Is it a good time to buy a 16-cores Xeon workstation?

Become a member of the CGSociety

Connect, Share, and Learn with our Large Growing CG Art Community. It's Free!

THREAD CLOSED
 
Thread Tools Search this Thread Display Modes
  07 July 2013
Originally Posted by earwax69: I'll ask the Titan then. perfect.

As for Intel roadmap... 45% more power in 2 months would be worth the wait. I dont think they'll be able to delay the purchase but I will try none the less. How to know how much to reserve is another thing. Nopbody know the price of the 10 cores xeon.


The new ones will be similar in price. Intel has kept their top end 2-socket Xeons around the same price for several generations.
__________________
Fix it in pre.
 
  07 July 2013
Originally Posted by dmeyer: The E5-2687w v2 chips will be out soon and will give you 10 core / 20 threads per chip at 3.4 Ghz. Roughly a 45% speed boost over the current E5-2687w. Not sure what the turbo speeds will be, but these will be the fastest of the new crop. The 12 core chips top out at 2.7.

Depends on your time frame. I'd wait 2 months if I could, but the current ones are solid as well.

I'd bet they'll start trickling out within the next month. They're being deployed in supercomputers now so they are clearly already being produced in some scale.



I'm curious what your evidence is for them being 45% faster since they're slated to be capable of 225 GFLOPS each compared to the current model's 185 GFLOPS
http://www.xbitlabs.com/news/cpu/di...formance. html

With the E5-2687w v2 being their flagship workstation chip, compared to the current R5-2687w it has 25% more cores, and 7% faster architecture which brings things to roughly 33.75% faster if the chip were to scale perfectly.

Most rendering software doesn't scale perfectly these days, so I'd think real-world speeds would be more like 30% faster rendering, possibly even less.

Last edited by sentry66 : 07 July 2013 at 10:39 PM.
 
  07 July 2013
Originally Posted by sentry66: I'm curious what your evidence is for them being 45% faster since they're slated to be capable of 225 GFLOPS each compared to the current model's 185 GFLOPS
http://www.xbitlabs.com/news/cpu/di...formance. html

With the E5-2687w v2 being their flagship workstation chip, compared to the current R5-2684w it has 25% more cores, and 7% faster architecture which brings things to roughly 33.75% faster if the chip were to scale perfectly - which most rendering software doesn't.


E5-2687w
8 cores * 3.1 Ghz = 24.8 ghz

E5-2687w v2
10 cores * 3.4 Ghz = 34 Ghz

So that gets you 38% faster. I added 5-7% estimate for architecture improvements. So 43-45%.
__________________
Fix it in pre.
 
  07 July 2013
Originally Posted by dmeyer: E5-2687w
8 cores * 3.1 Ghz = 24.8 ghz

E5-2687w v2
10 cores * 3.4 Ghz = 34 Ghz

So that gets you 38% faster. I added 5-7% estimate for architecture improvements. So 43-45%.



oh yeah, forgot about the 300mhz speed bump

still, LINPACK GFLOPS is about as granular as you can get with fine floating point operations. It's interesting that intel is rating the chip as 21% faster in a best-case scaling scenario with 225 vs 185.

It makes me wonder if they haven't been able to increase bandwidth latency as much as would be ideal to keep all those cores fed at those speeds.

Even still, we can't assume both core count and speed will scale linearly because it never does. It scales worse as they increase, usually around 85-90%. Ivy-bridge architecture isn't enough to negate the difference. Sandy bridge sure was though when it came out.

Last edited by sentry66 : 07 July 2013 at 12:23 AM.
 
  07 July 2013
My math was a bit off. The 7% estimate would need to be applied to the whole sum since you'd get the arch improvements across all cores, so that'd push it up to 47% if one assumed the same turbo values, and if the numbers posted out there are accurate. Not to mention faster memory interface.

The 225 number seems like a pretty general estimate. That slide could have been produce 2 years ago with as far out as they plan architecture migration.

Still it's on paper. It will obviously be faster for approximately the same price. Why not wait if you can?
__________________
Fix it in pre.
 
  07 July 2013
yeah but turbo mode doesn't really apply much when you're stressing all cores


Anyway, I could be completely wrong. The benchmarks will all be revealed all once it's released. I'm just saying, intel specifying the LINPACK score to measure GFLOPS is already a known benchmark in itself.

I hope it's 45+% faster....I just have serious doubts, at least with real-world CG rendering software.

I mean put it this way - with the last xeon upgrade, sandy-bridge was a major architecture improvement (like 25%) over westmere and increased the core count 33%. It managed to get cinebench scores of 38% higher over the older top chip.

This time the core count is increasing 25% more cores and a minor architecture improvement (7%). I'm not sure how that will get a 45+% higher cinebench score over today's top xeon.

And cinebench is generally considered to be about as linear as 3d rendering can get unlike a lot of modern scene files that utilize several single-threaded functions before multithreading kicks in.


That intel slide was released just last week, but it's true that intel could have created it long ago. Though it does include some very specific things about the new phi's which make me think it's fairly recent.




About waiting:
Speaking from experience, because I've been in situations where money was available to buy new hardware/software and I chose to wait a few months for the new chips or software version to come out. The money will find a way to evaporate. Someone else will need it, and it will slowly get chipped away at. Also just how businesses are ran with their fiscal year and tax deductions - the business wants to recirculate the money ASAP for tax deductions. Holding onto it is too expensive in the long run.

We still don't know if the chips really will be available to the mass market in 2 months. Probably some vendors will have them and will be updating their BIOS's for their workstations and then just like apple has said, will ship by the end of the year. You'll probably be able to buy the chips themselves online in a couple months, but I wouldn't automatically assume workstation vendors are going to be shipping the new systems to your door in 2 months. Most vendors don't do preorders for new systems before they're ready to ship, but who knows.

Last edited by sentry66 : 07 July 2013 at 11:59 PM.
 
  07 July 2013
well Apple has a close relationship with Intel and they know the next Mac Pro is going to be in high demand, so I'm sure they will be shipping the Mac Pro it as soon as vendors can buy those chips. It's happened in the past

and, considering they have Mari and OS X 10.9 running on the Mac Pro from the WWDC demos, it's safe to say they have close-to-final silicon already.
 
  07 July 2013
Look like I could reserve that budget until december.
__________________
So long and thanks for all the fish
 
  07 July 2013
if it was Haswell, I'd say buy now since there is almost no difference in performance since it's just about power savings but Ivy Bridge Xeon is actually a noticeable upgrade for performance from Sandy Bridge Xeon
 
  07 July 2013
From what I've seen, at the same clock speeds ivy bridge is around 7% faster than sandy bridge and haswell is 6-7% faster than ivy bridge - give or take
 
  07 July 2013
And 4 more cores. For the same price. Well worth waiting 2-3 months.
__________________
So long and thanks for all the fish
 
  07 July 2013
Originally Posted by sentry66: From what I've seen, at the same clock speeds ivy bridge is around 7% faster than sandy bridge and haswell is 6-7% faster than ivy bridge - give or take


really? unless I remember wrong, FPU performance was the same for the latter. edit: I think I realized the problem - I was comparing different clock speeds so it's actually faster than I thought since a 1.3GHz Haswell is the same as a 1.8GHz Ivy Bridge in Cinebench:

http://www.anandtech.com/show/7113/...core-i7-4650u/2
 
  07 July 2013
Originally Posted by cgbeige: really? unless I remember wrong, FPU performance was the same for the latter. edit: I think I realized the problem - I was comparing different clock speeds so it's actually faster than I thought since a 1.3GHz Haswell is the same as a 1.8GHz Ivy Bridge in Cinebench:

http://www.anandtech.com/show/7113/...core-i7-4650u/2


Don't forget to account for the "turbo" since they're the same for the single threaded Cinebench, but not the multithreaded.
__________________
http://www.whenpicsfly.com
 
  07 July 2013
Originally Posted by earwax69: And 4 more cores. For the same price. Well worth waiting 2-3 months.


The only thing is that I doubt you'll see dual-CPU machines for another few months after that due to scarcity. Nevertheless, a single-CPU Ivy Bridge Xeon machine at 2.7GHz and 12-cores will likely perform about the same as a dual 8-core Sandy Bridge Xeon at 2.4GHz and cost a lot less than the dual. For tasks that are poorly threaded (Photoshop, most video, almost anything in Maya that isn't rendering), the newer machines will smoke the older ones.
 
  07 July 2013
Originally Posted by cgbeige: The only thing is that I doubt you'll see dual-CPU machines for another few months after that due to scarcity. Nevertheless, a single-CPU Ivy Bridge Xeon machine at 2.7GHz and 12-cores will likely perform about the same as a dual 8-core Sandy Bridge Xeon at 2.4GHz and cost a lot less than the dual. For tasks that are poorly threaded (Photoshop, most video, almost anything in Maya that isn't rendering), the newer machines will smoke the older ones.



yeah, I agree
 
Thread Closed share thread



Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

vB code is On
Smilies are On
[IMG] code is On
HTML code is Off
CGSociety
Society of Digital Artists
www.cgsociety.org

Powered by vBulletin
Copyright 2000 - 2006,
Jelsoft Enterprises Ltd.
Minimize Ads
Forum Jump
Miscellaneous

All times are GMT. The time now is 06:09 AM.


Powered by vBulletin
Copyright ©2000 - 2017, Jelsoft Enterprises Ltd.