It's over Intel is finished!

>>59541278
the real test is whether or not Azure or AWS end up running

>>59541372
The 32 core Naples are targeted at cloud, but I don't see why a 16 core wouldn't work.

>>59541278
Xeon-D has built-in 10GbE MACs, which Zeppelin by all appearances lacks.

I -really- want a 16c/4xDDR4 Zen workstation though, even if I think the platform isn't a perfect substitution for Xeon-D.

>>59541534
These things will have boatloads of PCIe lanes, so I don't think including a 10GbE will be all that difficult.

https://www.servethehome.com/amd-naples-competition-re-ignites-zen-hits-server-market/

SFP+ 10GbE right there.

>8 core CPU barely competes with 4 core intel
>16 core CPU is expected to compete with 6, 8, or 12 core Xeons
sure thing buddy

A high TDP 16 or 32 core chip will hurt intel. They have been making workstations miserable for too long. Their xeons are all aimed at servers where efficiency is critical they offer nothing worthwhile for single socket systems. They have some 6 or 8 core 2P xeon that usually costs more the the unlocked i7 version typically leaving twin xeon builds barely faster than an overclocked i7 for 4 times the price.

File: 1488594364886.jpg (64KB, 855x566px) Image search: [Google]

64KB, 855x566px

>>59541612
>>59541278

Intel was already finished.

>>59541597
more importantly, all those PCIe lanes allow for better infiniband support

which means they're going to crush Intel in the HPC market

>>59541278
So I can assume somewhere around a 3.4GHz all core turbo?

Fucking awesome, that's practically double the 1700 + 100MHz

File: 1488594492280.jpg (70KB, 858x586px) Image search: [Google]

70KB, 858x586px

>>59541278

If I were building a high-throughput intranet server, I'd build a Ryzen server in a heartbeat.

File: Screenshot from 23-03-2017 17:06:49.png (275KB, 645x606px) Image search: [Google]

275KB, 645x606px

>>59541278
>not getting the much more efficient Naples one with twice the cores and lower TDP


Ha, those 'prosumer' 16 cores are the shittest voltage bins AMD could find, hence the high TDP.

>>59541733
>that barren board

integrated southbridge baby

>>59541733
when will the VGA meme die?

Seriously, somebody needs to come up with a VGA over USB or something

>>59541278
You'd think that the enterprise MCM dies would be binned for the absolute best perf/W samples that they can get out of the fab, so I'm wondering how much of the apparent extra power draw vs. 1700/1700X/1800X comes from where.

> AES-128 encrypted DDR4 controllers?
> inter-die IF links?
> inter-CCX IF links clocked up to full DDR rate?

>>59541772
These are for consumers, they won't do much voltage binning on these, and as anon already said this is a high performance 1P platform.

The Opteron ones will be though, see >>59541733

Skylake-EX 12 corelets already dead before launch.

>>59541764
It's not a meme for server or industrial applications.
Those things are built to last FOREVER and be compatible with EVERYTHING. There's a very, very good reason they still use it.
They don't need to push ultra 4k resolutions of 144hz refresh rates.

>>59541764
Never, because they're running off the motherboard GPU and those are essential for debugging.

Might as well ask when are we gonna get rid of PCI-X and Serial

>>59541772
Given the high TDP for those modest clocks its likely that these chips have different turbo scheme than consumer Ryzen CPUs. They'll probably sustain higher clocks on more cores. Give it more performance on a wider range of professional software vs intel's Xeons.

AMD doesn't have a unified GPU server, Intel+Nvidia are still going to own the datacenter because of this.

>>59541794
>>59541733
There are more 1S/1P servers floating around out there than you might imagine, so I think it's premature to assume that the 16c/32t Zen part will be prosumer only.

>>59541838
PCI-X has been dead for a decade

Serial is also dead, most of the new cisco routers have USB

>>59541847
There's a enterprise AND prosumer one, the prosumer one won't care about perf/watt all that much.

This is basically like Intel giving a glimpse of high end xeons with a 10 core Broadwell-E.

>>59541278
>Inb4 it flops on launch like the R7 and AMDumbfucks come up with more excuses like muh scheduler
>Inb4 AMDumbfucks starts telling everyone they are using it wrong

File: sg-2440-back.jpg (25KB, 650x320px) Image search: [Google]

25KB, 650x320px

>>59541864
>serial is dead

NEVER

>>59541844
The whole reason Ryzen exists as it is is that not every server workload is super SIMD/SIMT floating point heavy.

Everybody who runs a web server or a database on a Broadwell Xeon is paying an AVX tax they benefit nothing from.

AMD actually is working on Zeppelin-Vega hybrid MCM-based APUs, but nobody expects them to come out any time before the end of the year in the best case.

File: 2016-12-12-image-3.jpg (199KB, 1651x837px) Image search: [Google]

199KB, 1651x837px

>>59541844

Yeah but will it run CSGO at 700FPS

>>59541864

And that USB manifests as a serial device. So perhaps your point was that phy. / layer1 serial is dead (perhaps) - but serial is still the best way to bootstrap / initial config something.

When I have to work on 'prosumer' / SMB shit that the quick start guide says "SET YOUR IP TO 192.168.123.X" I want to shoot myself. Give me serial any day.

>>59541937
>R7 marketed as gaming CPU
>AMADs are still using the "muh gaming"

>>59541864
Serial is literally essential when everything goes south.

>>59541908
>6.6kw power supply

holy shit, is cloud gaming going to be a thing because of this?

That thing is like 100x PS4s

>>59541984
Actual power draw for that under a full load should only be 5000w~. The spec includes over provisioning for efficiency and cap aging.

>>59542041
If we're talking efficiency then the full load shouldn't be over 4000W at all, PSUs work best around 50-60% load

>>59542057
Its unlikely that the system would be at 100% torture test load levels during normal operation.

>>59542057
>>59542041
It's a 8GPU/4CPU system, the GPUs will be either 300W or 250W, the CPUs around 180W.

Roughly around 3200W total, ignoring NIC/SATA/SAS/IB

That is roughly in line of half the PSU for maximum efficiency

>>59541908

what's the x16 for ssds for, hwraid? or just some stupidly fast enterprise grade ssds I've missed out on?

>>59542057
(citation needed)
lots of high-end power supplies reach max efficiency at 80-90% load

>>59541278
>ryzen
>1800x can't even beat an i5-2500k released six fucking years ago

File: 1050ECEhot.png (132KB, 651x525px) Image search: [Google]

132KB, 651x525px

>>59542107
Here's a 1200W platinum seasonic, you'll see similar results with superflower and supermicro ones.

>>59542127
>thread about server chips
>shills gay men

wrap it up boys, we /v/ now

>>59541908
Wow, those lanes are lunatic, full 8x16GPU configuration? And still lanes leftover for other crap? Someone will love this.

>>59541984
>100x PS4s
STOP

>>59542322
PS4 Pro is 4 TFLOPS
This is 400

how am I wrong?

>>59542476
Computing power isn't linear.

>>59542476
hardware power isn't FLOPS alone.

e.g., 16xMI25 = 8.2 TB/s bandwidth = 37.6 PS4 Pros (218 GB/s each)

also, most modern rendering pipelines heavily rely on keeping previous frames' color/depth/normal/etc. buffers to do screen-space effects like TXAA, reflectons, etc., which would put way too much strain on RAM capacity even if art assets could be shared between client instances.

the list goes on, it would be a huge technical challenge getting things to work even if you could dedicate software efforts just to this, have bazillions of Jaguar core doing the CPU parts, etc.

>>59542869
so consoles will never die then?

>>59541278
>Server CPU with 16 core 32 thread 0,5ghz
>overheats when you open up nano
>no vmware support for amd

Yeah Intel better be afraid

File: good goy.png (604KB, 800x523px) Image search: [Google]

604KB, 800x523px

>>59541612
Go chase your shekels somewhere else, schlomo.

>>59543652
Literally everything you just said is retarded. I highly suggest ending your own life, faggot.

>>59541894
>>59542127
Kill yourself, shill.

>>59543736
What did that do? Unlock disabled cores?

>>59543824
https://en.wikipedia.org/wiki/Intel_Upgrade_Service

File: 1489903346057.png (504KB, 754x520px) Image search: [Google]

504KB, 754x520px

>more cores more better
>/v/

>>59541372

AMD needs to fix their RMA process before any serious provider adopts them.

I work at a contracted datacenter for a large "Cloud" provider. AMD has some horrid RMA process that takes way too long. It really needs some refinement, and there is consumer warranties that work faster then them. Intel, we have an account manager we call directly and they get the replacement components shipped out the same day.

>>59543846
Thanks.

>>59543888
...why are you even dealing directly with CPU manufacturers in the first place?

>>59541278
>>59541372
People like Google and Amazon pay extra to get tech before it is released they all ready on next gen Intel shit.

>>59544020

Facebook, Microsoft, AWS, Azure, Google, etc. All build-out their own servers and so do a lot of the contracted datacenters that work for them. It's much cheaper both upfront and in the long-run.

Buying a pre-built from Dell, HP, Supermicro, etc. is great for small-to-medium businesses that do not have the capital to hire an entire department to build/provide hardware support on custom servers. You really cannot even compare normal IT practices from that sort of SMB model to the large cloud environments.

>>59545239

I forgot to mention, typical deployments from these large cloud providers at a contracted datacenter typically get ordered in the megawatt range. We are not installing just a few servers or racks at a time...

>>59541612
>muh gayming fps!

>>59541908
will it run crysis?

>>59545588

Over Fabric.

How does this math work.

If Zeppelin has 20 PCIe lanes, and Naples is 4 Zeppelin dies, how can it have 128 PCie Lanes?

>>59546968
probably chipset or something

>>59547175
I'm pretty sure one of the slides said "direct I/O" which would clearly mean that its all on the CPU


Maybe AMD has a Zeppelin with a beefier southbridge, or maybe half the Ryzen southbridge is disabled

>>59541733

Intel is fucked!