[News] An EPYC Threadripper: Der8auer proves EPYC CPU Might Work on X399 Motherboard

[StyM]

https://www.techpowerup.com/241072/a...99-motherboard

So, maybe it isn't really working - but at least the system boots up all the way to the BIOS memory checks, where it then stops emitting life signs. Der8auer went through a sort of blind process to discover that there is a particular ID pin on EPYC that when covered, allows the CPU to be booted up by a X399 motherboard (in this case, an ASUS X399 Zenith Extreme). ID pins are nothing new, and basically tell sockets whether or not they should be powering up a particular CPU.

So what exactly does this mean? Nothing much - only that the sockets and pinouts are the same. The approach towards detecting the ID pin was a crude, brute force one, appending a piece of electrical tape to different parts of the CPU, narrowing down the search for a single pin. When this particular pin was covered, standby power finally kept on, and the motherboard ran through some initial boot steps until stopping at the D0 memory boot code. Der8auer thinks that a "simple" BIOS switch on this TR4 motherboard to an EPYC motherboard's BIOS would suffice to get the EPYC CPU running on this Threadripper motherboard. Check out the full video after the break.

[zanzabar]

Do we know the performance hit from not having direct memory access?

[Buckeye]

It doesn't work, can't even fully boot like he says in the video.
It will need a complete BIOS update to make it actually boot into an OS.
As far as I know AMD has no plans for this. If you want EPYIC you will need a EPYIC board.

[madmaxx]

What a bulls#1t title! I expect nothing less than 3rd rate click bait shiite from techpowerup, they don't disappoint.

At least by linking their drivel here I don't have to visit there and reward them with page hits.

[Charles Wirth]

From the same guy who said X299 was a "disaster"

[AliG]

Do we know the performance hit from not having direct memory access?

Assuming that he would get booting to work, I'm not sure DMA would make a difference in most Windows applications. Typically DMA is only used for tasks that have an extremely tight deadline. It might just be ignorance, but offhand the only applications I can think of that truly require DMA are sensory data i/o on embedded platforms (such as radar in a control system).

[Buckeye]

From the same guy who said X299 was a "disaster"

HAHA True

[kromosto]

I thought he said that because of the vrm problems on mobos.

If you refer to the complaints about unnecessary complexity of the x299 platform it is said by every youtube reviewer I watched not only Der8auer.

[Charles Wirth]

X299 was not a disaster, the original VRM design was on point and proven to work as intended for everyone. Its when you overclock, disable protection, increase power limits to maximum, and then load all cores with brutal stress test it can get hot and cause the machine to "clip" when forced. Its like going out to your car and holding down the gas pedal to run at max RPM, The engine will fail at some point, its called common sense.

Let me point out that when you overclock fizzle bits get hot and we put cooling on them. Remember? Overclock a VRM and fail to put cooling is more fail.

He was also playing with pre release stuff, he should have consulted the manufacturers instead of a click bait article that reviewers took notice to thinking they found a flaw.

I am yet to see a car reviewer hold 10K RPM until the engine blows to call the car manufacturer "a disaster".

[Charles Wirth]

The same thing goes for overclocking the CPU with a stock heat sink, it only goes so far and when it hits its limit did you find a flaw in the overall design that constitutes it being a disaster? Disable protection, max limits, max vcore and it will fail quicker...

[Charles Wirth]

Title fixed, hope it is more accurate and still gives him some credit to his testing. AMD has a track record of soft disabling features, cores and more through BIOS so this testing might be useful to some one.

[zanzabar]

X299 was not a disaster, the original VRM design was on point and proven to work as intended for everyone. Its when you overclock, disable protection, increase power limits to maximum, and then load all cores with brutal stress test it can get hot and cause the machine to "clip" when forced.

With multi core enhancement by default and the boards being marketed to overclocking it was a reasonable claim. He also ran into that problem with the 10 core; with a many core it would have been worse. Ether way newer boards are having real PWM heatpipes again, so we all won.

[kromosto]

Actually I have no intention to watch that video again but if I am not wrong it was about bad mobo manufacturing not about x299 on VRM side. I remember that what he did was making a better contact with heatspreader and VRMs and show it becomes cooler. And this is about mobo manufacturers I guess.

And my opinions on x299 is it is not a disaster but not good either. While I agree with them about the unnecessary complexity part of the x299 I don't care about it because complexity effects the designer, manufacturer of the chip and motherboard manufacturer not me as a consumer. But raid key, forcing optane etc. are disgusting. So I was planing to upgrade to a ryzen 1800x but it was not good for me so I wanted to buy a 7900x but at that price tag with these nonsense with x299 no way sorry. I know as an individual customer I can not effect the market but I always fell that I need to show my complaint. Because of my VM usage actually I was planing to get an at least 8 core cpu but these circumstances forced me to buy a 8700k with z370.

[zanzabar]

that was the exact problem. There were fake VRM coolers on almost every board, and there were a couple boards that could only push about 250W but had whatever the company called turbo enhancement on by default. There was nothign wrong with the chipset or platform other than OEMs basically using the consumer grade PWM/cooling for the high end platform that can suck down 4-6x the power.

To a lesser extent some boards also had only one cpu 8 pin, but were trying to pull over 300W through it.

[[XC] Lead Head]

I really don't understand what X299 has to do with this at all?

He didn't prove that it won't work. He didn't prove that it worked either. He showed that the pinouts are the same, and that the CPU at least begins to start.

[StAndrew]

I think what he did is prove that the epic and threadripper cpus are the a same silicon?

[Buckeye]

I think what he did is prove that the epic and threadripper cpus are the a same silicon?

Yes I think he did and prolly is correct.

[StAndrew]

Yes I think he did and prolly is correct.

You'd think he'd done it already when he sanded down the silicon on a threadripper cpu. But I guess he's just trying to stick it to the man (amd). Not sure why amd is so insistent that epyc and threadripper are different...

With the extra mem channels and pci channels, I don't see a bios update fixing anything... Maybe threadripper on the epyc platform would be more realistic. But why?

[Buckeye]

TR prolly wont work on a EPYC system, for the same reasons.

I bet they are micro code locked out.

[Particle]

I think what he did is prove that the epic and threadripper cpus are the a same silicon?

Not exactly. We already know that Epyc, Threadripper, and Ryzen are built from the same die. AMD has talked about that across various avenues. The die is called Zeppelin and then each of those three products have their own codenames as well--Naples, Whitehaven, and Summit Ridge respectively.

The big reveal here is that the socket pinout appears to be the same or at least close enough to be potentially compatible. This would at first glance appear to contradict a direct statement from AMD claiming that TR4 and SP3 are physically the same but electrically incompatible. A person could probably weasel-argue that "electrically incompatible" might still apply, but a direct lie wouldn't be a first for this topic. AMD had also stated that the other two dies onboard a Threadripper weren't even etched, and we later discovered that that isn't true. What is much more likely is that Threadripper was created from the Epyc platform and was never really its own design.

[StAndrew]

Not exactly. We already know that Epyc, Threadripper, and Ryzen are built from the same die. AMD has talked about that across various avenues. The die is called Zeppelin and then each of those three products have their own codenames as well--Naples, Whitehaven, and Summit Ridge respectively.

The big reveal here is that the socket pinout appears to be the same or at least close enough to be potentially compatible. This would at first glance appear to contradict a direct statement from AMD claiming that TR4 and SP3 are physically the same but electrically incompatible. A person could probably weasel-argue that "electrically incompatible" might still apply, but a direct lie wouldn't be a first for this topic. AMD had also stated that the other two dies onboard a Threadripper weren't even etched, and we later discovered that that isn't true. What is much more likely is that Threadripper was created from the Epyc platform and was never really its own design.

I thought AMD denied they were the same and claimed the other two dies were completely blank silicon. The way I followed it, this prompted Der8auer to sand down the TR but AMD still insisted they were different at the PCB level; not electrically connected? I don't know if this proves of disproves that but Gamers Nexus had an interesting theory that AMD is trying to hide the fact that Threadripper may have a new, 16+ core CPU on the way?

[Sparky]

I figured they weren't blank but rather just failed/non-functional dies (and if so, possibly not electrically connected).

[zanzabar]

I thought AMD denied they were the same and claimed the other two dies were completely blank silicon. The way I followed it, this prompted Der8auer to sand down the TR but AMD still insisted they were different at the PCB level; not electrically connected? I don't know if this proves of disproves that but Gamers Nexus had an interesting theory that AMD is trying to hide the fact that Threadripper may have a new, 16+ core CPU on the way?

AMD said they were blank, but with some sanding we found out that they were "failed" cores, then they came clean that the cores are inactive but some of the die was needed for interconnects. It would also be fair to assume that they will have a higher core count with all of the cores enabled at some point on the TR platform.