Question Zen 5 Turin builders thread.

[Soulkeeper]

Thanks to both of you.
I was under the impression that R-DIMMs did not buffer the clock. Most documentation only seems to mentions command/address.
I'm kinda excited about ddr5, cudimms, and other memory tech.
Turin having 12 channels is just amazing.
Maybe i'll open a thread in the memory forum section to not derail this thread.

 

[Markfw]

Thanks to both of you.
I was under the impression that R-DIMMs did not buffer the clock. Most documentation only seems to mentions command/address.
I'm kinda excited about ddr5, cudimms, and other memory tech.
Turin having 12 channels is just amazing.
Maybe i'll open a thread in the memory forum section to not derail this thread.

What you should do is monitor this thread as my hardware comes in, and I compare it to Genoa (which ALSO has 12 channel memory) and I benchmark certain things (like avx-512) It is supposed to be up to 40% faster than Genoa !

 

[StefanR5R]

I was under the impression that R-DIMMs did not buffer the clock. Most documentation only seems to mentions command/address.

"Buffering" may not be the right term for what is done with the clock signal. "Repeating" and thereby "boosting" maybe? Anyway, the RCDs on DDR5 (and DDR4 and DDR3) R-/LR-DIMMs have clock input and output pins, and that's obviously where the "C" in "RCD" is coming from. :-) And functionally, it makes sense to have a PLL located directly on such highclocking register devices, so I doubt that any manufacturer makes do without it.

Back on topic, as you may have read, Turin is spec'ed to support DDR5-6000 in general and DDR5-6400 in select configurations (CPU and board combos?). But due to the slow news on Turin-ready boards, I as a member of the unwashed masses have no idea which specific configs those latter are, and if any of them will target the retail market in the first place. Might be for special contracts only, for all we know.

 

[Markfw]

OK< motherboard (dual) and 2nd Turin 64 core coming tomorrow ! Stay tuned !

 

[Markfw]

You don't *have* to populate the second socket of a 2P board. Nor do you have to put it into a case.

(Still, waiting for stock of 1P boards with the necessary BIOS compatibility might make more sense. Compatibility to your ES, that is.)

Well, I could not resist. So case, 2P motherboard and 2 64 core Turins getting built ! Its all here !

 

[igor_kavinski]

2 674 core Turins

Typo? I think it's 384 cores or 768 threads on that mobo? Using a 2000W PSU to power this beast?

 

[Markfw]

Typo? I think it's 384 cores or 768 threads on that mobo? Using a 2000W PSU to power this beast?

2 64 core. 600 watt is all it needs using a Titanium 1000 EVGA though. It will be up soon

 

[StefanR5R]

Phoronix' article on Turin + Supermicro H13SSL-N is out:
https://www.phoronix.com/review/supermicro-h13ssln-epyc-turin

The main takeaway for me: Yep, board revision 2.0 is required. At least the author says so. (Same as what Supermicro says.)

 

[Markfw]

Well, that's the board I had ordered. They could not give me an ETA, and the word back from supermicro directly said the same. So I cancelled the order and get the gigabyte dual that was guaranteed to be compatible. I just emailed wiredzone. we will see.

Oh and the gigabyte motherboard will not even post, but I had 6400 memory which is supposed to be compatible. So I ordered 4800. We will see.