Community
Inventor Forum
Welcome to Autodesk’s Inventor Forums. Share your knowledge, ask questions, and explore popular Inventor topics.
cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 

Quad core performance question

20 REPLIES 20
SOLVED
Reply
Message 1 of 21
cbenner
3368 Views, 20 Replies

Quad core performance question

Hey,

 

We just started using our new CAD stationa a couple of weeks ago, and I'm noticing what appears to be a disturbing trend.  These (I am told by IT) are Quad Core - 8 Thread systems.... I have no idea what that really means.  I do know that as of IV 2013, IV is still only using one processor for most if not all functionality.  But with each core split into two threads, does this mean IV is only using one thread? 

 

Here is what I noticed besides slow performance.  On our old machines, when we were really pushing the limits with some of our largest designs, i could look at Task Manager's "Performance" tab.  It showed the 4 CPU's... and the one that was doing all of the work was, naturally, spiking toward the top.  The top being 25% CPU usage.  If we hit that magic 25% number.... usually that meant we were dead in the water.  With these new systems, the same screen shows 8 CPUs, with one doing all of the work and spiking towrd the top... at 13% CPU usage.  To me this means that we are going to crash a lot sooner than before.

 

Anyone here have any insights into these processors, and how they behave with Inventor?  Here are the Specs:

 

Win 7 x64, Dell Precision T7600, Intel(R) Xeon(R) CPU E5-2643@3.30GHz, 32 GB RAM; GeForce GTX580; 3DConnexion Wireless Space Mouse, Penguin Vertical Mouse

 

TIA!

20 REPLIES 20
Message 2 of 21
MikahB
in reply to: cbenner

I'm not an expert on CPU's by any means, but I think what you're seeing (13% now versus 25% previously) is the same indication of the same issue.  Inventor is single-threaded for all intents and purposes, so it can only utilize 1 CPU at any given time.  Since before 1 CPU was 1/4 of your CPU total capacity it was 25%, now it's 1/8 = 12.5% ~ 13%.  So long as the new CPU's are faster in speed, you should see a performance increase.  More directly to your question, I would not expect the lower % number is any indication you'll have more issues now.  If your new CPU's are faster than the previous ones, that should be a net gain.

 

About the only time I see 100% usage on any core with Inventor is during rendering.  Are you seeing it regularly?

Mikah Barnett
All Angles Design
Product Design Suite Ultimate 2014
Windows 7 Professional x64
Intel i7-3770k @ 4.5GHz
32GB DDR3-2400 RAM
GeForce GTX 670 4GB
Message 3 of 21
cbenner
in reply to: MikahB

So far today I have been seeing it all morning.  I am updating to the latest service packs and updates etc... to see if that makes a difference.

 

And I might add that the assembly I am working on is by no means a large one... one of our smaller ones in fact.  I should not be pushing a CPU to the max for any reason.... gotta be something going on here.

Message 4 of 21
mrattray
in reply to: cbenner

I'm running the exact same processor as you. Actually, my system specs are nearly identical to yours except I have an AMD GPU.
I'm far from an expert, but when I was doing a ton of FEA work (using a special electromagnetic add-in, not native Inventor FEA) I found the same behaviour in task manager as what you're describing. As an experiment I disabled hyper-threading (the feature that causes your quad core to behave like an 8-core) in the system BIOS. I saw no appreciable difference in solve times and ended up switching it back on.
My understanding is that these new processors are "smart" enough to know when it's appropriate to "split" processing power and when it's not and adjust themselves to suit.
If you're really curious do some googling. You'll give yourself a headache trying to understand this stuff!
Mike (not Matt) Rattray

Message 5 of 21
MikahB
in reply to: cbenner

Agreed, that doesn't sound right. Let us know if the updates help at all.
Mikah Barnett
All Angles Design
Product Design Suite Ultimate 2014
Windows 7 Professional x64
Intel i7-3770k @ 4.5GHz
32GB DDR3-2400 RAM
GeForce GTX 670 4GB
Message 6 of 21
swalton
in reply to: cbenner

IV is slowly becoming more multi-core/thread aware.  Idw rendering or FEA, for example.  Some of the graphics system in IV 2014 is too.

 

However, lots of IV is still single-threaded.  That means that IV must calculate each modeling feature or other program step one-after-another.  

 

Your task manager is showing you that IV is running a complex single thread.  Your processor has 4 two-thread cores for a total of 8 execution units, 100%/8 is 12.5%, round up to 13%. 

 

As long as you are not hitting any other system bottlenecks, and the new processor has a faster single-thread performance than your older processor, you should spend less time waiting on IV and more time designing.  Check http://www.cpubenchmark.net/singleThread.html for some benchmarks. 

 

Steve Walton
Did you find this post helpful? Feel free to Like this post.
Did your question get successfully answered? Then click on the ACCEPT SOLUTION button.

EESignature


Inventor 2023
Vault Professional 2023
Message 7 of 21
mrattray
in reply to: swalton

You guys all slipped in while I was typing. I must be slow!
Swalton, you bring up an important point: that FEA add-in I was benchmarking with is 100% single threaded.
Also, as Swalton and Mikah mentioned: as long as the clock speed is higher you should be running faster. It's a myth that the total clock speed is split among the cores; more is better but clock speed is more important than cores for some applications (including modeling in Inventor).
Mike (not Matt) Rattray

Message 8 of 21
swalton
in reply to: mrattray

Yeah, there were no responses when I started typing...

 

Its not just processor clock speed, but also how clock ticks it takes to run a segment of code.  If two cpus have the same clock speed, and the generation 1 cpu takes 100 ticks to run a segment of Inventor code and the generation 2 takes 75, it will take 25% less absolute time to run the task on the gen 2 processor. 

 

Like was said upthread, you can confuse yourself with lots of details from online research.Smiley Wink

Steve Walton
Did you find this post helpful? Feel free to Like this post.
Did your question get successfully answered? Then click on the ACCEPT SOLUTION button.

EESignature


Inventor 2023
Vault Professional 2023
Message 9 of 21
cbenner
in reply to: mrattray

IT has been looking at this as well, and also came up with the hyper threading question.  We're going to look into that.  Of course this whole thing could be very simply a case of Monday.

 

Thanks for the input guys!

Message 10 of 21
pcrawley
in reply to: cbenner

There are lots of references to hyperthreading and hardware setups.  Until recently I would have disagreed with anyone who suggested turning it off, but I have now seen first hand the effects, so I've reconsidered!  

 

I wish there was more information about this from Autodesk.  These are a couple of useful references on the subject;

 

This one: http://www.cadlinecommunity.co.uk/Blogs/Blog.aspx?ScoId=60dd95bf-8cad-4857-991c-f851d3583f45

And you'll find the link in the story doesn't work any more because Autodesk have changed all their tech-doc links... Grrr!

 

The new reference is here: http://knowledge.autodesk.com/support/inventor-products/learn-explore/caas/CloudHelp/cloudhelp/2014/...

 

Peter
Message 11 of 21
mrattray
in reply to: pcrawley

Thanks for the link, Peter. There's a lot of interesting tech info there.
Mike (not Matt) Rattray

Message 12 of 21
brian.cranston
in reply to: cbenner

What are you doing in Inventor when it runs a single core at full speed? That often happens to me and it's not necessarily an indicator of a problem.

 

As was mentioned already, Inventor is generally a single threaded application with a few exceptions.  To borrow from the car world, there's no substitution for Gigahertz! I'm partial to the E5-1650 myself. It starts at 3.2 GHz but can turbo boost to 3.8 GHz; a bit better than the 2643's max 3.5 GHz.  Plus it's a 6-core, which helps if you have multithreaded apps.  It's also cheaper than a E5-2643.

 

If you feel that your system is slower than it should be, here's some ideas...

 

-Check that your hard drive is plugged into the fastest SATA port available.  I've recently found some 6 Gbit 10K RPM drives plugged into the 3 Gbit port on the motherboards. Although, it probably won't make any appreciable difference in overall performance.

 

-See what speed RAM shipped.  That CPU is capable of using DDR3-1600.  Sometimes vendors install slower ram and that can be a huge bottleneck during intensive CPU activity.

 

-The CPU is also capable of quad channel ram configuration.  That means the best performance is achieved by installing ram modules in groups of four in the same color ram sockets.  Not doing this can reduce performance, but not as much as using slower clock speed ram.

 

Some reference material...

http://ark.intel.com/products/64587

 

http://ark.intel.com/products/64601

 

http://en.wikipedia.org/wiki/List_of_Intel_Xeon_microprocessors

 

-Brian Cranston

 

Message 13 of 21
pcrawley
in reply to: brian.cranston

Just seen the effects again.  An Inventor IDW of a 33,000 part assembly:  HT = on, drawing opened in 6.5 minutes.  HT = off, drawing opens in about 2 minutes.

 

I would love to know why HT makes such a difference if anyone has any facts? (Opinion is OK as long as it is substantiated!)

Peter
Message 14 of 21
mrattray
in reply to: pcrawley

That's not what I would expect to see. I'll do some experiments on my end and see if I come up with the same results.
Mike (not Matt) Rattray

Message 15 of 21
dgorsman
in reply to: pcrawley

I'm assuming this is repeatable both ways, so you aren't dealing with data being cached or similar effects?

 

Hyperthreading deals with "virtual" cores rather than physical ones, and gets handled with code rather than hardware.  If the calling software isn't optimized for multi-threading then you could be dealing with the overhead of spooling up and reconciling not only different threads between cores but between the virtual cores as well.  With hyperthreading turned off, any miscellaneous threads are being forced to a different physical core rather than sharing a virtual one.

 

Ah, for the days with an 8088 and an LED probe...

----------------------------------
If you are going to fly by the seat of your pants, expect friction burns.
"I don't know" is the beginning of knowledge, not the end.


Message 16 of 21
pcrawley
in reply to: dgorsman

No caching.  The advantage of testing HT on/off is that it's changed in BIOS which forces a reboot between tests.

The tests were on the same drawing and followed exactly the same procedure, so repeatable... Yep.

 

The actual numbers were 6.5 minutes with HT on, and 2.5 minutes with HT off.  4 minutes saved per file, per number of open's, per user, per day...  Adds up really fast!  Just going to do some more testing on view-creation and update times.

Peter
Message 17 of 21
blair
in reply to: mrattray

Interesting, I never tested IV with Hyperthreading. I know that the lV2014 did make use of multi cores in the IDW environment in addition to Stdio and FEA. Will have to try some tests when I get back from holidays.

Inventor 2020, In-Cad, Simulation Mechanical

Just insert the picture rather than attaching it as a file
Did you find this reply helpful ? If so please use the Accept as Solution or Kudos button below.
Delta Tau Chi ΔΤΧ

Message 18 of 21
blair
in reply to: cbenner

Just turn on the Turbo-boost and quite complaining about the cold. 😃

Inventor 2020, In-Cad, Simulation Mechanical

Just insert the picture rather than attaching it as a file
Did you find this reply helpful ? If so please use the Accept as Solution or Kudos button below.
Delta Tau Chi ΔΤΧ

Message 19 of 21
pcrawley
in reply to: blair

Ha ha!  I tend to render things in 3ds max if I need to warm the room up a bit!

Enjoy your holiday.

Peter
Message 20 of 21
mrattray
in reply to: pcrawley

I conducted an experiment this morning. I'd like to do more testing with different scenarios, but I have to do SOME real work, too.

 

HT On:

  Drawing A:

    Open: 39s

    Update all sheets: 16s

  Drawing B:

    Open: 33s

    Update all sheets: 27s

HT Off:

  Drawing A:

    Open: 27s

    Update all sheets: 15s

  Drawing B:

    Open: 23s

    Update all sheets: 28s

 

These were both multi-sheet drawings with about 65-80 sheets of an assembly and details with around 1000-1500 instances of 120-160 unique parts. Note that we have approx. 30% faster opening times with HT off, but no appreciable change to the update time.

I'm going to work with HT off for a while and try to get a seat-of-the-pants feel until I can get around to doing more testing.

Mike (not Matt) Rattray

Can't find what you're looking for? Ask the community or share your knowledge.

Post to forums  

Autodesk Design & Make Report