AMD THREADRIPPER VS INTEL SKYLAKE X

[Deleted member 66701]

Lets compare to this super sexy Prescott board. As you can see, aesthetically a lot of work was put into these boards during this period.

Spoiler

I'd happily buy a board like that I'd it was cheaper.

 

[Silent_Scone]

Instead you bought the Aorus Gaming 7 lol . Bit of a contrast

 

[Curlyriff]

M.2 cooling is nice but I don't see why a fan is required. Surely a decent heatsink is sufficient? I'm sure that motherboard that had the M.2 slot mounted so the cards were sticking up from the board rather than laying flat had no throttling issues either so there are clearly passive solutions that work. Why would I want another small, noisy fan in my case?

Em I am in agreement with you. The one SS posted by Asus has clean lined passive heatsink blocks and covers. The EVGA one has the fan, it has awful lines, overlay detailed heatsinks, awful balance of aesthetics.

 

[Plec]

Said it before and i'll say it again. I wish motherboard manufacturers would do cheaper versions without the bling and general pointless aesthetics but keep the functionality and component quality. Hate paying for all the crap they slap on

For my needs i agree. My own personal builds are silent, little light emission with non windowed cases (i know boring ). Plus, simpler board designs have less points of failure - which really suits my needs - not that i would see it if it was an aesthetic issue.

I basically treat my components like a self-harming prisoner in isolation - padded case, no window, and all lights disabled.

 

[Deleted member 66701]

Instead you bought the Aorus Gaming 7 lol . Bit of a contrast

I didn't buy it for the looks though. If they'd have stripped all the rgb **** off it and charged 100 less I'd have bought that model instead.

 

[JonJ678]

That same office would also have no use for a core i9, this argument is as pointless as it is insane, arguing some offices not needing multi-threaded performance and there in need the core i9 is bonkers

...

Also: If you think Anyone running server rooms don't care about heat and power consumption you are deluded, those things are of primary concern.

Bit weird. Our server room is multisocket xeons which we access via crappy laprops, roughly one server per engineer. I don't know what you're imagining.

Neither heat nor power matter to me. It's minimising wall clock time per simulation. I'm in this thread because I'm interested in sixteen cores at higher frequency than my current numa box.

 

[8 Pack]

Bit weird. Our server room is multisocket xeons which we access via crappy laprops, roughly one server per engineer. I don't know what you're imagining.

Neither heat nor power matter to me. It's minimising wall clock time per simulation. I'm in this thread because I'm interested in sixteen cores at higher frequency than my current numa box.

We are working on some really crazy rack mount servers with hcc binned Intel cpu's. Ofcourse designed for time critical tasks. Please contact our business team or me direct [email protected] and we can discuss our unique bespoke solutions.

 

[humbug]

Bit weird. Our server room is multisocket xeons which we access via crappy laprops, roughly one server per engineer. I don't know what you're imagining.

Neither heat nor power matter to me. It's minimising wall clock time per simulation. I'm in this thread because I'm interested in sixteen cores at higher frequency than my current numa box.

At the time i wrote that Intel only had the 10 core which is slower than Threadripper at any speed.

If all you are looking for is outright speed then you should be looking at the 32 core EPYC.

 

[Blackbadger]

If all you are looking for is outright speed, a high end Xeon system will outperform an Epyc system by huge amounts. At certain price brackets and use cases epyc systems can be competitive it depends on workloads, for total system performance AMD are no where close at the high end. Due to the way EPYC is constructed, more than 2cpu is not currently possible, (as each epyc chip is already 4cpus). Both platforms have advantages and disadvantages.


Yes we are way past the realms of consumer pricing, However it is a sign of things to come. There quite rightly has been a lot of frustration that Intel have been sleeping and not pushing development or performance, it is not the case at the high end. A new dual Xeon system is approximately the same performance as a top end quad Xeon ivy bridge system from 3 years ago. For us as consumers, for the prices we want it has indeed been stagnant. Hopefully AMD's presence again will bring this available power closer to accessible in the coming years. Thread ripper certainly has made a big impact for the price, kudos to Amd for that.

Choosing cinebench as a guide with the best chips currently available.

Dual EPYC scores around 5500
Dual Xeon scores around 7200
Quad Xeon are scoring around 11500 However the benchmark is failing at this stage due to completing so quickly the resources are not fully utilised before the benchmark completes and scheduling issues.. Basically a more intense scene is needed. Cinebench effectively has a ramp up time whilst resources are allocated, making the initial few seconds wasted regardless of the system used.
Octo Xeon scores are only available to lottery winners

I would have preferred Intel to have released SkylakeX on the LGA 3647 platform as they have with SkylakeXS. It's pretty similar in size to the Threadripper/EPYC and would provide a easier migration/handmedown from the server chips, when they decide us lowly consumers are ready for the power. I am sure there are reasons, technical, market segmentation etc.

 

[Rroff]

Neither heat nor power matter to me. It's minimising wall clock time per simulation.

Depends a lot on your setup though - I used to do some work for one of the bigger GSPs - we had rented space with specific power wattage availability and cooling capability where the price went astronomical for higher levels of provisioning so being able to get the maximum performance for the lowest power draw and heat output would be a massive advantage.

 

[humbug]

If all you are looking for is outright speed, a high end Xeon system will outperform an Epyc system by huge amounts. At certain price brackets and use cases epyc systems can be competitive it depends on workloads, for total system performance AMD are no where close at the high end. Due to the way EPYC is constructed, more than 2cpu is not currently possible, (as each epyc chip is already 4cpus). Both platforms have advantages and disadvantages.


Yes we are way past the realms of consumer pricing, However it is a sign of things to come. There quite rightly has been a lot of frustration that Intel have been sleeping and not pushing development or performance, it is not the case at the high end. A new dual Xeon system is approximately the same performance as a top end quad Xeon ivy bridge system from 3 years ago. For us as consumers, for the prices we want it has indeed been stagnant. Hopefully AMD's presence again will bring this available power closer to accessible in the coming years. Thread ripper certainly has made a big impact for the price, kudos to Amd for that.

Choosing cinebench as a guide with the best chips currently available.

Dual EPYC scores around 5500
Dual Xeon scores around 7200
Quad Xeon are scoring around 11500 However the benchmark is failing at this stage due to completing so quickly the resources are not fully utilised before the benchmark completes and scheduling issues.. Basically a more intense scene is needed. Cinebench effectively has a ramp up time whilst resources are allocated, making the initial few seconds wasted regardless of the system used.
Octo Xeon scores are only available to lottery winners

I would have preferred Intel to have released SkylakeX on the LGA 3647 platform as they have with SkylakeXS. It's pretty similar in size to the Threadripper/EPYC and would provide a easier migration/handmedown from the server chips, when they decide us lowly consumers are ready for the power. I am sure there are reasons, technical, market segmentation etc.

if you're talking about Cinebench a single EPYC "scores around 5500" Dual would be around 11000

One EPYC chip has 64 threads, Dual has 128 threads.
Soon AMD will be launching Starship, a 96 threads per chip leviathan. That's getting on for 200 threads on a dual platform, nothing Intel has can get even close to that behemoth.

 

[jigger]

The only CPU's Intel offer that AMD has no answer for are the very highly clocked Xeon quad cores with 20Mb+ of cache and maybe the 45 watt i7 greenlow Xeons.

 

[Blackbadger]

Understandably there is not as much information available for the higher end, especially with these new platforms so the scores may be off.

The scores are from the verified benchmarks I can find.

Dual EPYC 7601 - 5544cb 64 cores - 128 threads
(I do note that it is running in vmware so it could indeed be higher, however does match the chart posted above. And multiple other bench's. I have also seen talk of dual EPYC scoring around 6800, which is not unreasonable to believe either, however I do not believe it will even remotely close to 11000 as suggested) Neither is bad.

Dual Xeon SkylakeSP 8168 - 7188cb 48 cores 96 threads
Dual Xeon SKylaleSP 8180 - 8300cb 56 cores 112 threads

Quad Xeon SkylakeSP 8180 - 11504cb 112 cores 224 threads.

The Amd systems do offer good value for money and as explained do offer competitive platforms at certain brackets. however cannot compete at the top end with intel for multi threaded performance, nor can they for single threaded. You will pay for the extra speed though.

The Amd Starship will not be available until 2019 and hopefully a 24core Threadripper to go alongside it

As said both platforms have advantages mainly Amd price, Intel performance. Both have very high importance.

 

[jigger]

Single thread AMD have desktop i7 performance and make a very compelling case when it comes to big CPU counts. The only missing areas I can see are the greenlows and £10,00 £20,000 specialised quad cores.

 

[humbug]

@Blackbadger

Again that's not a Dual CPU, that's a Single CPU, EPYC 7601 has 32 cores.

Your only saying its a Dual CPU system because Windows server 2012 is reading "2 CPU's" Windows server 2012 is ancient, it doesn't know what it is looking at, all it does know is there are (as listed) 32 cores, there isn't even such a thing as 2X 16 core EPYC.

Here... 1 CPU 32 cores 64 threads. https://www.amd.com/en/products/cpu/amd-epyc-7601

Its actually pretty obvious its a single 32 core CPU.

So it is scoring 5500 (per CPU), 2x that would be 11000.

Dual EPYC (Naples) "64 cores 128 Threads"

StarShip is coming Q2 next year.

 

[jigger]

I want 128 threads. I'm not sure what I'd do with them but I want them
in my life.

 

[Doobedoo]

Ouch that Xeon SKylaleSP 8180 cost's almost £9300 each, damn that is one expensive cpu!

That's over £18,500 to reach that score and about £37,000 for the quad which is a little over twice the performance of the Epyc 7601.

Epyc 7601 cost's less than half of what the 8180 cost so you could buy almost 9 7601's to every quad socket 8180's.

 

[Vince]

Man I want that EPYC chip, justifying it may be difficult but that doesn't stop the want. Ill swap my 1950x for one right now. Thank you please!

Twice as fast as 1950x in one package is always a winner.

 

[Doobedoo]

Thinking about it, some of those cinebench scores seem a bit off. I've seen 16c 32t 1950x's get about 3000, so 5500 for 64c 128t doesn't seem right.

Perhaps cinebench is not the correct tool to measure the performance of these server chips.

 

[Blackbadger]

You are correct server 2012 is not a ideal OS to use, and from my own testing 2016 does give better performance for things such as cinebench.
However it is reporting correctly. The screen shot clearly shows 2x 32 core processors and cinebench clearly shows 64c/128t thats how windows and cinebench display things.
The scores could be off and I expect them to be better once it is released, however the hardware is being correctly detected and displayed.

Here is a video that is actually relevant, showing the dual epyc running on server 2016 showing a similar score that is posted. It's a very pro AMD video, showing the price performance ratio against old xeon's.

C'mon be better humbug, less strawman nonsense. You are smarter than that.

In a high end build Xeon CPU prices although to us look astronomically expensive (which they are), they are not as significant as the whole platform cost. After all the rest of the parts are in place even at £8k per cpu, that is usually only a fraction of the actual system cost.