Yes. Those are the successors to the Opteron line.
i'd probably be correct in also stating that joe user probably won't need to grab these unless for very specific needs, yeah?
Yeah probably 
But that's pretty much the case with any high core-count CPU.
I will just point out now having thought this over more, AMD is offering more PCIE lanes from ONE cpu than TWO Skylake-SP CPU's can provide. That shit is FIRE for AMD.
Also.
Supermicro has a 1U chassis with:
10 Hot-swap U.2 x4 NVMe drives
up-to 64 combined cpu cores and 128 threads
up-to 4TB ram
two 16x expansion cards and one x8 low profile slot.
ho lee sheet. Dat density doe. If you need cpu compute, in 4u using four of those 1u servers, you could have 256 cores and 512 threads, 16TB of ram, forty NVMe drives, 8 gpus, and 4 100Gbps networking cards. It's insanity.
2 Likes
Looking forward to Threadripper. Lots of PCIe lanes means that the system should last much longer as there will be plenty of bandwidth for add-on cards later.
Do you, or anyone else, have a source confirming SEV on current Ryzen chips? I tried to find it but wasn't quite able to. There is a quote about not disabling any virtualization features on consumer chips, which sounds good, but I'd like to know for sure.
As far as I know it's in all Zen cores.
But I haven't personally tried activating / testing it
Maybe that's something to try out and setup. for another weekend
https://lwn.net/Articles/685215/
AMD Memory Encryption whitepaper:
(Currently getting bad gateway)
http://amd-dev.wpengine.netdna-cdn.com/wordpress/media/2013/12/AMD_Memory_Encryption_Whitepaper_v7-Public.pdf
AMD64 Architecture Programmer's Manual:
http://support.amd.com/TechDocs/24593.pdf
SME is section 7.10
SEV is section 15.34
Support for the extension exists within the kernel
1 Like
Ok so here is the relevant section from the documentation regarding SEV
This is the CPU register as read on my machine (1700x)
0x8000001f 0x00: eax=0x00000007 ebx=0x0000016f ecx=0x0000000f edx=0x00000000Converting register EAX to binary we get
EAX
0000 0000 0000 0000 0000 0000 0000 0111
^
#So since bit 0 and 1 is 1. That's a yes on SME and SEV
EBX
0000 0000 0000 0000 0000 0001 0110 1111
#Here's where it get's funky
#Bit 5 - 0 (10 1111 = Dec 47) I'm not sure how to interpret that. Clearly used to enable memory encryption.
#Bit 6 - 11 ( 001 01 = Dec 5) Reduction of physical address space in bits.
ECX
0000 0000 0000 0000 0000 0000 0000 1111
#To DEC = 15 Secure Virtual Machines supported? This might be offset by 1
2 Likes
Very much appreciated! This is really really good for home servers / lab.
Updated with more details
Read them, unfortunately I can't give any insights - my understanding seems to be at the exact same level as yours. I do have a feeling that the example in the pdf mentioning reduction from 48 bits of address space to 43 bits, using up bits 43 : 47, does coincide nicely with the numbers in your EBX register... but it is now to late at night to be trusting my gut.
I continued this as a separate thread here so others can hopefully chime in.
2 Likes
Interesting too read the secure VM so an admin can't peek. As with tech...physical access to a machine is game over.
Or is it
Seems like availability still sucks
I would imagine with more than 2 sockets the inter-die latency is real bad. bad enough we have inter-CCX latency.
can you even remember the last server chip that was game changing? opteron was not very relevant.
you probably wont be able to get as high clock speed and overclocks on a server board as with a tr4 board. made to be cheap and reliable not fancy with zero regards to overclocking.
1 Like
This is exciting for me. Very exciting. not only because of the disruptions that this platform will provide, but also the amount of cheap thread ripper/epyc could 5 years from now on the used market. Even in the upcoming years I can foresee tonnes of Intel machines sold off for cheap.
I’m amazed to see so much interconnectivity and power being packed into such a small system
What about adoption? This cpu clearly wins, but how many businesses will run out and buy a bunch of replacement machines with epyc/TR? anyone in the enterprise sectors know how this would play out?
With all the stuff that happened in Florida and Texas and such, I suspect the timing of these systems is good for the businesses who lost a bunch of hardware. It’s probably a good time to drive adoption of this. Hopefully nothing bad happened to the folks there, but if you’re a sysadmin or part of the higher ups, you know what must be done
Yeah jumping off of the non-numerical xeons up to the 51XX series on intel. Also, the 970FX was an insanely OP chip.
Oh I forgot the original post lol.
Well coming off of the K5 and K6, the K7 was used in a lot of servers.
This is mostly distributors. I’m working with supermicro to populate a datacenter with EPYC and it seems like they’re dragging their feet. I wonder if they didn’t get enough time to get their systems designed and tested.