motherboard as well.

What do you think?
Def don't go by cpuz, ive seen it identify epycs as threadrippers etc. Seems highly unlikely to work unless you are very capable in bios modding
 

bob_dvb

Active Member
Sep 7, 2018
171
74
28
Not quite London
www.orbit.me.uk
Thanks for the comments, I've taken a leap and bought a 2990WX, bold move but I figure if it doesn't work I'll resell it and cut my losses while learning something.

Bob

Yeah a link to that video would be nice. I can't find it at first glance.

Actually I might have messed up, they talked about TR but then used an Epyc.

Well, let's see what happens.
 

bob_dvb

Active Member
Sep 7, 2018
171
74
28
Not quite London
www.orbit.me.uk
I found a website where they had done a comparison of the pinout from the perspective of SP3 vs TR4 and it's largely exactly the same.

Some TR4 pins are RSVD which means reserved and IMHO they might not be an immediate blocker (there's just a risk that those pins are floating not grounded and that causes issues with the TR), then there's many pins that get an assignment change which is just where the naming of the memory interfaces and PCI bus changes, the function of those remains the same. Even the PCIe doesn't change at all.

On the first table I found there were four pins with different voltages which panicked me, but I checked the pinout table on wikichips for SP3 and TR4 and the voltages are the same there. I thought it was odd that a chip voltage and purpose changed, and verifying it elsewhere, it didn't.

There was talk of IDs changing, which could be an issue.
 

mirrormax

Active Member
Apr 10, 2020
190
77
28
there was an der8auer video where he attempted to get an epyc cpu working on in a TR4 mobo unsuccesfully. He managed to get Mobo to boot by blocking the ID pin but it obv didn't recognize the CPU without the epycs bioses. Its just not happening unless you are able to mod the bios with the right epyc profiles
 

RageBone

Active Member
Jul 11, 2017
570
141
43
Be warned, putting a TR4 CPU in a TRX40 socket may have killed the CPU.
It was damaged before but after i did that, it wouldn't even Power on the main VRs.
So going with a 2990 is a very very bold and potentially stupid move.
 
  • Like
Reactions: bob_dvb

bob_dvb

Active Member
Sep 7, 2018
171
74
28
Not quite London
www.orbit.me.uk
Just to document what I did somewhere public, rather than losing it, nothing special but for history it might save someone some time.

There's a Google spreadsheet here: http://www.hwbattle.com/bbs/board.php?bo_table=cpumbram&wr_id=169038&ckattempt=2
Then I added conditional highlighting and filters to make it more obvious what the differences are.
 

RageBone

Active Member
Jul 11, 2017
570
141
43
I don't know what happened with the table but it does not make sense to me.
Especially the pin info seems weird when i compare it to the tables here:

But maybe the weather is making me especially stupid today.
 

bob_dvb

Active Member
Sep 7, 2018
171
74
28
Not quite London
www.orbit.me.uk
I removed the filter that was applied, which might have been confusing.

It's basically two pairs of columns, the SP3 and the TR4, with the Pin designators the same and the Pin name applied. I've then selectively coloured the cells so that you can see what matches, what almost matches and what doesn't, then it also shows "RSVD" as a grey area.
 

RageBone

Active Member
Jul 11, 2017
570
141
43
I removed the filter that was applied, which might have been confusing.

It's basically two pairs of columns, the SP3 and the TR4, with the Pin designators the same and the Pin name applied. I've then selectively coloured the cells so that you can see what matches, what almost matches and what doesn't, then it also shows "RSVD" as a grey area.
I was looking for basically that, but i could only spot mostly memory channels that are not used, B,C F and G, the rest was either Reserved or Unknown.
Couldn't spot any Reset and logic signals or Pcie.
And the line numbers didn't really line up with the ones that i posted, kinda kinda not.

Now it looks completely different, so something probably went wrong on my end.

Edit:

Since TR uses the same 8 channel pinout of Epyc from A to H, of which only A, D, E and H are used, they changed the Letter scheme to ABCD.
THAT is the reason P0_ME_C0 now is P0_MC_C0.
Same with the others, so those aren't an issue.

Pins 1987 and 1988 might be 3.3VSB on all of them, will need to check as well as 243 and 2044.
If they really are 3.3, how isn't epyc instantly dying in those sockets?

Can't spot any possible explanation why my 1920 went completely dead after putting it in a TRX40.
Not that i can be sure that its the TRX40s fault, but it is sus.
 
Last edited: