• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

Details of new 64c128t Epyc (Rome) emerge

Soldato
Joined
14 Apr 2014
Posts
2,585
Location
East Sussex
Crikey!

The impact of leapfrogging Intel and using 7nm is several-fold. First, Rome will have up to 64 cores and 128 threads in a single socket. The new CPUs will be socket compatible with the current SP3 socket motherboards with a small caveat. At STH, we expect Rome to adopt PCIe Gen 4 so motherboards will have to support the higher signaling rates to achieve PCIe Gen 4. We also expect the next generation to have greatly improved Infinity Fabric, an area that the first generation product has room to improve upon.

The other key disclosure is that AMD already has silicon in their labs with the next generation AMD EPYC Rome CPUs in their labs. They will be sampling to partners in the second half of 2018 and will launch in 2019.

Source: https://www.servethehome.com/amd-epyc-rome-details-trickle-out-64-cores-128-threads-per-socket/
 
Man of Honour
Joined
13 Oct 2006
Posts
90,805
Not sure about the leapfrogging bit - people keep forgetting that Intel 10nm is more an approximate of 7nm at GF and TSMC (and part of why they have having a lot more trouble with it than "10nm" processes at those two).

Sounds like a pretty crazy CPU for those that need a lot of threading - gonna need some big advances in bus bandwidth and other IO to keep feeding that beast effectively.
 
Soldato
Joined
26 Apr 2013
Posts
4,826
Location
Plymouth
I bet I could open 10 tabs in Chrome with that! In all seriousness, when specs come out for this, I'd love to see how they avoid bottlenecks with this.
 
Caporegime
Joined
18 Oct 2002
Posts
33,188
Not sure about the leapfrogging bit - people keep forgetting that Intel 10nm is more an approximate of 7nm at GF and TSMC (and part of why they have having a lot more trouble with it than "10nm" processes at those two).

Sounds like a pretty crazy CPU for those that need a lot of threading - gonna need some big advances in bus bandwidth and other IO to keep feeding that beast effectively.


I mean, if Epyc is currently competitive on a worse process node, then I'd expect it to leap from Intel just moving to a comparable node. But the issue is Intel, I'll be straight up shocked if they get 500mm^2 and larger cores on 10nm next year, while AMD to have EPYC out only needs 200mm^2 cores yielding. AMD is going to have 7nm EPYC next year, Intel I'll be surprised if they have the small die server chip in 2019 now, the large part I think will be 2020 H2 at this rate. They just canned full production on 10nm because they couldn't get what I assume to be around a 50-60mm^2 dual core part yielding with the GPU working at all. They pushed production of again what I would presume to be something in the range of 100-150mm^2 desktop parts back to likely 2019 H2 because that isn't close to ready to go.

As such for 2019 I see it that AMD will be selling 64 core EPYC on 7nm while Intel is still selling 28 core on 14nm.
 
Soldato
Joined
30 Jul 2004
Posts
2,836
Location
Auckland
Lol, I suspect it has far more to do with Intels legal team looking real close at AMD patents and IP rights.
The Ryzen core architecture is probably the biggest revolution we have seen since dual core and 64bit chips. We are only just starting to see how explosive it is. If AMD can go to 6 or 8 core ccx on 7nm they can massively increase core count again.
If we now get some really smart software that loves cores and can split threads... Or at least a movement to massively threaded applications
 
Soldato
Joined
30 Jul 2004
Posts
2,836
Location
Auckland
It dropping into the existing socket gives them an amazing sales opportunity right now. I was just scanning the article as I had not heard any definite confirmation of a doubling core count.

They have actually had confirmation that we are going to see a 48 Core variant before we see a 64 Core chip.

That points to initially a 6 core CCX possibly due to using some of the extra silicon for something else or possibly because for a server chip they cannot guarantee a chip that size with good enough yields out of the gate.

Intel are going to have a hard time with Epyc having already proven itself and with that level of incremental increases incoming.
 
Permabanned
Joined
2 Sep 2017
Posts
10,490
Soldato
Joined
6 Aug 2009
Posts
7,070
I haven't got much interest in the server world however I do love to see progress at the high end knowing it will filter down eventually. Innovation is one of the reasons I like AMD as a company. To be fair they have to be to try to compete with Intel.

I'd really love to see what Intel, with all their resources, could do if they really tried. Sadly the very thing that provides large corporates with funding, shareholders, seems to hold them back when they focus on profits over innovation. That may just be the type of people in charge rather than being a large corporate issue.

The great thing is I can see the next 5 years as being very interesting with this renewed competition.
 
Soldato
Joined
1 Dec 2015
Posts
18,514
@EsaT not sure you seen the post above and others. one of them has nice mention of Infinity Fabric being greatly effected by PCIe 4.0 . Guessing in theory, those that are keeping their ryzen 1000/2000 will see a boost going to b550/x570 boards hypothetically :D
 
Soldato
Joined
26 Sep 2010
Posts
7,146
Location
Stoke-on-Trent
Wffctech (yeah, which is why I'm not linking) reporting:

"162 PCIe Gen 4 Lanes"

on EPYC Rome

Doable, read BongoHunter's linked article on page 2, it covers the potential of diverting some of the socket-to-socket lanes to bolster connectivity.

Essentially, use 3 16x links instead of 4 for socket-socket IF and redirect those 32 outwards for 160 lanes. Each CPU has an additional 129th lane dedicated to low-bandwidth devices rather than eating into the main pool, which gives you 162.

Edit: and yes, WCCFTech are just paraphrasing the same ServeTheHome article BongoHunter posted.
 
Back
Top Bottom