AMD Epyc 7000 server CPUs released, take fight to Intel Xeons

Quote:

---

We look at in-depth architecture, models, market potential and lots more.

---

Read more.

It seems Dell has announced servers based on Epyc:

http://www.servers-maintenance.com/2...elling-server/
http://images.anandtech.com/doci/115...0_12.56.20.jpg

Apparently the picture is of some AMD test servers but with Dell and HP branding.

Another poster on OcUK saw this:

https://www.glassdoor.co.uk/job-list...yRedirect=true

Mentions Amazon,so they might be a customer??

That picture is taken from the launch event - I'm sitting in the same building right now. HP has also announced upcoming servers.

Some initial power consumption figures:

https://www.servethehome.com/amd-epy...-observations/

Edit!!

LOL Intel:

http://www.barrons.com/articles/amd-...-it-1497997334

Quote:

---

We take all competitors seriously, and while AMD is trying to re-enter the server market segment, Intel continues to deliver 20+ years of uninterrupted data center innovations while maintaining broad ecosystem investments. Our Xeon CPU architecture is proven and battle tested, delivering outstanding performance on a wide range of workloads and specifically designed to maximize data center performance, capabilities, reliability, and manageability. With our next-generation Xeon Scalable processors, we expect to continue offering the highest core and system performance versus AMD. AMD’s approach of stitching together 4 desktop die in a processor is expected to lead to inconsistent performance and other deployment complexities in the data center.

---

Quote:

---

Originally Posted by CAT-THE-FIFTH

LOL Intel:

http://www.barrons.com/articles/amd-...-it-1497997334

---

Comical reply from Intel considering their xeons are basically x desktop dies stuck together....
Really does scream of a company thats worried because AMD has a good chip again, bit like when (s939) athlon x2/opterons came about and were considerably better than intel options.

Quote:

---

Originally Posted by LSG501

Comical reply from Intel considering their xeons are basically x desktop dies stuck together....
Really does scream of a company thats worried because AMD has a good chip again, bit like when (s939) athlon x2/opterons came about and were considerably better than intel options.

---

Its called Epyc desperation.

Quote:

---

Originally Posted by CAT-THE-FIFTH

LOL Intel:

http://www.barrons.com/articles/amd-...-it-1497997334

---

Wait, wasn't the Zen architecture designed primarily for this market?

I love it, Intel are losing their minds and doing everything they can to downplay what AMD has done. I bet Lisa and her teams are sat back in their chairs, arms on back of heads smiling...for five seconds then she'll jump up and spearhead the next disruptive technology.

Also I really hope that software makers really work with AMD to eke performance out (especially VMWare). The business I work at is growing and EPYC nodes look like the right way to go!!!

Quote:

---

Originally Posted by LSG501

Comical reply from Intel considering their xeons are basically x desktop dies stuck together....

---

Erm ... not really. High core count Xeons are distinct monolithic silicon. I don't think Intel offer any MCM Xeons at the minute - iirc the last serious Intel investment in MCM was during the Core 2 days, when they were doing quad core by stick two dual-core dies on a package, while AMD were rushing to produce a monolithic quad core. Anyone remember how that turned out for AMD? ;)

AMD have been producing MCM server chips for ages - pretty much all of their high core count Opterons were MCMs, right through the Phenom and FX desktop lines. That would've been fine if their cores were performance and efficiency equivalent to Intels, but they weren't - Phenom was inefficient and couldn't clock highly enough, Phenom II was a big improvement but came out after Bloomfield, and FX ... well, 'nuff sed ;)

The Zen core does have worse IPC than Skylake/Kaby Lake, but it's been tuned for server workloads/constraints - you can tell that by the fact that they can pack in so many more cores than Intel, at similar clocks, and with similar thermal design (although interestingly if you look up AMD's "equivalent" processors on ark Intel are claiming lower TDPs than AMD...). If I was Intel I'd be pushing how AMD's desktop chips are actually just massively overclocked server chips, rather than vice-versa...

Quote:

---

Originally Posted by scaryjim

If I was Intel I'd be pushing how AMD's desktop chips are actually just massively overclocked server chips, rather than vice-versa...

---

Hm, and risk having their Skylake-X HEDT chips accused of being the same!

Quote:

---

Originally Posted by scaryjim

Erm ... not really. High core count Xeons are distinct monolithic silicon. I don't think Intel offer any MCM Xeons at the minute - iirc the last serious Intel investment in MCM was during the Core 2 days, when they were doing quad core by stick two dual-core dies on a package, while AMD were rushing to produce a monolithic quad core. Anyone remember how that turned out for AMD? ;)

---

To be fair, maybe die wasn't the correct term... should have probably used the term core.

Quote:

---

Originally Posted by kompukare

Hm, and risk having their Skylake-X HEDT chips accused of being the same!

---

Well, they are! Only I believe they fuse off ECC memory support, and probably one or two other interesting features, just for "product differentiation"... ;)

Quote:

---

Originally Posted by scaryjim

Well, they are! Only I believe they fuse off ECC memory support, and probably one or two other interesting features, just for "product differentiation"... ;)

---

I believe they have a list of things they might remove, put them on a wall, and then throw darts!

I don't like the branding, IMO 7281-7301-7351 should be 7301-7351-7371 or similar. Then you can see core count from the second digit. Shouldn't be a major issue for the server market, of course.

The 7501 looks very interesting. If the 180W chips can downclock to 165W, and the 7501 takes 155W when running 2.4GHz memory, then how little power do they need to run 32 cores? Scaling off the % reduction of the 180W chips gives ~140W, which is incredible

Quote:

---

Originally Posted by Xlucine

... If the 180W chips can downclock to 165W, and the 7501 takes 155W when running 2.4GHz memory, then how little power do they need to run 32 cores? ...

---

Hardly any - CAT posted some charts of voltage:clock when Ryzen first came out and there's a linear voltage drop right down to < 2GHz. The scaleable TDP range on the 155W OPN parts is as low as 140W (second slide on page 3), which works out at 35W per die, and actual draw could be lower than that. I suspect with manual tweaking you could probably peg 32 cores as low as 100W (run 2133MHz memory, manually tweak voltages, clocks at around 1.8GHz?)...

EDIT:

You know, it's only just struck me what AMD has offered with the 7501 ... 64 threads, 8-channel memory supporting up to 2TB of DDR4, 128 lanes pf PCIe I/O, all in as low as a 140W thermal envelope. That's ... gobsmacking....

Quote:

---

Originally Posted by scaryjim

You know, it's only just struck me what AMD has offered with the 7501 ... 64 threads, 8-channel memory supporting up to 2TB of DDR4, 128 lanes pf PCIe I/O, all in as low as a 140W thermal envelope. That's ... gobsmacking....

---

And that's why we might see a second panic from Intel while they quickly put back in all the features they were going to remove in the Skylake based Xeons

Interesting new video about SEV:
'EPYC Enables New VM Boot Security'

https://www.youtube.com/watch?v=P05d6sp0m04