Der8auer Gets Epyc Working on X399

R

rgMekanic

[H]ard|News
Joined
May 13, 2013
Messages
6,943
YouTuber and extreme overclocker De8auer has gotten an Epyc CPU working on an X399 Threadripper motherboard.. sort of. In his testing Der8auer has found that there is a single ID pin on Epyc that when covered, will allow you to partially boot an Epyc CPU. The BIOS stopped loading at memory checks, so a BIOS that is compatible with Epyc is required, but this proves that the sockets and pinouts are the same.

Quite amazing stuff. I have my doubts if a partner will ever release a BIOS that allows Epyc in an X399 motherboard, but it opens up possibilities for sure.
 
What would be the purpose of this? To save money on a motherboard?
 
32 cores would require an octachannel memory interface, so i am not sure, it would work on x399. :/
 
bobzdar said:
32 cores and overclocking on an enthusiast platform maybe?
Click to expand...

I'd much rather have an overclockable Epyic board with all the extra RAM and PCIe slots than to put a chip on a platform that gimps it. The net effect is like Intel's Kaby-X where the IGPU free Kaby chip was on an X-board, but lacked the quad channel memory and PCIe lanes, an altogether waste.

Now as a simple experiment, I applaud the Der8auer's efforts.
 
Someone could always just hack the BIOS to support it. What ever happened to the people that used to hack motherboard BIOS's left and right?

giphy.gif

 
yumnmycandy said:
32 cores would require an octachannel memory interface, so i am not sure, it would work on x399. :/
Click to expand...

I'm not sure it would... As is each Sub Cpu/Ryzen is wired to it's own block of ram and handles the IO for that block -- if code/data is in the other block the system requests it through the other memory controller at a penalty to latency. Or at least this is the understanding I gathered reading the release info back when tr dropped.

In theory all the memory access could be through just 1/2 banks instead of 4, with a latency penalty.

Existing TR owners, do you need to have ram in both banks for the system to work? Does the ram need to be symmetrical? I am very curious to know.
 
Title is misleading, it's not exactly working, and he's in the process of grabbing some ECC memory to maybe get it actually working ...

"der8auer1 day ago
mh interesting idea. Didn't think of that :D Thanks a lot! Will grab some ECC sticks"
 
C'mon you guys know you want to see Asrock make an itx epyc board like they did for x99 (in my signature).

My board doesn't have all the memory it is capable of addressing? Who cares only really matters in synthetic benchmarks.

Hell I'd even be all about threadripper in an NCase.
 
Yeah, overall, I would not suggest that you would want to be doubling up to four cores on a X399 board and OCing. That package at 4GHz pulls 350 watts under load. I doubt many HEDT boards are suited for overclocking EPYCs, if they can be. But still very cool to see, however not likely to be practical at all. Fun is fun though.
 
tunatime said:
Unless they have much better binned cores that might run better at lower vcore?
Click to expand...
Considering that AMD has already spoken to Threadripper being cheery picked cores, I highly doubt that is going to come into play. Also, TR is already 350w package power. That bumps up against 500w load on the PSU in experience. Chip binning is nice, but it is not a magic wand for CPU per usage.
 
erek said:
Title is misleading, it's not exactly working, and he's in the process of grabbing some ECC memory to maybe get it actually working ...

"der8auer1 day ago
mh interesting idea. Didn't think of that :D Thanks a lot! Will grab some ECC sticks"
Click to expand...

How's he going to get around the fact that epyc uses an 8 channel memory controller? AMD stated from day one that Epyc and TR share the same socket and pin outs but they are electrically different. I haven't seen anything that suggests otherwise.

darrpara said:
C'mon you guys know you want to see Asrock make an itx epyc board like they did for x99 (in my signature).

My board doesn't have all the memory it is capable of addressing? Who cares only really matters in synthetic benchmarks.

Hell I'd even be all about threadripper in an NCase.
Click to expand...

I suppose it would be interesting as a one-off but its not anything remotely attractive for 99.9% of us.
 
  • Like
Reactions: erek
like this
thesmokingman said:
How's he going to get around the fact that epyc uses an 8 channel memory controller? AMD stated from day one that Epyc and TR share the same socket and pin outs but they are electrically different. I haven't seen anything that suggests otherwise.



I suppose it would be interesting as a one-off but its not anything remotely attractive for 99.9% of us.
Click to expand...

Yah, just glad someone else isn't calling this "working" cause ... it's not
 
So many rose-colored glasses, seeing what you want to see. So TR and Epyc share a socket, minus a blocker pin. This isn't really news LGA771 and LGA775 were the same socket with a few pins shifted, precisely to keep things isolated. Point of putting Epyc on an X399 board? Nearly zero, aside from e-peen. "Proof" of AMD planning/releasing anything? Not even slightly. All that's been "proven" is that it's possible, NOT that AMD is planning to do it, or even seriously considering it. Just look at Intel's Xeon and HEDT lines. You can buy a nice big fat E5 with way more cores than any of the HEDT offerings, I know erek has, but you get a lower clock speed in return. There are limits that must be kept. By buying HEDT you get higher clocks, unlocked multipliers, and other desktop-ish features. By buying Xeon/Epyc, you get better thermal management, lower clocks, more cores/threads.

But back to the OP, it doesn't sound like much of a success at all. I'd bet cash it's hanging because the memory controllers aren't being initialized correctly.

As for why doesn't someone just "hack" an Epyc firmware for a consumer board? Because firmwares are signed and/or encrypted these days, and complexity has gone up several orders of magnitude. In the old days you could kitbash a 440BX firmware out of modules from other 440BX boards, and the only "hacking" was in the OEM modules. These days firmwares are far more specific and complex.
 
  • Like
Reactions: erek
like this
maxius said:
im calling it enhanced 12nm 32c 128t quad channel threadripper incoming
Click to expand...

I wonder if the memory channel setup would be 2+0+2+0 or if it would be 1+1+1+1 and what the benefits / downsides would be...
 
I dont get it... Of course AMD will eventually release a 32 core threadripper.
But at those power levels it is going to take a couple process changes before it happens.
I predict it will be after zen3 is out.
 
You don't actually have to populate all of those memory channels, do you?

I understand why you'd want to, and why they're there; but they're not necessary are they?
 
ecmaster76 said:
He believes AMD might be planning a 32 core TR product. This almost proves it
Click to expand...

Not really. I've said it before. People didn't think it was possible because of how Epyc is configured for RAM and PCIe slots. I don't think that's the case. I have believed from the start that a 32 core processor was possible on this platform, albeit with some changes vs. the Epyc version of the chip. I also said that it wasn't likely until at least Zen+, but probably not until Zen 2 or whatever it's called shows up.
 
EPYC 32 core processors have the same TDP paper spec as a 1950X, so theoretically, a X399 board should be able to at least run from a power perspective.

But of course if the memory config, etc differs there's nothing much to do about it.

It does, however, suggest that the TR platform could support 32core procs in the future. Doubling the die package, or perhaps routing traces to the disabled dummy cores.

The issue with this, is the usefulness of 2.2base/3.2boost 32 cores in a HEDT platform. At least TR has a super aggressive boost that allows it to recoup some ground in thinly threaded applications.
I'm sure there's a market for this in HEDT, question is if it's large enough for AMD to care. Zen+/2 would likely allow these to run at a more useful range.
 
Yea you might be able to get it to boot on an consumer board, but the chances of a consumer board delivering the power needs without substantially limiting it's lifecycle.... that' I don't see. Though it reminds me of doing the old pencil traces in order to overclock early pentium's or was that late 486's I can't remember.
 
Grimlaking said:
Yea you might be able to get it to boot on an consumer board, but the chances of a consumer board delivering the power needs without substantially limiting it's lifecycle.... that' I don't see. Though it reminds me of doing the old pencil traces in order to overclock early pentium's or was that late 486's I can't remember.
Click to expand...

If the TDP is truly the same on Epyc as the 1950X, then it shouldn't be a problem.
 
Grimlaking said:
Yea you might be able to get it to boot on an consumer board, but the chances of a consumer board delivering the power needs without substantially limiting it's lifecycle.... that' I don't see. Though it reminds me of doing the old pencil traces in order to overclock early pentium's or was that late 486's I can't remember.
Click to expand...

That was AMD. Worked great too.
 
It is definitely interesting to get the epyc to boot, but what if there are multiple "Id pins"?
Trial and error with electrical tape is quite a shot in the dark.
Also the dots on the epyc and threadripper CPU pcbs are inverted. These may be chip orientation dots. What if the pin layout on SP3 is a mirrored version of TR4? Sure, the sockets look the same, but what if the AMD design spec was to mirror the pinouts of the cpus for SP3 and TR4 so the motherboard manufacturers follow two different guidelines for Threadripper and EPYC?
 
Title: Epyc working on X399
Story: But not really

Do clickbaits like this really resonate with the target audience on [H]? Honest question.
 
ecmaster76 said:
He believes AMD might be planning a 32 core TR product. This almost proves it
Click to expand...

How will it be cooled?. AFAIK it would need 480mm Radiators just to reach ~3.7Ghz. This is a curve ball that AMD can only launch on a more efficient process node.
 
I know this is a [H]ardcore overclocking site but there exists a market for stock speed 32core workstation. :D

It doesn't have to be a gaming oriented product.

So this is possible though perhaps not likely on the current Zen cores. Maybe Zen+ or Zen2. Like I said I think he has proven AMD is keeping their options open not necessarily a relase this gen
 
ecmaster76 said:
I know this is a [H]ardcore overclocking site but there exists a market for stock speed 32core workstation. :D

It doesn't have to be a gaming oriented product.
Click to expand...

It doesn't have to be anything oriented...I thought this was an enthusiast site, why are there so many people crying about market segments and (unbelievably) too many cores?
 
Ehren8879 said:
I doubt this prediction.

I do not believe AMD will rely on the Infinity Fabric in place of direct to core memory channels.
Click to expand...

AFAIK, all zen products work fine with one stick of ram installed (ie single channel mode) - when I say fine, I mean they work without issue, but not optimal. However, I can see cases where more cores would be beneficial but 8 ram channels vs. 4 (or even 2) not being worth anything at all. It all depends if the application is memory latency sensitive. Some are, some aren't. Point is, Epyc on TR should be the same as only running 4 sticks of ram on Epyc.
 
darrpara said:
C'mon you guys know you want to see Asrock make an itx epyc board like they did for x99 (in my signature).

My board doesn't have all the memory it is capable of addressing? Who cares only really matters in synthetic benchmarks.

Hell I'd even be all about threadripper in an NCase.
Click to expand...

Apart from just being a cool feat, what's the point? Your ITX board pretty much lacks all the features folks desire from the enthusiast platform apart from the extra cores you may get due to having 2011-v3 socket. Cool? Yes! Practical or useful? Heck no... (I love mini boards, even have some Pico ones from VIA but they are severely gimped)
 
What I would like to see attempted is the unlocking of the other two dies on the Threadripper chip by matching the components configurations of the EPYC on the outside in terms of the resistors, capacitors, etc. In his first comparison video between TR and EPYC he made note of the components configurations differences and it seems the TR has the components solder pads ready to accept them.

Currently this EPYC processor is not really working in that ASUS Zenith board.
 
darrpara said:
C'mon you guys know you want to see Asrock make an itx epyc board like they did for x99 (in my signature).

My board doesn't have all the memory it is capable of addressing? Who cares only really matters in synthetic benchmarks.

Hell I'd even be all about threadripper in an NCase.
Click to expand...

Not really. You need at least four slots for quad-channel memory support. That's one of the big draws of HEDT motherboards. HEDT boards and the processors that drive them are also about PCIe lanes which go unused by stuffing one of those chips into a mini-ITX motherboard. As zkostik said above, it's virtually pointless for most people as all the enthusiast oriented features that draw people to HEDT motherboards and processors are missing from any mini-ITX motherboard. There are some niche type reasons to use something like that when all you care about are cores, but very few people are in that boat. Mini-ITX motherboards aren't big sellers anyway. ASUS barely breaks even on the R&D to design them.
 
You must log in or register to reply here.
Back
Top