AMD’s Epyc pummels Intel’s new Xeon-W workstation CPUs

>>62157712
It's funny how even Intel admits they're going to have performance regressions with 10nm.

File: 1491972340310.png (162KB, 633x900px) Image search: [Google]

162KB, 633x900px

>>62157767
DELET

i was quietly optimistic about amd returning with something good and it looks like they delivered so fucking hard

>>62157712
>With Epyc you get more cores, more performance per socket, more memory capacity, more PCIe lanes, and all for notably less money. Intel still wins, barely, on single threaded performance, something that has no place in the workstation market. In short Intel wins on nothing with the Xeon-W and SemiAccurate’s talks with large OEMs confirm this is reflected in sales figures. AMD wins this round, Intel didn’t even show up to fight.
Ayy.
The kikes are dead.

File: 1475211353488.jpg (108KB, 782x782px) Image search: [Google]

108KB, 782x782px

>>62157712
OY VEYYY

DELETE THIS REVIEW RIGHT NOW

WE NEED TO KEEP THE ISRAEL PLANT OPEN

SHREEEEEEEEEEEEEEEEEEEEEEEEE

>>62160190
>>62159004
You do realize you're not funny at all, huh?

>>62160189
>Intel didn’t even show up to fight.
dumb kikes got thrown right into the oven in more ways than one.

>>62160204
DOA "Xeons" are even less funny.

>>62157712
https://mobile.twitter.com/GFXChipTweeter/status/902648904440725504
He's back from India to destroy Nvidia... I hope

>>62157767
10nm is already better than what Skylake-X is using so it's.

>>62161503
That really depends on NGG pipeline being ready.
Right now it's not.

>>62157712
MODS! Delete this anti-semitic thread.

Reminder: if you buy AMD you're supporting nazis.

>>62160204
>NOOOOO STOP RUBBING IT IN MY FACE GOY

>>62157712
This could very well all be true but fucking hell I don't think I've ever seen an article this fucking biased. Hard to take it serious desu.

>>62161876
>calling out Intel's retarded marketing and even more retarded segmentation is bias
?
I too love buying shitty Xeon with 4 channel memory instead of 7551p.

File: 1502259085076.png (32KB, 600x600px) Image search: [Google]

32KB, 600x600px

>>62160204

I would buy a 32 core epyc but no one sells either a workstation board or the CPU itself.

Goodbye Intel, caught with their pants down. Jesus.

>>62161876
So it's now biased to compare different hardware and point out flaws?

What is wrong with you?

>>62161876
for it to be bias the competition must exist and in that case there isnt

>>62162263
Well technically you can waste money on these abortions called Xeon-W.

When AMD will start selling EBYN?

>>62162441
You can order servers with top three SKUs right now AFAIK.
Don't know about the rest.

AMD for life

File: intel amd and g.jpg (458KB, 1600x1200px) Image search: [Google]

458KB, 1600x1200px

>>62157712
I for one genuinely welcome the return of a competitive AMD. Shit's going to be fun, and good for consumers.
And I gotta watch Toradora again.

File: 1499933108581.jpg (88KB, 768x539px) Image search: [Google]

88KB, 768x539px

>>62157712

At Intel, our products are more stable than four cores glued together. Please buy our servers? We know we have forced you to buy more hardware than you needed for inflated costs. But that was all in the past! We learned our lesson, please baby, please. We have now segmented our product lines that we no longer understand what we are selling any more. Just to give you more options for your robust battle tested needs. Give us one more chance?

>>62163125
That sounds lazy, Rabbi.

>>62160189
>dead
Dead, rotten, desecrated and their (((souls))) devored

do xeons still have better single core (db implementations)?

>>62163407
At this rate, accounting price and shit, no.

does epyc have psp on it?

>>62163416
not price perf but purely single core. if it does some cases wont bother with epyc.

File: IntelMonolith-vs-EMIB.png (540KB, 1284x357px) Image search: [Google]

540KB, 1284x357px

G L U E
L
U
E

>>62163436
>wanting single core performance
>on a $10,000 28-core CPU
You're a dumbass.

>>62163465
you run 28 instances you retard, not every programme can be optimised for multiple cores

>>62160189
ding dong, the kikes are dead! which old kikes? the intel kikes! ding dong, the intel kikes are dead!

>>62163488
>muh unoptimized programs
Use less shitty software and you won't have to blow 10 grand on a piece of silicon shit, then.

>>62163407
Barely. They can't clock these too high.

>>62163508
not everything can be optimised though.

>>62163535
thats something i noticed, the clock rates are terrible.

>>62163570
Reading comprehension. Don't use shit that "can't" be optimized. If the author won't optimize it and you can't optimize it yourself it's time to either look for a new program or suck it up and stop paying thousands for CPUs that are crippled in JUST the right way to make it play nice with your shitware.

>>62163608
again not everything can be multithreaded, which is why running more instances is a workaround. i was never talking about certain softwares, rather there is no known way possible to multithread any implementation of a program.

>>62163488
Sometimes you want 28 instances, sometimes you want 28 threads on the same task.

You can't make one into the other, and more often it is better to thread out than spawn new instances.

>>62163642
>rather there is no known way possible to multithread any implementation of a program
You mean there is no known way possible to multithread every implementation of a program.

>Intel announces the MAGNYFICENT CPU, composed of 64 intel atom dies, each glued on top of their own 4MB L3 cache, glued on top of the multiplexed serial interposer
>"you want glue? we have the glue, we have a lot of glue!"

>>62163741
Whoa, why would I buy that over Navi?
Navi would have even more glue.

>>62163723
no, any.

>>62163705
of course

>>62163642
How the fuck did you find a process that can't be multithreaded at all?

>>62163779
"Any" would mean that can't multithread any program, which is just incorrect.

>>62163791
Somethings need to be done linearly, so you can't thread them out.

>>62157712

Xeon-W are fucking workstation CPUs

Epyc CPUs are server CPUs.

Knowing the difference may save your career.

Epyc aren't so hot in workstation loads because they don't scale much beyond 12-threads. You still need clockspeed/IPC to back it up.

Xeon Ws have the advantage here. If anything Xeon-W are Intel's attempt to fight off Ryzen and Threadripper.

2213 and 2215 are sleepers if your workloads don't require a ton of cores just clockspeed/IPC.

$300-400 for 48-PCIe and quad-channel DDR4. AMD has nothing comparable to it at this time. They might release a binned "Threadripper" a.k.a Ryzen in TR4 packaging to compete against it.

File: Mysterious Merchant.gif (4KB, 452x523px) Image search: [Google]

4KB, 452x523px

DELET

>>62163804
Explain what your goal is. Unless each and every step requires the previous step's result, it can be threaded to some degree.

>>62163819
1900x is 8cores and 60 lanes at $549.
And most werkstation loads are I/O bound anyway.

File: 1501250367381.png (33KB, 1060x722px) Image search: [Google]

33KB, 1060x722px

>>62163819
really cooks my macaco

>>62163819
Xeon-W is Xeon CPU with very small cache relative to amount of Cores.
So in Workstation loads it's going to choke just like 7700k does

>>62163892
It's 1MB L2$@core though.

>>62163819
EPYC 7251 128 PCI-E lanes. 400 USD. INTEL BTFO

I just realized, you can buy about 2,000 RasPis for the price of one top-of-the-line Xeon. What's the singlecore performance on a Pi?

>>62163984
That would be an interesting project

>>62161503
I just remembered this fucker was working for that Bollywood movie while he should've been working on vega.

>>62164051
>create Docker instance with scripts to launch necessary programs to start workflow, plus blinky lights for status
>set up rack of raspberry pis
>create 2,000 job bakery
>draw almost as much power as a Xeon

>>62163984
Like 1/20th of whatever Intel offers.

>>62164125
Numbers? I want the proofs, anon

>>62163984
the latency would be incredible. buy a tesla instead

>>62163488
Why stop at 28 instances precisely when you could be running 64 instances? Or 128 instances in a 2S configuration?

>>62163125
>Please buy our servers?
Please kill yourself?

File: 1499957247669.png (84KB, 653x726px) Image search: [Google]

84KB, 653x726px

>>62163819
>Epyc aren't so hot in workstation loads because they don't scale much beyond 12-threads.

hehe yeah infinity fabric has nothing on good ol bingbus

>>62160204
Maybe you're not funny?

>>62163836
I was just saying that sometimes that is required.
I'm not saying you can't thread most things.

>>62164532
So you're saying you're full of shit?

>>62164566
No, I'm saying that there are some things you can't thread, but most things you can thread.

I'm not arguing for single core performance, just that sometimes your application needs to be a single path.

>>62164656
That is extremely rare so why are you even arguing for it?

File: amd-epyc-7000-table.jpg (102KB, 658x374px) Image search: [Google]

102KB, 658x374px

>>62163819
you do know both supermicro and tyan builts epyc workstations right?
epyc dont scale ? we fucking know epyc scales almost linear since the launch of ryzen they dont have any restriction in terms of ccx
i cn totally see how amd doesnt have anything at that price bracket like literally

>>62164711
Because it does exist.

Don't just discount it.

>>62163461
Why the fuck intel doesn't simply sell their fabs?

I'm thinking on blow a shitton of money on a epyc workstation.
Only for e-penis.

>>62164941
go for it, buddy. post Crysis benchmarks plz :-----DDDDDDDDD

>>62160142
This

>>62160204
DELID THIS

>>62163488
you know that it only has better single core perf because it does a >4GHz turbo when a single one of them is in use right?