Intel Xeon E5-2600 v2: More Cores, Cache, And Better Efficiency

[Brainstorms]

I bought a Dell T7500 with dual 3.3 GHz 4-core Xeon W5590s, 24 GB RAM, and a Quadro 4800 card on ebay. The machine came off lease after 3 years; the leasing company who sold it to me spiffed it up and put 2 new 2TB hard drives it in (with a fresh Win7 Pro64 install & COA sticker). I found a review of almost the exact same machine when new in 2009, which included a link to a scanned invoice: $12k. I got it for $1500 in 2012, and only added a nice sound board and a PCIe card for USB 3. You couldn't tell it wasn't new, and it's still an excellent machine at a great price.

 

[Brainstorms]

I bought a Dell T7500 with dual 3.3 GHz 4-core Xeon W5590s, 24 GB RAM, and a Quadro 4800 card on ebay. The machine came off lease after 3 years; the leasing company who sold it to me spiffed it up and put 2 new 2TB hard drives it in (with a fresh Win7 Pro64 install & COA sticker). I found a review of almost the exact same machine when new in 2009, which included a link to a scanned invoice: $12k. I got it for $1500 in 2012, and only added a nice sound board and a PCIe card for USB 3. You couldn't tell it wasn't new, and it's still an excellent machine at a great price.

 

[Draven35]

In order to show the effects of various cache sizes, we'd literally have to run a half dozen to a dozen different 3D rendering tests. Then people would say we are running too many rendering tests

Thanks for your reply. I agree it would make a dull read to see dozens of similar tests all lined up but it might offer an insight into how different cache sizes affect performance (games, mainstream apps, multiuser systems etc). At the moment we sort of assume that more is better but is there a sweat spot and what type of cache memory is more important for what sort of apps? It might make an interesting article for the geeks out there. Similarly with number of cores (with and without hyperthreading). Would even highly threaded games make use of all the available cores? how does the benefit change when you move to 2 CPU's on the same board? would an end user see any noticeable benefit? Obviously Xeons aren't aimed at gamers but they do offer a taste of the future. If I had money to burn then 2 Xeons would be nice but would they be any faster than an i5 running much of it's processing through a moderate GPU?

I don't know how many games that multithread have some kind of limit on how many threads they can launch. Look at the i5 vs i7 tests that include games, and i7 tests that include the same games, to see if they continue to benefit from the added cores. there *may* be a limit to how many threads they are designed to launch, and how many threads can run asynchronously, but the testing of something like that is beyond my purview.

 

[Draven35]

Definitely something wrong with Tom’s CineBench results.Every benchmark I have seen online scores a pair of E5-2687w’s at around the 26 mark for the multiple test, and my own E5-2687w’s lowest score has been 25.6 and highest 26.22, so I’m guessing the results should be 26 for the E5-2687w and 28 for the V2.

you did catch that its CineBench R15, right?

 

[InvalidError]

I don't know how many games that multithread have some kind of limit on how many threads they can launch.

The number of threads an application or game is running is a really poor indication of how well-threaded the software actually is since the bulk of those threads is usually background/housekeeping stuff and if you monitor the process with Process Explorer or equivalent, you will find that on a process with 20+ threads that might be using around 25% total CPU time, you might have one thread using ~20%, another thread using ~5% and all the ~20 other threads being under 0.01%. So, despite having tons of threads, the application itself barely qualifies as dual-threaded in terms of (probable) meaningful load on extra threads.

 

[Draven35]

Yes, that's my exact point. (also why renderers break the screen into tiles or bands and renders each tile/band in a separate thread to do multithreaded rendering)

 

[Jess Castro]

Rendering a playblast in maya is not really "rendering". If you want to do some real torture tests for maya renders, load up a scene with mental ray chrome and glass textures, hdri lighting, turn on final gathering and crank up the max sample quality etc. then do a Render of the current frame with mental ray...not a playblast. That will show you what a dual cpu can do vs a single socket. I could design that static scene for you if you cannot find a freebie floating around

 

[Jess Castro]

Also, any way we could we get the same benchs with the 12 core xeon in single and dual config to see what $2700 per really nets you in productivity?

 

[Draven35]

The mental ray playblast test is a CPU/GPU test, the actual rendering test isn't on that page. I've already mentioned that to Chris. For the mental ray render test we have a Maya version of the Tom's Hardware logo scene, rendering the same frames as rendered in the LightWave version (and the Max version, which uses V-Ray as the renderer). We're using the logo scene because at 1.7+ million polygons (because the N-gons didn't make it through FBX) it at least has the poly count of a production-level model.

 

[Draven35]

Also, any way we could we get the same benchs with the 12 core xeon in single and dual config to see what $2700 per really nets you in productivity?

For multithreaded workloads it is approximately double. (realistically, you gain about 95% more CPU speed, depending on how memory bandwidth intensive your test is.) for single threaded workloads it is no gain 😀

 

[Jess Castro]

What kind of texture is applied? ...putting mental ray material chrome or glass vs a base lambert will beat a system up. Final gather just adds extra insult to injury for the render time. If I'm not mistaken doesn't v ray leverage the graphics cards as well for the renders?

 

[Jess Castro]

What kind of texture is applied? ...putting mental ray material chrome or glass vs a base lambert will beat a system up. Final gather just adds extra insult to injury for the render time. If I'm not mistaken doesn't v ray leverage the graphics cards as well for the renders?

 

[Jess Castro]

Love the article btw.

 

[Jess Castro]

Love the article btw.

 

[Draven35]

What kind of texture is applied? ...putting mental ray material chrome or glass vs a base lambert will beat a system up. Final gather just adds extra insult to injury for the render time. If I'm not mistaken doesn't v ray leverage the graphics cards as well for the renders?

V-Ray has a GPU renderer (VRayRT) with lots of limitations, or a CPU-based renderer (V-Ray), just like iray versus mental ray. (have an iray test too, its the chair scene from Auodesk's content)

1.7 or so million polygons, cubic mapped generic tile texture, Final Gather and motion blur are ON... its not exactly a 'quick' test. I've been considering painting a custom map for it but it would end up being a HUGE UV map in order to cover it decently.

http://www.tomshardware.com/reviews/precision-t5600-xeon-e5-2687w-workstation,3610-4.html

 

[mapesdhs]

I bought a Dell T7500 with dual 3.3 GHz 4-core Xeon W5590s, 24 GB RAM, and a Quadro 4800 card on ebay. ...

I built up a Dell T7500 a while ago, certainly a very quiet system. Sometimes if I've switched monitor inputs
to something else, I forget it's turned on. 😀 I started with a bare-bones case (everything except CPU, RAM,
gfx, drives) for 200 UKP; Dell were selling some cheap because the side panel case had 'damage' (which was
really just a moderate scratch). I obtained two X5570s off eBay for a good price (so 8 cores & 16 threads total),
24GB RAM (low cost 2GB modules), a used 600GB 15K SAS (80 UKP) which I'll replace with an SSD soon.
GPU varies as I test different things (variety of Quadros and gamer cards). Only slightly tricky part was sourcing
the right parts for the 2nd CPU board, but got them ok for a good price. It runs nicely, stays at 3.2GHz most of
the time; decent threaded performance (faster than a 5GHz 2700K), but of course it's utterly hammered for
single-threaded tasks by any consumer oc'd setup, and beaten for threaded stuff by a stock 3930K:

http://www.sgidepot.co.uk/misc/tests-jj.txt

I've read that since the T7500, Dell's builds haven't been quite so good, which would be a shame.

Ian.

 

[Draven35]

The T5600 we tested as kinda loud...

 

[mapesdhs]

The T5600 we tested as kinda loud...

That's a shame, but mirrors what I'd read about their newer builds. Pity, as the T7500 is almost silent.

Ian.

 

[mapesdhs]

Draven35 writes:> My personal feeling is that the Premiere test is getting a little long in the tooth,> as is the AE test (hence why i went back and redid it for HD). I meant to say, did Chris mention to you that a friend of mine (also called Chris)is creating a more 'real-world' AE test for you guys to use? Not finished yet, butit hammers a system pretty hard (30 second animation).Ian.

 

[Draven35]

No, didn't mention it to me at all. My test is a motion graphics test with three picture-in-picture panels, similar to what you'd see in a DVD menu. So it isn't a test that is not real world at all, IMO. just having footage that I can use as i want that is useable as a comp has been problematic- a lot of the footage I have that I *could* use, is technically from a film made under the SAG student agreement, and it might be a violation to use said footage in a for-profit venture outside of the film, even....

I was planning on doing a separate comp test, but that is going to be in.... another product we're going to add to our testing suite.... (not gonna discuss that in an open forum, sorry)

 

[mapesdhs]

No, didn't mention it to me at all. My test is a motion graphics test with three picture-in-picture panels, similar to what you'd see in a DVD menu. So it isn't a test that is not real world at all, IMO. ...

I should clarify, when I said real-world, what I meant was that it's an animation typical of what one might
create for a proper complex movie sequence, advert, video or suchlike. It takes 2 hours to render with
the default settings using two 3GB GTX 580s (faster than a Titan). One can of course create a shorter
test simply by reducing the AA level in the RayTrace3D options.

I was planning on doing a separate comp test, but that is going to be in.... another product we're going to add to our testing suite.... (not gonna discuss that in an open forum, sorry)

No prob! Feel free to email if preferred )see below); btw, Chris has a screenshot I sent him
of the test being constructed, a reasonably recent snap, I'm sure he could send it to you.


Hey, do you happent to know if AE is able to make any use at all of 64bit CUDA ops? The
other issue I'm trying to resolve is whether most of the main AE user interface uses OpenGL
or D3D for the display, ie. is one better off having a Quadro card or something like a Titan
or 780Ti for the primary display. If it's mostly OGL then I'd have thought a Quadro would be
better. The system I built for the AE guy has a Quadro 4000 and three 3GB GTX 580s.

Ian.

 

[Draven35]

AE only uses CUDA for the 3d rendering stuff, and that is only useful for things like flying logos (and i mean 3d rendering, not 3d compositing

 

[mapesdhs]

AE only uses CUDA for the 3d rendering stuff, and that is only useful for things like flying logos (and i mean 3d rendering, not 3d compositing

Yes, I know, the RayTraced3D function. But ask Chris to send you the image I sent him,
you'll see what I mean.

Ian.