The 8 core Intel thread.

[ALXAndy]

The only part of this thread I am even remotely interested in is the £110 8 core Xeon.

Where did that come from?

Would be nice to upgrade my old workstation to one... It currently has a pair of E5430's (quad core 2.66GHz socket 771 chips)

hehe auction site.

It should be here today if the seller is a man of his word. Board was dispatched last night, so shouldn't be long now.

TBH after the fun I had with the 6 core Westmere ES I'm really looking forward to this. The Westmere had a lame clock and I couldn't overclock it (1.78ghz on all cores loaded) but tbh? with 670 SLI it absolutely roars. I was soooo surprised how well it ran games, especially stuff like Watchdogs and Metro : Last Light.

When I get the Ivy all set up I am going to run a second screen which I can snapshot to show core use. I will do it for everything.. Asus Realbench covers handbrake and GIMP etc, I will run the games and see what happens over the 16 threads.

Whilst people may think it's a waste of time it may prove very valuable to those wanting to spend £900 on the 8 core Haswell E.

 

[Martini1991]

Link is in the post you quoted under the graphs.

You've linked to completely different Xeon's effectively that aren't part of my argument ; http://ark.intel.com/products/76161/Intel-Xeon-Processor-E5-2687W-v2-25M-Cache-3_40-GHz

And the other being the Sandy bridge variant.

These CPU's costing 1,500 quid. Not the high core low clock CPU's I'm talking about.

EDIT : http://ark.intel.com/products/75267/Intel-Xeon-Processor-E5-2640-v2-20M-Cache-2_00-GHz

It'd seem it's only 2GHZ clock speed on all 8 cores (As Andy only quotes the Turbo's), so that'd put its performance lower than I thought.

 

[Mal X]

those Xeons must be powerful in one way or another, because they use them in Autocad 3D solid Model rendering with those quadro cards, they dont use our chips.

e.g .... two Xeons and 6 quadro cards, so yes they must be powerful, but in what way i've got no idea, and i'm not sure that others here know why either

it might be because combined they have something like 24 threads, but running at at a much slower clock speed of maybe 2 to 2.8 ....rendering a 3D image takes ages, i use my rig to render 3D DAZ women and it's as slow as hell, it takes two minutes to render an image at 1080p, maybe this is because i like big boobs ....ha ha

my guess is it's great processing power but slow clock speed.............and ours are the opposite, not sure, maybe the great processing power gives it the ability to render a complicated 3D shape really easily on multi cores all at once ( so it doesn't need to be superclocked at 4.6)

http://regmedia.co.uk/2012/04/23/dell_precision_t7600_workstation.jpg

 

[Martini1991]

You can make a powerful workstation with Xeon's, but the Xeon in the OP is far from "powerful" on its own as cheaper chips will outperform it.

 

[RobCrezz]

You've linked to completely different Xeon's effectively that aren't part of my argument ; http://ark.intel.com/products/76161/Intel-Xeon-Processor-E5-2687W-v2-25M-Cache-3_40-GHz

And the other being the Sandy bridge variant.

These CPU's costing 1,500 quid. Not the high core low clock CPU's I'm talking about.

EDIT : http://ark.intel.com/products/75267/Intel-Xeon-Processor-E5-2640-v2-20M-Cache-2_00-GHz

It'd seem it's only 2GHZ clock speed on all 8 cores (As Andy only quotes the Turbo's), so that'd put its performance lower than I thought.

You clearly didnt read my post...

 

[Martini1991]

You clearly didnt read my post...

I did, but the performance results you showed weren't relevant to what I was saying.

The rest? I have no rebuttal to it, you make accurate points, I don't fully agree with them, but that's it.

The Xeon from the OP, its only justification is that you can get two of them? But you're paying through the absolute nose for that justification, that to me is pointless.

 

[ALXAndy]

You can make a powerful workstation with Xeon's, but the Xeon in the OP is far from "powerful" on its own as cheaper chips will outperform it.

In a crude form yes.

However, you're forgetting many traits of the Xeon here. Firstly, they have the second QPI unlocked, so you can run two. That means up to 24 cores and 48 threads. At which point no 4770k can touch them.

Intel make sure you pay for it mind.

But the point is threads. I've never ever rated CPUs on software that doesn't support them. The only fair way to judge them is on their full merit, not by running software that deliberately avoids half of what they offer.

IPC is not important. Core count, and support, is important. Because of the position Intel are in they have been calling the shots. IE - "Here look, have our quad core CPU because IPC is all that matters !" - it isn't, not to me.

What I want is IPC, plus cores, plus support. I want it all and I don't want Intel dictating to me what they think I want. It's a typical trick of a corporation, tell people what they want.

At last they are releasing an 8 core CPU that's sort of mainstream. I mean, it isn't strictly mainstream because it goes in a ridiculous socket and costs ridiculous money, but Intel are going to have to do a complete Uturn now. No longer can they play the IPC card if they are charging £900 for an 8 core CPU. Now it comes down to cores and support for those cores.

And tbh? it's about the only avenue Intel have left now. As predicted they have hit a brick wall when it comes to gains and shrinks, so now they need to start offering out more cores.

And, with any luck that means support. So these Enterprise chips will slowly trickle down into the enthusiast sector, then into the regular desktop sector. And that means that it won't be just Cinebench for example that uses them properly.

8 Ivybridge cores with 8 HTs, even at the 2.3ghz (and 2.5ghz I should get if I tinker with the FSB) should be immensely powerful.

 

[RobCrezz]

I did, but the performance results you showed weren't relevant to what I was saying.

The rest? I have no rebuttal to it, you make accurate points, I don't fully agree with them, but that's it.

I said it wasn't the same CPU as I couldn't find benchmarks of the CPU in question, it was purely to demonstrate the vast difference in multi thread performance.

Do you have any benchmarks showing the 3770k beating this xeon in question in multithreaded operation?

 

[Martini1991]

I said it wasn't the same CPU as I couldn't find benchmarks of the CPU in question, it was purely to demonstrate the vast difference in multi thread performance.

Do you have any benchmarks showing the 3770k beating this xeon in question in multithreaded operation?

Get Andy to post up a result and I'll get my 4770K at 4.4GHZ to try and beat it (Or compare it to the Cinebench results on the forum)

I think I'll beat it in Cinebench for sure's.

Hell, I think with it running at its 2GHZ stock, it will come up trumps to a stock 4930K in Cinebench.

The problem is, that with the CPU's you've chosen, they're much higher clocked than the OP's Xeon, ergo, the results aren't directly comparable to the OP's Xeon, like at all.

All you've shown is X CPU being Y CPU. X CPU being undeniably better.

All I'm using is speculation on an educated guess, but if I'm wrong, when Andy posts a result, we'll see that.

 

[JonJ678]

hi thanks
i dont mind, i'm not sure which Xeon either, i was wondering if anyone knew of a good candidate worth testing. ...maybe

Out of the box a 3.3GHz 8 core chip can do about 190 Gflops on Linpack.

Size LDA Align. Time(s) GFlops Residual Residual(norm) Check
10000 10000 4 3.820 174.5529 8.700884e-11 3.068020e-02 pass
10000 10000 4 3.825 174.3367 8.700884e-11 3.068020e-02 pass
15000 15000 4 11.830 190.2386 2.225641e-10 3.505422e-02 pass
15000 15000 4 11.838 190.1086 2.225641e-10 3.505422e-02 pass
18000 18008 4 20.206 192.4489 2.894987e-10 3.170367e-02 pass
18000 18008 4 20.181 192.6887 2.894987e-10 3.170367e-02 pass
20000 20016 4 27.544 193.6553 4.097986e-10 3.627616e-02 pass

Two processors aren't getting anywhere near double this. NUMA is tricky.

Size LDA Align. Time(s) GFlops Residual Residual(norm) Check
10000 10000 4 2.534 263.1786 8.916344e-11 3.143993e-02 pass
10000 10000 4 2.556 260.8748 8.916344e-11 3.143993e-02 pass
15000 15000 4 7.439 302.5359 2.165846e-10 3.411244e-02 pass
15000 15000 4 7.441 302.4324 2.165846e-10 3.411244e-02 pass
18000 18008 4 11.969 324.9001 2.945255e-10 3.225417e-02 pass
18000 18008 4 11.952 325.3454 2.945255e-10 3.225417e-02 pass
20000 20016 4 16.133 330.6323 3.831049e-10 3.391318e-02 pass

Cinebench seems to be Windows only, so I won't have any results from it.

 

[Mal X]

thanks for that but we need to know how this compares to the 4790k, because i dont know enough about cpus to understand this

 

[413x]

for me it all depends on if my software can run on more than 4 cores.

AND

if single/dual core isnt compromised too much

my problem is i have software that is sometimes single core limited.. ie excel has a fair few single core functions
but then i also have software that uses at least 4 cores. but i dont know if it can use more.


my main software uses are a

excel - single for macros etc and multicore use for mass calculations - but more than 4 core support?
stats package - exactly same as excel
photoshop - def multi core in demanding tasks - but more than 4?
games - single and dual core has benifits - ie other CPUs
video transcoding - forgot this one - def 4 cores but dont know about more

as long as games arent limited i dont mind, and an i7 920 keeps up fairly well so i doubt it will

more what i am concerned about is impact of these low clocks on single threaded processes

 

[JonJ678]

thanks for that but we need to know how this compares to the 4790k, because i dont know enough about cpus to understand this

I can't find raw numbers for the 4790k. The 4770k scored 177 according to one website and seems to be essentially the same processor at a lower clock, so the 4790k can probably beat one of the ivy oct cores. I think the factor of two in Gflops/core is mostly due to AVX 2.0 vs AVX.

more what i am concerned about is impact of these low clocks on single threaded processes

This is probably a non-issue these days. Single threaded performance on a new processor is still far ahead of the C2D days. There are still programs out there that run on a single thread, but they're few and tend to be old enough to have run acceptably on the C2D systems of yesteryear.

 

[wazza300]

turning off hyperthreading will give big increase in gflops though

atleast its the case with intel burn test,i dont use linpak

 

[413x]

I can't find raw numbers for the 4790k. The 4770k scored 177 according to one website and seems to be essentially the same processor at a lower clock, so the 4790k can probably beat one of the ivy oct cores. I think the factor of two in Gflops/core is mostly due to AVX 2.0 vs AVX.



This is probably a non-issue these days. Single threaded performance on a new processor is still far ahead of the C2D days. There are still programs out there that run on a single thread, but they're few and tend to be old enough to have run acceptably on the C2D systems of yesteryear.

TBH I don't think it is. There are a fair few non threaded tasks that are either not able to be broken up for some reason or another that would definitely benefit from a hypothetical doubling of single core clock

Excel macro is probably the one most would be familiar with

 

[JonJ678]

Photoshop and video transcoding should scale rather well - the scene can be split into rectangles that are processed individually in both cases. R & S are single threaded, but all the expensive subroutines are C or Fortran (& either are, or should be, parallel). Games are gradually getting there, but it's hard to justify the development cost.

Excel macros are single threaded. Probably forever. Spreadsheets in general should be amenable to parallel computation, but I wouldn't expect to see it anytime soon! Are you really doing things in Excel that are so complicated as to bog down a modern cpu? Scary thought

I'm absolutely in agreement that there are tasks that are essentially impossible to parallelise. Lots of algorithms fit under that category. I haven't come across any programs that are both single threaded and annoyingly slow though. Perhaps I'm lucky.

 

[413x]

Photoshop and video transcoding should scale rather well - the scene can be split into rectangles that are processed individually in both cases. R & S are single threaded, but all the expensive subroutines are C or Fortran (& either are, or should be, parallel). Games are gradually getting there, but it's hard to justify the development cost.

Excel macros are single threaded. Probably forever. Spreadsheets in general should be amenable to parallel computation, but I wouldn't expect to see it anytime soon! Are you really doing things in Excel that are so complicated as to bog down a modern cpu? Scary thought

I'm absolutely in agreement that there are tasks that are essentially impossible to parallelise. Lots of algorithms fit under that category. I haven't come across any programs that are both single threaded and annoyingly slow though. Perhaps I'm lucky.

Yeah that was my take (educated guess) on how photoshop handles it

For me in excel I'm doing a lot of data checking, amending and loops. Some of the tasks populate cells down to the bottom of Excel 2010.
I am pretty sure many tasks I could do in a database.. And this is something I need to learn.
But.. I'm still not quite sure if everything can.

Some tasks in my SAS app are definitely single threaded. It might just be lazy programming.

 

[JonJ678]

The thought of a spreadsheet with that much data... it's the stuff of nightmares. How do you find mistakes in the logic? How do you check the dataset is still sane? I think I'll stick with SQL! Which is also not very fast, and thinking about it, also tends to be single threaded (transactional).

The first rule of parallel programming, is don't do parallel programming... it makes the programmer's job much more difficult interesting. They may just not have the budget for it

 

[mmj_uk]

turning off hyperthreading will give big increase in gflops though

atleast its the case with intel burn test,i dont use linpak

That's just the way IBT calculates it, I'm presuming it divides the overall result by the number of threads so non-HT processors tend to get a bigger score even though they'll be doing less.

Has OP come to his conclusion that Haswell-E 5xxx will be overkill and everybody should buy AMD yet?

 

[Andy7]

That's just the way IBT calculates it, I'm presuming it divides the overall result by the number of threads so non-HT processors tend to get a bigger score even though they'll be doing less.

Has OP come to his conclusion that Haswell-E 5xxx will be overkill and everybody should buy AMD yet?

That is the bit I am waiting for