[News] AMD Demo's Naples Server SoC, Launches Q2 2017

[StyM]

http://www.tomshardware.com/news/amd...ver,33819.html

AMD demoed its 64-core/128 thread dual socket Naples server platform against a comparable Intel server at its recent Tech Day but timed the announcement to coincide with the OCP (Open Compute Project) Summit this week. AMD presented a narrow workload that doesn't do much to reveal Naples' broader performance trend, although it does compare favorably against the Intel comparison platform. More importantly, AMD also revealed more details on the architecture and design of its forthcoming server SoCs.
In the past, AMD's desktop PC market penetration has lagged woefully behind Intel's, but a quick glance at its data center share highlights an even larger disparity between the two companies. Intel currently commands ~99% of data center server sockets, and its dominance in the high-margin segment has been a key to its success on many fronts. Intel is leveraging its commanding Xeon lead as a springboard to attack other lucrative segments, such as networking with Omni-Path/silicon photonics and memory with 3D XPoint, but it leverages locked-down proprietary interconnects that have raised the ire of the broader industry.
AMD, by contrast, leverages open protocols where it can and participates in developmental efforts on several new open interconnects, such as CCIX, Gen-Z, and OpenCAPI, so for the broader industry, a competitive AMD represents more than just a cost advantage and second source. Let's dive in.

AMD Prepares Zen-Based "Naples" Server SoC For Q2 Launch

“’Naples’ represents a completely new approach to supporting the massive processing requirements of the modern datacenter. This groundbreaking system-on-chip delivers the unique high-performance features required to address highly virtualized environments, massive data sets and new, emerging workloads.”

AMD NAPLES – COMPETITION RE-IGNITES AS ZEN HITS THE SERVER MARKET


my next cruncher...

[AliG]

The memory controller is most interesting to me. Twice the channels of the Intel counterpart.

I guess this will be another round of "get the chip that supports your applications." These tests demonstrate Naples will blow the pants off some tests, but like the Ryzen pre-launch benchmarks, I have a hunch that the specificity of these demos suggest it'll get spanked where AVX512 instructions can be used.

[StyM]

[vario]

Do these have 64 cores on one CPU package ?! Or is this a value considering threads as cores ?
16 memory channels.
I wonder if these use the same chips as consumer zens or are they 2x times bigger and mcm on a package.
Anyhow, seems kinda overpowered, like desktop zens somewhat underpowered.

[StyM]

32 cores 64 threads.

[Greg83]

this is definitely where my interest lays. can't wait to see what type 1P workstation motherboards come out for this.
guess in 1P there is a full 128 pci-e 3.0 lanes also. since not connecting to another processor.

[zanzabar]

The memory controller is most interesting to me. Twice the channels of the Intel counterpart.

I guess this will be another round of "get the chip that supports your applications." These tests demonstrate Naples will blow the pants off some tests, but like the Ryzen pre-launch benchmarks, I have a hunch that the specificity of these demos suggest it'll get spanked where AVX512 instructions can be used.

i would bet it is 4 modules, and MCM. that gives you a 16 cores per cpu, 2 cpus dies per socket, and 2 sockets.

[sergiu]

Impressive if turn out to be representative. Any pricing and single thread max speed until now?

[vario]

i would bet it is 4 modules, and MCM. that gives you a 16 cores per cpu, 2 cpus dies per socket, and 2 sockets.

Uhm, 128threads, 2 sockets, thats 64 cores total, meaning 32 cores per cpu. At least thats what the slides in the 1st post say.

[zanzabar]

Uhm, 128threads, 2 sockets, thats 64 cores total, meaning 32 cores per cpu. At least thats what the slides in the 1st post say.

it means 32 cores per socket. each module is 4 cores 8 threads. so 4 modules per die (or cpu) is 16 cores. then assuming it is a multi chip module like g34 you put 2 dies on one socket so 16x2 is 32. then 2 sockets is 64 cores.

[vario]

it means 32 cores per socket. each module is 4 cores 8 threads. so 4 modules per die (or cpu) is 16 cores. then assuming it is a multi chip module like g34 you put 2 dies on one socket so 16x2 is 32. then 2 sockets is 64 cores.

I only corrected because you wrote 16 cores per CPU, and a CPU is one physical package, no matter how many dies on it :-)
Anyhow, i really hope they will make workstation version, Stilt in his topic at anandtech forums made a case that ryzen is extremely efficient energy wise up to around 3.5ghz, its just later that power consumptiom sky rockets (and the reason why it wont clock really much after 4ghz).

[Zoran]

this is definitely where my interest lays. can't wait to see what type 1P workstation motherboards come out for this.
guess in 1P there is a full 128 pci-e 3.0 lanes also. since not connecting to another processor.

32 lanes per die, 64 per MCM package I believe.

[Greg83]

32 lanes per die, 64 per MCM package I believe.

its 128 actually , seen in the slides here. http://wccftech.com/amd-naples-serve...hread-preview/
1p http://cdn.wccftech.com/wp-content/u...tecture-IO.jpg
2p http://cdn.wccftech.com/wp-content/u..._2P-Server.jpg

[StyM]

[NEOAethyr]

32 cores, / 4 = this is a 4 module cpu, and 256bit mem controllers (quad chan).

This will have 4x the latency of the intel and will need 4x the mem speed of the intel to compete equally in some scenario's... (edit:, tha's wrong, it must be 4x latency, but only need 2x overall in ram speed increase...)
Complete disaster in my opinion, I wouldn't consider this a deal unless you plan on light load, like server's or something.
It certainly isn't a workstation.
It'll get you good cinebench scores though lol.

[zanzabar]

so up top i was wrong, the ryzen is an MCM so this is likely a quad chip module.

32 cores, / 4 = this is a 4 module cpu, and 256bit mem controllers (quad chan).

This will have 4x the latency of the intel and will need 4x the mem speed of the intel to compete equally in some scenario's... (edit:, tha's wrong, it must be 4x latency, but only need 2x overall in ram speed increase...)
Complete disaster in my opinion, I wouldn't consider this a deal unless you plan on light load, like server's or something.
It certainly isn't a workstation.
It'll get you good cinebench scores though lol.

amd uses infinity fabric, they do not need the same kind of bandwidth or speed that QPI has since they have many more direct connections and every core on a module can pull from the l3 of its module and the one next to it (this means on the same socket apparently.)

the big problem with the infinity fabric is that it runs at the same clock speed as the ram (the real clock not the ddr) so there are huge problems with slow ram on the consumer part, and since the server one wants FB it could be a huge problem.

[Lightman]

There will be no problem with memory or latency on SERVER parts, purely because applications this is aiming for will be aware of how to utilize Naples to its full potential. Server market is not like desktop and gaming, totally dominated by Intel. Server workloads run on many architectures like Power from IBM or ARM and all that thanks to workloads being compiled for specific platforms.

On desktop ZEN has high memory latency and it looks like AMD took the same mem controller from Bristol Ridge and slapped it on. There might be something in there we don't yet know about, which will explain why it is so slow. I have a gut feeling it has something to do with infinity fabric and direct link from CPU to upcoming AMD GPU ...

[AbortRetryFail?]

There will be no problem with memory or latency on SERVER parts, purely because applications this is aiming for will be aware of how to utilize Naples to its full potential. Server market is not like desktop and gaming, totally dominated by Intel. Server workloads run on many architectures like Power from IBM or ARM and all that thanks to workloads being compiled for specific platforms.

On desktop ZEN has high memory latency and it looks like AMD took the same mem controller from Bristol Ridge and slapped it on. There might be something in there we don't yet know about, which will explain why it is so slow. I have a gut feeling it has something to do with infinity fabric and direct link from CPU to upcoming AMD GPU ...

Yup.
Naples configured as the MI25 "Falcon Radeon Instinct Cluster" with 16X Vega.

Not likely to be playing much Crysis ...