Ryzen HEDT will kill Skylake-X, but not for the reasons you think

Freshly released specs show Zeppelin is a full SOC. Even with 48 PCIe lanes, 10GbE can be enabled as a virtual afterthought.

Other urls found in this thread:

support.amd.com/TechDocs/54945_PPR_Family_17h_Models_00h-0Fh.pdf
asus.com/us/Motherboards/X99WSIPMI/
twitter.com/AnonBabble

Each Zeppelin die has 32 general purpose IO lanes that can be divided between PCIe, interconnects for 2P systems, up to 8 SATA ports, and up to 4 10GbE ports.

SFP+ ports can be added for basically the cost of the cages, and 10GBASE-T RJ45 ports can be added for mostly just the cost of the PHYs.

Hey neat, Intel is even more fucked than we realized.

The hilarious thing was that Intel was hiding the 12c model as their recent big reveal, while Ryzen HEDT still has 16c, more PCIe lanes, and apparently can also throw in 10GbE because hey, why not?

It's pretty safe to assume Ryzen HEDT will have 12c and 16c models, but I wonder if they'd bother selling 8c parts just to people who just want more RAM/PCIe/whatever than the AM4/R7 platform supports.

>SkylakeX

Lol

Skylake-E has its own problems.

But can it hardware accelerate 10 bit HEVC?

>wanting a integrated GPU on your $1000+ CPU

b-but muh hardware drm needed for UHD blurays

And nothing of value was lost.

kek.

Raven Ridge APUs (4c Zen + mini Vega) is expected towards the end of the year, and we might see a Zeppelin+Vega hybrid MCM platform next year if we're extremely lucky.

B-but you can watch your chinese cartoon goyim

Call me back when AMD cpu's can get past 4.1 ghz on air

As if i'd ever let my GPU do the decoding when I have that much CPU horsepower.
Even on my 4 core Intel I have hardware acceleration disabled in mpv

this is a pretty weird lineup even if you allow for a conspicuous typo or two (18c*3.2GHz@130W, 18c*3.0GHz@105W).

W 2.0 2.1 2.2 2.3 2.4 2.5 2.6 2.7 3.0
-- --- --- --- --- --- --- --- --- ---
14| 125
16| 120 145
18| 130* 140 165 200
20| 125 150
22| 140
24| 150 205
26| 150 165
28| 165 205

>8 SATA ports
>HEDT
Pick one.

Not really, some parts don't have AVX512 or 6 channel memory.

That's just stupid. Do you force CPU rendering for DX applications too?

Decoding video isn't intensive, there's no need for hardware acceleration of video on a desktop if you have more than 2 cores.

That's inside EACH cpu alone. Add chipset. Add whatever motherboard manufacturers decide to implement.

DELID THIS

So intel is getting fucked sideways, neat, do you have any links?

Unless it clocks higher, it literally makes no difference for my needs.

reference manual:
> support.amd.com/TechDocs/54945_PPR_Family_17h_Models_00h-0Fh.pdf

everything on the 2-die HEDT is still rumor at this point though.

enjoy your wait for Coffee Lake (and still being disappointed) then.

>MUH CLOCKS
We're talking server shit here, Sup Forumstard.
This is the dumbest comparison I've ever seen in my entire life.

Even my powermac g5 quad can decode 1080p h265 on mpv, hardware decoders are for laptops.

The RZ2700 platform looks weird as shown.
With 44 PCIe lanes and a x4 link to the chipset, there should be 16 lanes left for more PCIe, SATA, 10GbE, or whatever.

The idea of workstations with built-in 10GbE makes my dick diamonds though. Local bulk storage as a concept needs to die off, and finally getting universal 10GbE will help people finally move all their shit to NAS where it belongs.

Welcome to niche-use-caseville, population you. And only you. Enjoy your stay.

This is HEDT, it doesn't appeal to >muh single core fags like you anyway.

>imagine being such a retard

M-MOAR RING BUSES :------------D

>MUH GIGGAHURTZ
why don't you check the blazing fast clockspeeds of Xeons? oh wait, they're all slow as fuck because efficiency.

>server shit
Actually HEDT, but same theory applies. Nobody goes to these platforms for clockspeed.

>add another bingbus :------DDDDD
>ebin :---------DDDDDDDDDD

Did you guys forget that Skylake-X isn't only going to be in Xeons? Have we forgotten what HEDT stands for? I didn't say these improvements were useless period. They will do very well in certain market requirements. Especially Naples. However to claim these improvements will across the board kill Skylake-X is humorous given how many markets the two companies will be competing in.

Intel will kill it themselves by pricing it 50% higher than AMD while offering nothing but higher TDP and lower performance.

>Have we forgotten what HEDT stands for?
Intel's HEDT has no ECC support so it's entirely pointless for any kind of workstation builds.

delirlake-X can't be competitive without massively depreciating intels current lineup.

Most of the production machines that I still see use Q or Z chipsets on mainstream sockets, with the occasional HEDT 4-8 core with gaming RAM. I think ECC RAM isn't all that common for serious users.

Because Intel hasn't offered it on anything and saves it for their server motherboards for ((((((((((reasons))))))))))).

Also, pic related for the big details from the reference guide, for the lazy.

Current Broadwell-E chips range are:
> 6c*3.4GHz - i7-6800k ($400)
> 6c*3.6GHz - i7-6850k ($600)
> 8c*3.2GHz - i7-6900k ($1100)
> 10c*3.0GHz - i7-6950k ($1700)

Though they've dropped in price finally thanks to Ryzen, you can't pretend these feel remotely "high end" anymore, when the R7 1700 is nearly on par with them for $300.

Not all workstations are the same. The CAD house I work for doesn't have ECC in any of our machines as it's just not necessary for our workload.

Yes, the platform they go with is geared towards high-end consumer space with Intel's X99 chipset. The X399 chipset coming from AMD is geared much more towards enterprise requirements and such to compete against the Xeons. Which is absolutely fine. But all I really want is something equivalent to "prosumer" focused chipset (though I hate using that term) with a higher clocked R7. R7 has the CPU chops to compete against Intel's HEDT offerings, but is let down by the platform compared to X99 (or the upcoming X299) and while X399 looks to solve that, it seems to be more competing against the Xeon C- series platform offerings, rather than the high-end consumer market.

Which - again - is fine, but it does leave a niche buy very profitable crack in their lineup against Intel. That's really my main point here.

lol, a lot of usage scenarios don't require ecc (especially workstations)

A flipped bit once a decade is not worth the performance hit for most people

You call me when Intel can get their TDP under 140W.

post ur cheapest ryzen clusters

Maybe Zen+ will have something like that. Just speculating.

Hmm? The X390 is AMD's version of HEDT, high clocks single socket board with up to 16 cores.

Their enterprise stuff is clocked way lower and has multi socket support, among support for IPMI and Infiniband

>no 4.0GHz 16 core with two sockets

Life is shit.
Yeah I know you'll need a motherboard designed to handle that much fucking current and I'll gladly dish out $250 extra for it

>no IPMI support out of the gate
Trash
Even my workstation X79 and X99 motherboards had IPMI 2.0 support

A 7700k at 5.0 ghz uses less wattage than an 1800x at stock dumbass

X390 mobo wouldn't be that expensive with less chips if the CPU is a SoC.

I've only seen the Naples I/O layout and it has IPMI, I haven't seen the 16 core chipset version but I simply assume it's not there since it's not really a necessary workstation feature.

And then you woke up

yeah but can it play crysis on max settings xD

>4 cores uses less power than 8 cores
kek

Irnoic shitposting is still shitposting.

In Intel's case it doesn't.

oops

>PCie x32
>x32

There's nothing precluding a 180W or so SKU for 16c Ryzzen HEDT parts.

Really though it's probably a lot better to have 2 Zeppelin dies sharing an MCM rather than having their own sockets. A GMI link between dies goes up to 100 GB/s, while the xGMI links on the general purpose PHYs would only go to 20 GB/s with 16 lanes used, which isn't even enough to stripe data across all the DDR4 channels.

xGMI is 64 lanes, so enough bandwidth.

Anyhow, the 180W 16 core SKUs for HEDT are all but confirmed at this point.

The only rumors at this point for 180W parts are for presumably low-clocked Naples parts.

Also, xGMI is only 16 lanes per die and needs 4-die Naples for 64 lanes.

Honestly I'd rather have a 180-190W 3.6/4.0 16 core than a 2.7-2.9GHz 32 core at 180W

That's because multi sockets are limited to the Naples lineup, which I think won't go under 24 cores.
And that provides 64 xGMI lanes still.
The 16 core enterprise parts are BGA Xeon-D competitors and aren't dual socket.
And I doubt anyone will make HEDT dual socket motherboards since the market is beyond niche.

ayy lmao

I can see 16 core naples with 2 core per ccx cut, but that just seems a waste of packaging.
But then again having that much fucking I/O and memory against the completely hopeless 16 core Intel parts would be hilarious in itself

I'm betting that high-clocked 2*8c and 4*4c parts will actually both come out, since they will take next to no engineering effort to productize and would have potentially really high margins.

Remember that Intel sells a 4c E7 Broadwell for $7k just because it has 60 MB L3 by virtue of being natively a 24c die. I'm sure AMD would love to sell some 64 MB L3 4-die parts with only 8 or 16 uncut cores, since they could turn some of their absolute shittiest chips into gold.

>x99 motherboards with IPMI
What?

asus.com/us/Motherboards/X99WSIPMI/

>two cpu coolers
user

wrong thread

That's some insane muxing there.
So is it possible to get 40GbE with bonding?
Finding a motherboard with that much NICs will be difficult

Man, AMD wasn't fucking around, the die has too much non-consumer features it's unreal, and it still does phenomenally in the consumer space, I can't wait to see what a trimmed down die for plebs next or 2 years later will have.
As it stands now this is a born and bred server die.

...