Encoding Sequence... starts, stops, starts, stops, starts... etc...

[John Hewat]

This is a strange.

Today I told Premiere CS5 to export a 2 hour 30-40Mbps 1920x1080 mp4 from my Cineform project.

It took about 5 hours.

Maybe that's about expected. Maybe not.

It seemed to take longer than it should have primarily because the encoding would start and the CPU usage would jump up to 70% or so and then they would suddenly drop to 0% and the encoding would stop.

Then it would repeat for the next five hours until the encoding was done. The final product was just fine when tested afterwards.

So my questions are:

1. Why would the CPUs suddenly stop and start like that? No other programs running, still 12GB of memory not being used (CS5 was using only 2.5GB out of 20 the whole time), 3 TBs of free HDD space on the RAID 0 drive being used.

2. As I watched the temperatures in CPUID Hardware Monitor, the CPUs were getting as high as 66 degrees celcius when the encoding was happening and then, ten seconds later, when they would stop, the temperatures dropped down to 35 degrees. Ten seconds later, at 70% use again, the temps were at 60 degrees again. Surely this is a terrible scenario for a CPU to be in? Is it going to have caused damage? Could the temperature inccrease be the cause of the encode stopping? It just seems strange to me...

The CPUs in question are 2 x Xeon 5450s quad cores at 3GHz each on a Supermicro X7DWA-N. And the temperature increase was only that significant on one of the physical CPUs (all four cores though) - in fact, the maximum temperature of the other core never got above 38 degrees.

[Harm Millaard]

Hard to answer with these details.

Can you start ResMon by typing in Resmon in the Search programs and files field of the Start menu and immediately after the encode, make a screenshot of it and post it here. Also make a screenshot during the encoding of Process Explorer with the 'View System Information' panel open. Both screenshots will tell a lot more than HWMonitor.

[John Hewat]

Hi Harm,

I've attached the screenshot during the encode - I included both programs as well as info about the encode in case it's relevant.

Thanks once again for your help.

If I can stay up til 2am I'll give you the other screenshot at the end but if not, I'll try it again during daylight hours.

Regards,
-- John.

[Harm Millaard]

John,

Don't worry about taking a screenshot tonight. This info is great, I'll study it and get back to you. Meanwhile, have a good nights rest, after you have done me one more favor. In Process Explorer and System Information, tick the bottom left checkbox to show one graph per CPU and post a screen shot. I have the feeling you may be in for a surprise, but again, I may be mistaken.

I'll get back to you ASAP.

[John Hewat]

Sorry Harm, I was out cold by the time I got that next response so I didn't get a chance to do so. I'm off to work at my real job now and when I come home I'll do the export again and get those next two screen shots.

For what it's worth though, I can confirm from having watched the Performance tab in Task Manager, that all cores of both processes were at 0% at the same time. Is that what you were looking for by seeing each graph?

Anyway, I'll get te screenshots for you this afternoon.

Thanks again,
-- John.

[John Hewat]

Ok - I got home from work, ignored my girlfriend and came straight into my computer room and started the encode again.

A few things (that may be) worth noting:

1. When I turned the PC on, the CPU fans (Noctuas) were making an unusually loud noise. This stopped when Windows booted. This has never happened before.
2. The noise of the fans also subsided slightly when I applied pressure to them. As in, I actually grabbed the fan and forced it a little more toward the heatsink. I'll attach a photo to explain why I did that.
3. This time, I reduced the 1 hour 50 minute work area to just 10 minutes for the sake of this experiment.

So, here's the attached screenshots.

One from during the encode, the second taken mimediately after - but before the CS5 Encoding Sequence window had a chance to close (I got paranoid because I wasn't sure how immediately afterward I should take the shot).

The third shot is a photo of my heatsink & fan setup. I am worried that the root of my problem could have something to do with my lousy install of either: a) thermal paste, or b) the heatsink.

As is kind of obvious from the picture, the two heatsinks are not lined up perfectly - the fan that sits between them clearly fits snuggly into one of the heatsinks, but a gap of a few mm is left between it and the other heatsink.

Now these heatsink/fan combos are designed to be used on dual CPU setups and I've seen pictures of the whole outfit sitting snugly together. So given that mine do not, could I have screwed up the installation of one of them, which could explain why one CPU is always hotter than the other? But, that said, the top heatsink is always hotter to the touch too, so I don't know. I kind of assumed that was because of the upwards flow of air.

Thanks again,

-- John.

P.S. - I'm dying to know what the surprise is - I'm excited but terrified at the same time...

[Harm Millaard]

John,

I'm not familiar with Matrox Power Desk, but could it be that that application is doing something weird? You could try to either Kill Process Tree or Suspend these processes and see if it makes any difference.

The disk activity is too low to be the cause of this and apart from this Matrox PD, your system looks organized and clean / lean.

You are using the Export button in PR and not Adobe Media Encoder and the queue. I have found that sometimes Export takes nearly three times as long as when using AME.

If you try AME, also try the following:

In Process Explorer, right click on Adobe Media Encoder.exe and set the priority to High (13). Do the same with dynamiclinkmanager.exe. Then start the queue.

[John Hewat]

Thanks Harm,

Tried it through AME but same thing: starting, stopping until complete. I exported the exact same 10 minute section and the export was a little quicker.

Changing the priority didn't seem to have any detectable effect. I also tried the export again from within Premiere, changing its priority to High 13 as well, but still, no noticable change.

Matrox Power Desk is the software that handles my display setup for the Triple Head2Go - I use it to spread my editing space over two or three moniors and use the second output from my GPU to feed the preview monitor. I was hesitant to kill the process fearing I would lose my displays - but I did so, and nothing happened to the displays... but there was no affect on the encoding process: still starting and stopping...

Do you have any comment on my install of the heatsinks on my CPUs - could the heat be too much and be prompting the slowing down of the CPU? And if so, could my installation of the heatsinks be the root cause of all of this?

Out of it all, should I be worried about the rapid temperature increase and decrease of my CPUs? These changes of 30 degrees are sometimes happening within a 20 - 30 second period.

[Harm Millaard]

John,

Looking at the max temps on both CPU's, it appears that the second 5450 is not properly cooled, and that can be the explanation for the extra noise during boot. It may be vibrating and one of the attachment screws may have come loose a bit. It is easy for you to check whether both coolers have the same gap between the cooling body and fan. I have the U12, so that is different.

But the real surprise is the relatively long periods of inactivity with CPU load at 0%, but all 8 cores fully operational and no visible explanation for it. Even the max temps on the 2-nd 5450 are a long way off from the TDP, so there is no throttling down.

During encoding there is no hardware assisted MPE activity, so that factor should be ruled out.

Maybe this is helpful: Case of the Unexplained 3 :: 2009 :: North America :: Microsoft Tech·Ed

[John Hewat]

Thanks Harm, mine are te U12(DX) also. I don't quite follow, where you've said "it's easy to check whether both coolers have the same gap..." do you mean that the gap between the coolers and the fans is different to the gap between the coolers and the CPU itself? I get that part, but I suppose that my only choice is to remove them and reapply the thermal paste and reinsert?

Maybe I should join the fans and heatsinks together before I apply them to the CPUs? I don't really know how best to do it. It was a fairly simple installation, or so I thought...

I'll reinstall them tomorrow after work and see how I go.

It's my least favourite part of building a pc; I still don't know whether to spread the paste or just put a small dollop. Why are there different opinions on this???

Thanks again Harm. I'll use that site to try to see if he can narrow down the problem. And I'll reinstall the coolers. Despite having no real knowledge on the matter, part of me is hoping that by simply fixing the installation of the coolers, my CPU starting andstopping problem might go away. You don't think they're related?

[David Chilson]

Here is a good pdf on applying paste for your processor. It looks like you should take the time to reseat it.

http://www.arcticsilver.com/pdf/appm..._line_v1.1.pdf

I had a start/stop problem before and it was a hard drive that that was going bad. You wrote about a raid but I am guessing you use another drive for program files. I don't think it will be as simple as reseating your coolers, but one can hope! After you fix the cooler I would back everything up. Good luck!

[Steve Kalle]

Hey John,

Who told you that those Noctua's are good for a dual-CPU setup? Just by looking at the photo, I can easily see that there is no room for air to be pulled thru the top cooler, which is causing the high temps. If you can, remove the top fan from the bottom cooler, which should give the top cooler some room to pull air. You probably will see a 3-5 degree increase under load on the bottom cpu but the top cpu should be much cooler

For thermal paste, this is what I have always done: I cut a piece of plastic the size of the cpu (the top part where it contacts the cooler). This gives me a nice flat surface to spread the paste, which I always spread on the cpu itself. I cover almost the entire cpu but keep 1-2mm from the edges clean so the paste has somewhere to go when the heatsink is secured onto it. I squeeze a small amount onto the cpu and use the piece of plastic to spread it around. I make a square and then start scraping away/thinning out the layer of paste and use some paper towel to wipe the excess paste from the piece of plastic. When I scrape, I go from one side to the other. I can tell when there is a good amount of paste on the cpu when only a very small amount of paste builds up on the plastic.

And someone correct me if I'm wrong but I think it takes a day or two for the paste to 'settle' in and provide maximum heat transfer.

Hope this helps.

Questions:
1) does this start - stop problem occur with other projects?
2) can you try moving all data to a single drive (source video, pppro project file, media cache, export) and try again. This might help to single out hard drive problems.

[John Hewat]

Hello all,

I haven't had a chance to re-attach the fans yet, but thought I'd respond to a couple fo points:

1. The Noctuas are designed (aren't they) to be put in sequence like that to send air through both heatsinks in the same direction over dual CPU sockets. I've seen plenty of reviews of them that show them set up like that with one fan on either side and one fan in between. Harm, if you have them too, how do you set them up? The manual that comes with them specifies that the fan should be pushing air over the heatsink, which means that middle fan is in fact required as far as Noctua are concerned.

2. I haven't tried moving the files required to another drive yet, but I do have all files, project, scratch and output on the RAID drive at the moment. I am dealing with a 60GB Cineform AVI so I kept thinking, maybe it's just too big... Maybe next time I try, I'll move it all over to my C: drive and try again. My RAID has always been... questionable. Running HD Tach, I could never get as good results as I thought, despite having a hardware setup.

[Harm Millaard]

John,

I only have a single CPU, so not directly comparable.

In your case make sure that the orientation of all the fans is the same, so that the labeled side on all three point in the same direction, or alternatively, change the coolers direction by 90 degrees so they are side by side with the fans, instead of in a row. That will allow 4 fans to be used in a push-pull configuration.

The cooling issue is not likely related to the stopping-starting, but may help to improve your temps.

You may have done this before, but in that case I have forgotten, what were your problems with the raid? Please give as much details as possible.

[John Hewat]

I have a feeling the coolers can only go in one way... I could be wrong about that. They are currently all going in the same direction. When I get home tonight, I'll re-attach them and try turning them 90 degrees. That might also cool the RAM, as there is no airflow over them other than a fan that I custom mounted to keep the sticks from getting to 120 degrees C.

I can also confirm that the stop, start issue is not happening on other projects - or at least one other project. Exporting a different 1 hour 40 EX1 HQ 25i footage as PAL DV 25i results in uninterrupted Premiere export with CPUs constantly at 99 - 100%. It was the most beautiful thing I've ever seen.

Harm, this thread here is where I documented my problems trying to improve my PPBM score. You helped me immensely throughout the process and if you give it a quick look, from about post #6 onwards you'll see that you thought that my RAID was not performing as you would expect. By post #34 I think, I discuess the effect of the hardware RAID and I am still getting average results, even with my Highpoint 2640x4 (which device manager sees as a SAS controller rather that a SATA one I just realized... is that bad?) and 4x1TB drives.