EPYC almost runs on X399

[BluJay614]

Okay, going to start this off with a slight warning.

ATTENTION: It may be prudent to take the following information with a grain of salt, as it is coming from wccftech.

https://wccftech.com/amd-epyc-threadripper-mobo-compatible/

Quote

As reported earlier, AMD's EPYC and Ryzen Threadripper are not the same as they feature different dies and a different substrate. However, there were speculations if the server-based processor is just another repurposed Threadripper chip which ultimately cannot be used with the X399 platform.

Okay, the first part was actually proven by der8auer's video on YouTube, showing via x-ray the differences between Threadripper and EPYC.

Quote

YouTuber der8auer manages to take the EPYC server processor through its paces in order to find out if the chip is indeed just another rebranded Threadripper chip. To narrow the problems down as much as possible, the current processor that is mounted in the motherboard is a Threadripper 1950X, but without a CPU cooler. There is no storage or GPU attached, but RAM is present so that the motherboard is able to register.

This will help narrow down the errors when it comes down to attaching the EPYC CPU.

Okay, so we have a means of verifying that the mobo and the RAM are good and functional, though we are posing a bit of a risk at a $900 cpu......

Quote

After removing the 24-pin power connector and removing Threadripper, der8auer is ready to take on the challenge to mount the server processor in the ASUS ROG Zenith Extreme. Not surprisingly, there where a few complications that where encountered.

Yeah, saw that one coming.

Quote

Even though the chip can be effortlessly attached to the motherboard, standby power does not show up and this is due to the ID pin present on the chip. This helps the socket of the motherboard to determine which processor is being attempted to get mounted and if it is not the correct one then standby power will not show up.

Okay, a little info I didn't know, but hey, you learn something every day!

Quote

Now, der8auer was able to find a clever workaround and he did this by taping a portion of the CPU so that the ID pin does not come in contact with the socket pins.

Given that there is, what, over 4000 pins on this behemoth, I'm glad I'm not the one who had to find out which one it was.

Quote

After applying the isolation tape over the ID pin, the CPU finally powers on, which is a very good sign. Unfortunately, the debugging LED lights up, which according to the YouTuber is because the motherboard has not been tailored to accommodate the EPYC processor. It could also be that the memory controller is preventing the system from booting properly.

In short, the CPU can work on this board if not for the ID pin interfering, along with other issues. However, if an EPYC BIOS is made available for the motherboard, then system builders can easily get the server processor working on it. Unfortunately, this is something that we probably will not see ocme to fruition due to monetary interests.

Alright, so, again, While it is possible for EPYC to work on X399 boards (something I'm sure Linus would love to see happen), there seem to be issues standing in the way.

 

Either way, this does prove that there is some relation between EPYC and Threadripper besides the physical layout and the socket they go in.

 

Der8auer's English video this was bassed on:

https://www.youtube.com/watch?v=MWKnLvqjquw&app=desktop

same video in German:

https://www.youtube.com/watch?v+mAaUdcams4o&app=desktop

(I apologize if the links do not work)

 

[Guest]

im not exactly sure who would need an EPYC processor in the consumer marketplace excluding giant server farms. Even for productivity EPYC seems overkill

[FloRolf]

Is a 32core SMT Threadripper @4ghz too much to ask? 

[BluJay614]

3 minutes ago, NovaMan01 said:

im not exactly sure who would need an EPYC processor in the consumer marketplace excluding giant server farms. Even for productivity EPYC seems overkill

Agreed..... Though I guess you could say shits and giggles, or home server with something like the ASRock X399 Professional Gaming

[The Benjamins]

I thought it was common knowledge that TR4 is the same socket as SP3, and all that would be needed is a bios update to make Epyc work on x399.

 

Also if TR had its own socket it would need about 2000-2700 pins, AM4 is 1331 pins per die and SP3 is 1024 pins per die.

 

AMD HEDT could use a socket close to the size of x299

[BluJay614]

1 minute ago, FloRolf said:

Is a 32core SMT Threadripper @4ghz too much to ask? 

I don't know if I should hit funny or agree on that

[The Benjamins]

2 minutes ago, FloRolf said:

Is a 32core SMT Threadripper @4ghz too much to ask? 

it will not happen, AMD is not stupid enough to make a CPU where dies do not have direct access to ram. AMD already has latency issue no need to make a CPU with worse latency issues 

[Jito463]

6 minutes ago, The Benjamins said:

I thought it was common knowledge that TR4 is the same socket as SP3, and all that would be needed is a bios update to make Epyc work on x399.

I'm sure it was speculated on, but I don't think anyone has attempted to confirm it before now.  Though I still question whether a BIOS update actually would be enough.  Look at the work that went into getting an 8th Gen i3 running on a Z170 board, and then not everything functioned correctly.  This could be a similar situation.  Even if it's possible to get it to boot, not everything may work as expected.

[The Benjamins]

1 minute ago, Tedny said:

we still on new gen of Cpu's from AMD, everything will be better in the future 

yes, that's why Zen 2 HEDT will be 24c, because it will be 12c per die or 6c per CCX. but AMD would not shoot them selves in the foot by releasing a CPU that can't have a symmetric memory set up.

 

AMD already stated that their CCX layout will be symmetrical on all chips, why would they care about that if they would release a unsymmetrical design later.

[BluJay614]

1 minute ago, The Benjamins said:

yes, that's why Zen 2 HEDT will be 24c, because it will be 12c per die or 6c per CCX. but AMD would not shoot them selves in the foot by releasing a CPU that can't have a symmetric memory set up.

 

AMD already stated that their CCX layout will be symmetrical on all chips, why would they care about that if they would release a unsymmetrical design later.

ummm...... When did that info come out? Or is that just rumor being stated like fact?

[FloRolf]

11 minutes ago, BluJay614 said:

I don't know if I should hit funny or agree on that

[BluJay614]

1 minute ago, FloRolf said:

Because it will only let me do one

[Dragonheart]

15 minutes ago, BluJay614 said:

I don't know if I should hit funny or agree on that

 

3 minutes ago, FloRolf said:

You hit agree, I hit funny

[BluJay614]

Just now, Dragonheart said:

 

You hit agree, I hit funny

I LIKE IT!

[Dragonheart]

Just now, BluJay614 said:

I LIKE IT!

problem solved

[The Benjamins]

3 minutes ago, BluJay614 said:

ummm...... When did that info come out? Or is that just rumor being stated like fact?

for HEDT its rumor, but its confirmed Zen2 Server part will be 48c and be socket compatible witch means it will have 8 channel ram. which means it will be the same 4 die design. which means each CCX must be 6c or it would break socket compatibility.

 

and since AM4 and TR4 will be able to accept zen2 due to it being support by 2020. AM4 CPU's will be 12c, and TR4 will be 24c.

[DocSwag]

IIRC somebody theorized it could be because of the lack of ECC RAM. Last I heard Der8auer was gonna test using ECC RAM

[BluJay614]

1 minute ago, The Benjamins said:

for HEDT its rumor, but its confirmed Zen2 Server part will be 48c and be socket compatible witch means it will have 8 channel ram. which means it will be the same 4 die design. which means each CCX must be 6c or it would break socket compatibility.

 

and since AM4 and TR4 will be able to accept zen2 due to it being support by 2020. AM4 CPU's will be 12c, and TR4 will be 24c.

I did not know that was confirmed about EPYC 2, though the rest is still speculation. It's easily possible that everything else could maintain the same core counts, espechally since the leaks about the R5 2600

[The Benjamins]

2 minutes ago, BluJay614 said:

I did not know that was confirmed about EPYC 2, though the rest is still speculation. It's easily possible that everything else could maintain the same core counts, espechally since the leaks about the R5 2600

don't mix 2xxx with zen2. 2xxx is zen+, zen2 is the 7nm version due early 2019

[BluJay614]

7 minutes ago, The Benjamins said:

don't mix 2xxx with zen2. 2xxx is zen+, zen2 is the 7nm version due early 2019

That is true, and I do apologize for the mixup. However, as far as I was aware, the only things that where confirmed about Zen2 was that the process is shrinking, it is NOT a refresh, and that it is on schedule to be released in that time period. I hadn't heard any information beyond that.

Edit: My apologies, the design is finished for Zen 2. The other info that has been announced is that production will be split between Global Foundries and TSMC. http://fudzilla.com/news/processors/45453-amd-zen-2-design-is-finished https://pcgamesn.com/amd-7nm-tsmc-globalfoundries https://news.google.com/news/search/section/q/zen 2 amd/zen 2 amd?hl=en&gl=US&ned=us otherwise, I don't know of anything else that has been announced about it, even the sever bit that you where talking about.

[The Benjamins]

5 minutes ago, BluJay614 said:

That is true, and I do apologize for the mixup. However, as far as I was aware, the only things that where confirmed about Zen2 was that the process is shrinking, it is NOT a refresh, and that it is on schedule to be released in that time period. I hadn't heard any information beyond that.

Their was a slide during the Epyc launch that showed next gen would be 48c, and a representative during a video stated that the next gen Epyc CPU's would be socket compatible.

 

here is a leaked slide I found, trying to find the public slide that has 48c

[BluJay614]

3 minutes ago, The Benjamins said:

Their was a slide during the Epyc launch that showed next gen would be 48c, and a representative during a video stated that the next gen Epyc CPU's would be socket compatible.

 

here is a leaked slide I found, trying to find the public slide that has 48c

This is just going over their Enterprise CPU Products. The only socket they said would be supported was AM4. They said nothing about TR4/SP3 being supported for as long, so those could change. It's also possible information from this slide may have changed as it is almost a year old.

[The Benjamins]

3 minutes ago, BluJay614 said:

This is just going over their Enterprise CPU Products. The only socket they said would be supported was AM4. They said nothing about TR4/SP3 being supported for as long, so those could change. It's also possible information from this slide may have changed as it is almost a year old.

It was stated in a video, I would need to crawl through a lot to find it. 

[cj09beira]

2 hours ago, BluJay614 said:

This is just going over their Enterprise CPU Products. The only socket they said would be supported was AM4. They said nothing about TR4/SP3 being supported for as long, so those could change. It's also possible information from this slide may have changed as it is almost a year old.

ya they maintain support for a socket on a consumer sub 300 dollar motherboard, but wouldn't do the same for 3000 grand plus server, they need every thing they can get to convince more people to use epyc, support is a huge part of it, knowing you can get better perf with a simple cpu swap is a huge bonus

[Jito463]

2 hours ago, cj09beira said:

ya they maintain support for a socket on a consumer sub 300 dollar motherboard, but wouldn't do the same for 3000 grand plus server, they need every thing they can get to convince more people to use epyc, support is a huge part of it, knowing you can get better perf with a simple cpu swap is a huge bonus

Generally speaking, servers don't get upgraded in the same way that consumer systems do.  It's not common for server admins to swap out CPUs.