Hyper Threading + LW

[ldgree]

Hey all,

I buy a new computer here..... with HT feature... but before i install the new computer.. i create a test scene in old PC to create one Benchmark.. and look this

In my old computer - P4 .2.0 512mb ram - geforce 4 the test render :
13minutes 55 seconds

And in my new... - P4 3.2 1gb ram - geforce 6 the test render :
16m 34 seconds...


this is weird dont you think? a faster machine very slowler in render?!?! :((

Anyone has some tip? i need adjust something??

Thanks,

LD

[Nathonix]

well ive found out that your video card does nothing in the rendering phase, only in your viewports. but i think to take advantage of hyperthreading, you'd have to split lightwave into two processes, which i belive is done through the hub.

[JamesCurtis]

In order for Hyperthreading to work, it must be enabled in the bios, I believe, and you must have 2 memory chips in place to utilize it [2 x 256, 2 x 512] and then Lightwave itself must have Multithreading set to 2 or above. Of course you also need to be using a P4 processor with Hyperthreading capability.

[jasonazure]

I have some experience with network rendering on Hyper threading PCs, and I have come across some weird stuff - mainly that hypervoxel heavy frames render a **** of a lot slower on HT processors than on single non-HT processors !!!

I have actually changed processor affinity during a LWSN render and seen an immediate speed-up.

[ldgree]

Thanks for the answers ^^ :thumbsup:


My HyperT. is active..... i have 2x512mb in different slots here...
its all ok, in Render Options i change to 2 threads too

But its a slowler than one without HT feature, i think when i need render something i will turn off HT feature in Bios.. this is the only way i think


Thanks Again :)

LD

[hstewarth]

This is interesting, it would be really interesting to see what the results are wth Bios with HT off. But make sure that you test it in the same way. With same # of render threads.

Also it may depend on version of Lightwave, Older version might not work well with HT Machines.

[Nathonix]

thats a good point, also, hyperthreading is not the same as having a dual core. it was meant for multitasking, by spliting the processor's work into two "virtual" cores, which i belive would impact your performance. could be wrong, im an amd man usually, and i have a celeron in my lightwave pc, so this is all what ive gathered over the years.

[hstewarth]

Well I don't believe its entirely Hyperthreading fault here..

I did a test with Mojoworld with does not internally use multiple processors. I basically render the same scene 6 times with 2 seperate processors concurrently in the same time as 4 process running right behind each other. Basically Hyperthreading gave about 50% improvement in time but only if you run with 2 processors with this application.

I believe that hyperthreading will speed up render time if it is propertly used. For Lightwave, a lot depends on how it is written. I would assume that two many threads in lightwave could be possible slow things down .. but I am not sure on that.

I would also think that lot else depends on what is installed on computer.

I will give it test run tonight.. Just to see what happens.. My test
is with 3.2Ghz P4 with 2Gig of memory and 6800GT with
the following tests on it - with same scene

1. HT On as normal
2. HT On with affinite on Lightwave.
3. HT Off in Bios.

With single and dual threads in Lightwave.

Of course this may depend on what you rendering.. Some operations like Surface Baking don't work with multiple threads.

[hstewarth]

Ok I did the test on my 3.2Ghz P4 with Lightwave 8.5 (32bit)

Here are the results - with TeaPot Benchmark scene supply with Lightwave

Hyperthreading ON

1. Thread 138.2 -22.4% on 2 Thread HT
2. Threads 112.9 0.0%
8 Threads 185.9 -64.7% ( just for fun )

Hyperthreading ON affinity set to CPU 0

1. Thread 140.9 -24.8%
2. Threads 148.2 -31.3%

Hyperthreading OFF.

1. Thread 118.6 -5.0%
2. Threads 138.5 -22.6%


My conclusion is that Hyperthreading is better. Dual core is like to be a lot better but this limited test at least shows it is better.

Also in my experience here setting Affinity to CPU 0 on my test, show worst results at all. Also interesting to note that HT with 2 threads is only slightly better than Non HT with 1 thread and also HT with 1 thread is similar to non HT with 2 threads.

The # of threads appears tio have an impact in rendering time, 8 threads was signfiantly worst than 2 threads - did not even care to test it in other tests.- so it is obvious there is some overhead in threads. I would assume that 8 threads is only good on Quad HT or 8 Non HT machine.

I have a feeling that depending on the scenes, the results will differ. But I find the advantages of Hyperthreading outway these because of better multitasking experence.

Hope this help - like to see other results.

On a side note: If you using NVidia card, I would recommend getting 8x.xx series drivers - these drivers are suppose to be multi-threaded and could have a performance impact performance.