Premiere Pro Hyperthreading Benchmark

[Kent Diamond]

I've seen lots of comments about Premiere Pro and the advantage of hyperthreading. I could not find any actual tests so I decided to do my own benchmark.

Result: Premiere Pro transcoded a video 13 percent faster with hyperthreading turned on.

Test: Transcode a 7 minutes 12 second video. Video is combination of 50 or 60 short dv segments and 10 or 15 still images. Various transitions between elements. Three video and two audio tracks are used.

Transcode by File | Export | Adobe Media Encoder
Format: MPEG2-DVD
Preset: NTSC DV High quality 7Mb CBR 1 Pass

This seemed both long enough and complex enough to be a good test.

Computer is a Dell Precision 360 workstation running at 2.8 GHz and 1 GB of RAM running Win XP Pro. Premiere Pro v7.0 was used. The computer was rebooted between each run so none of the files were in memory. The computer left alone while the test ran.

With hyperthreading turned off the encoding times were 15 minutes 59 seconds and 15 minutes and 58 seconds.

With hyperthreading turned on the encoding times were 13:52, 13:54, and 14:00. That works out to 13 percent faster.

I was surprised by the results. I thought much of the hyperthreading noise was marketing hype. In a CPU bound computation I would expect the overhead of hyperthreading would be great than the gain. I was wrong.

Disclaimer: I have done some operating system benchmarking at my job. This test was not done to professional standards. For example, I kept the network card plugged in. I did not defrag the drive between runs. Still, it's a pretty good first approximation. The consistency of the numbers lends some credibility. Repeating the tests under much greater control might change the results a few percentage points.

Anyway, I did the test and thought people might like to see the results.

[Kevin King]

Thanks Kent. That's good to see. I have a 2.6 w/ HT. Never turned HT off for a comparison. I think it especially helps when rendering in the background. I'll kick off a complex render in AE that will run an hour or so, and I can still fluidly check mail, surf, or enter my bills, and this doesn't seem to significently impact the render time.

[Khaled Chatila]

Hi,

I've been searching a long time for a community like this and its really awesome, i've learnt so much in the few days i've been reading posts here than in the last few months combined of random articles etc... real people and real experiences - its great!

Now, I have just experienced a scenario that is confusing me greatly!

I have an Intel P4 2.8Ghz machine (currently Overclocked to 3.06Gz) running smoothly for over a year now here's the specs:

CPU: see above
Motherboard: MSI GNB MAX
RAM: 2Gb
GFX: ATI X800XT


C:\ = 120Gb 7200 RPM Disk
R:\ = 4 x 250Gb 7200RPM Disks connected in RAID 5 (project files)
X:\ = 2 x 120GB 7200RPM Disks connected in RAID 0 (scratch disk)


I ran those PPBM DV tests and the numbers were ok, i've got em below:

219, secs Total Benchmark Time
45, secs AVI Encoding Time
94, secs MPEG Elapsed Time
80, secs Rendering Time

Recently, i acquired a second machine which was being upgraded and it has the following specs:

CPU: 2 x Intel Xeon 3.06 Ghz HT Enabled
MB: Supermicro P5DPE-G2
RAM: 4Gb
GFX: ATI On board (new card coming soon, this one only has 8Mb of Video RAM!!!)


C:\ = 80Gb 7200 RPM Disk
R:\ = 4 x 320Gb 7200RPM Disks connected in RAID 5 (project files)
X:\ = 80GB 7200RPM Disk (scratch disk)

Now, the PPBM test i did on this new machine had the following results:

364, secs Total Benchmark Time
118, secs AVI Encoding Time
71, secs MPEG Elapsed Time
175, secs Rendering Time

As you can see, less than ideal especially when this system is Dual Xeon CPU and the other is normal P4?! The only thing that was faster was the Mpeg creation time!

Immediately i thought that the scratch disk configuration had something to do with it...

When i transferred the preview files locations etc... to "same as project" the following were the results:

201, secs Total Benchmark Time
44, secs AVI Encoding Time
72, secs MPEG Elapsed Time
85, secs Rendering Time

Much Better, but still a long way off some of those results on that website mentioned in a previous post above!! does this look normal? and also it's only marginally faster than the single CPU system i have with half the RAM?!!


Remember the project files are on a Raid array, and in the first machine both project files and scratch are raid arrays - but still this means that even with ONE cpu, a proper disk configuration plays a HUGE part in performance and for me almost outpaces a Dual CPU system!!

Any comments most welcome!

Thanks!

[Calvin Wallace]

How do I turn on hyperthreading in premiere?

[Even Solberg]

You don't. It's a feature of your CPU. If the CPU has it (or if you have a multi-core CPU), Premiere Pro will use it. Not sure about 1.5, but 2.0 definitely will. It's using my quad core Dell just nicely.

[Mike Teutsch]

Don't believe you have to turn it on. Check as PPro boots and it will say what type of processor it is detecting. Mine says multi-processor detected. I think they are talking about tuning it off or on for the whole system, or something.

I am very happy with my simple system. Have an HP Pavilion 3.4 ghz P4 with hyper-threading, and 2 gig of ram, running Media Center 2005. When it is encoding or rendering it uses 100% of the processor, and when it is done the processor sits and idles at between 1% & 3%, usually 1 or 2.

Mike

[Khaled Chatila]

Calvin,

I have two machines with the Hyperthreading feature, and both have options in the BIOS to enable or disable hyperthreading.

So if it's not on by default, and you dont see two processors in your task manager (obviously when you only have one) then you may need to enable it since some of the first P4 processors were not capable of being used in a hyperthreading setup.

[Ray Bell]

Hey, I'm looking to get a Dell 360 workstation...

any idea's on how to configure it... anything you guys would suggest??

[Steve Oakley]

not all apps benifit from hyper threading. in fact some take a serious performance hit. one example is invigorator. with HT ON, it only uses 50% of the CPU cycles, turn it off, it uses 100% cutting render time in half. likewise plugins running in PP may exhibit similar perfromance. the only way to know is to try.

I've been running with HT off. one thing I notice is the desktop responds differently. While some tasks run faster, its also harder to interupt or do a second task at the same time. YMMV

Steve Oakley

[Mikko Lopponen]

Originally Posted by Steve Oakley

not all apps benifit from hyper threading. in fact some take a serious performance hit. one example is invigorator. with HT ON, it only uses 50% of the CPU cycles, turn it off, it uses 100% cutting render time in half.

Eh? 50% with ht on is the same thing as 100% with it off. Atleast with dual cores.