Does Hyperthreading (HT) turned on improve things?

[Trixter]

I searched the forum but didn't find an answer: I have the opportunity to upgrade my ailing P4 system to a CPU that has hyperthreading capability. Does anyone know if turning HT on affects CCE encoding speed? If so, does it make it better or worse?

[Boulder]

It should improve the performance as CCE does use multithreading when possible.

[Fishman0919]

CCE does not support HTT

http://www.cinemacraft.com/eng/faq_offline.html#01

I want to say... and maybe it just me but it seems a bit slow w/HT on.

[artoor]

I confirm what Boulder has said above.
I've got P4 2.4HT. While CCE encoding when I turn one (logical) processor off in "Set Affinity" encoding is slowing down. I can't tell You how much but it is noticeable value. I guess it could better in case of 2Cores CPU :)

[Boulder]

When you encode something, the CPU usage for CCE goes above 50% so CCE uses both virtual CPUs. Usually the usage hovers somewhere around 50-65%.

Then there's also tsp's multithreaded Avisynth build which is really handy.

[artoor]

When you encode something, the CPU usage for CCE goes above 50% so CCE uses both virtual CPUs. Usually the usage hovers somewhere around 50-65%.


Frankly I notice when both logical processors are enable cce uses at about 70-80% power, when it use only one processor the value is equal 50% That's my observation :)

[Trixter]

Could someone do some tests to make sure? I don't mean "put CCE on one virtual CPU", I mean "turn off HT in the BIOS and try it". I'd be real curious to see if HT improves CCE or not -- CinemaCraft says it hurts performance, not helps.

[artoor]

I'll try to do what You wish :) Not this day because I'm busy now. Maybe tomorrow, ok?
Question - can it be done with CCE SP2 trial or CCE SP 2.7?? I assume that it doesn't have to be whole DVD just a part :)
Very welcome if there will be someone else, to comfirm my results. For the meantime - bye :)

---== EDIT ==---

Ok. I did the test and I don't know what to say. It is very strange. I'll tell You my observation, but You should know that at present I don't know what to think about it.
First of all my hardware configuration: P4 2.4HT + 768 MB Ram (that's important I guess)
Test was done after restarting my PC. All process (like antivirus, antispam, etc) were tuned off.
CCE SP2 trial

1st test - HT ENABLE
Total frame: 9738
Max Speed (1st pass): 1.85
Max Speed (2nd pass): 1.84
Total Time: 3:31

2nd test - HT DISABLE (disabled in BIOS)
Total frame: 9738
Max Speed (1st pass): 1.70
Max Speed (2nd pass): 1.72
Total Time: 3:48

Then I thought "hehe I was right" but to be honest I decided to confirm my 1st test by myself and...

3rd test - HT ENABLE
Total frame: 9738
Max Speed (1st pass): I didn't notice accurate (I've already thought it's no matter) at about 1.50-55
Max Speed (2nd pass): 1.45
Total Time: 4:30

The conditions of 1st, 2nd and 3rd test were the same, so I don't know what to think about it :( I think I can't help You to choose between HT or not HT CPU. All I know CCE supports dual processors, so it should supports multi core as well. Think about it :)

Regards!

[Trixter]

All I know CCE supports dual processors, so it should supports multi core as well. Think about it :)


Ah, but HT is not true multi-core. It's a way of simulating a second core with spare unused cycles, but it is not true multi-core. So that's why the numbers were worse on the third test.

And that answers my question: HT would not be worth the cost of upgrading (I have 2.6 now, best I could do would be 2.8 w/HT, and it's not worth the money from what I'm seeing).

Thanks for testing!

[Boulder]

I'll do some tests in a while as well.

When I chose the CPU for my current computer, I chose a HT-capable P4 because it makes working simultaneously on multiple CPU intensive applications much smoother than it used to be on my AMD processor which didn't have such capabilities. And as I said, tsp's multithreaded Avisynth build will boost performance because Avisynth is usually the bottleneck. Multithreading should be officially available in v2.6 whenever it comes out.

[Boulder]

OK, here are my results. HT enabled does make encoding faster on my system regardless of heavy or light filtering in the Avisynth script.

HT enabled, heavy filtering
Source : 100,000 seconds (2500 frames)
Elapsed: 357,687 seconds
---------------------------------------------------
>> File reading 357,014 99,812 %
>> Decoding 0,000 0,000 %
>> Resizing 0,000 0,000 %
>> Deinterlacing 0,000 0,000 %
>> RGB -> YUY2 0,000 0,000 %
---------------------------------------------------
>> MPEG encoding 0,673 0,188 %
HT disabled, heavy filtering
Source : 100,000 seconds (2500 frames)
Elapsed: 370,516 seconds
---------------------------------------------------
>> File reading 357,522 96,493 %
>> Decoding 0,000 0,000 %
>> Resizing 0,000 0,000 %
>> Deinterlacing 0,000 0,000 %
>> RGB -> YUY2 0,000 0,000 %
---------------------------------------------------
>> MPEG encoding 12,994 3,507 %
HT enabled, light filtering
Source : 200,000 seconds (5000 frames)
Elapsed: 115,765 seconds
---------------------------------------------------
>> File reading 115,090 99,417 %
>> Decoding 0,000 0,000 %
>> Resizing 0,000 0,000 %
>> Deinterlacing 0,000 0,000 %
>> RGB -> YUY2 0,000 0,000 %
---------------------------------------------------
>> MPEG encoding 0,675 0,583 %
HT disabled, light filtering
Source : 200,000 seconds (5000 frames)
Elapsed: 122,313 seconds
---------------------------------------------------
>> File reading 97,879 80,023 %
>> Decoding 0,000 0,000 %
>> Resizing 0,000 0,000 %
>> Deinterlacing 0,000 0,000 %
>> RGB -> YUY2 0,000 0,000 %
---------------------------------------------------
>> MPEG encoding 24,434 19,977 %

Notice how small the time it takes to encode to MPEG is compared to Avisynth processing :D

[Trixter]

OK, here are my results. HT enabled does make encoding faster on my system regardless of heavy or light filtering in the Avisynth script.


I appreciate the tests, but the filtering borks the caches in respect to how fast CCE can operate. Can you do an HT and no-HT test for encoding only (no filtering)?

[Boulder]

I try to find the time for some testing tomorrow.

[Boulder]

And here are the results:

HT disabled Source : 400,000 seconds (10000 frames)
Elapsed: 166,500 seconds
---------------------------------------------------
>> File reading 72,198 43,362 %
>> Decoding 0,000 0,000 %
>> Resizing 0,000 0,000 %
>> Deinterlacing 0,000 0,000 %
>> RGB -> YUY2 0,000 0,000 %
---------------------------------------------------
>> MPEG encoding 94,302 56,638 %
HT enabled Source : 400,000 seconds (10000 frames)
Elapsed: 150,750 seconds
---------------------------------------------------
>> File reading 149,495 99,167 %
>> Decoding 0,000 0,000 %
>> Resizing 0,000 0,000 %
>> Deinterlacing 0,000 0,000 %
>> RGB -> YUY2 0,000 0,000 %
---------------------------------------------------
>> MPEG encoding 1,255 0,833 %

With HT enabled it's still faster. The script only had the MPEG2Source line with upConv=1 to get YUY2 output.

[Fishman0919]

Try feeding an AVI or Quicktime file straight into CCE.... you'll see just about no diff at all.... well I did see any speed change.

[Boulder]

Depends on the decoder. Using DivX6 for AVI files should be faster on HT because the codec supports multithreading.

All in all, for my usage, HT enabled is faster.

[Trixter]

Thanks very much for the tests, I really appreciate it. It is faster, but I have to decide whether or not I want to spend $70 to upgrade my CPU for what looks like 10% more speed.

[Sharktooth]

Get a true dual core instead of a faked one...

[Trixter]

Obviously, but that means CPU+motherboard+memory since my P4 system is 3 years old.