Solution to Vegas crashes on modern high end CPU’s (e.g. i9 13900K)

bitman wrote on 12/25/2023, 4:04 AM

First merry Christmas to you all!

We all know Vegas crashes from time to time, however lately, for me it is on the increase, which I found strange, given the good stability of the program the last couple of years. So, I started to investigate and found at least one reason, and a solution to it:

Vegas has difficulties surviving CPU Thermal Throttling of one or more of its cores.

Reproducible on Vegas 19, 20 and 21. I have recently discovered that I can work for hours without a crash using the most complex edits and rendering and yet I can make it crash with the simplest project and CPU only rendering. So, what is going on? Let me list an example of the conditions for a crash:

Rendering of a 4K file into FHD (no edits, just a HD crop via pan/crop) using the standard Magix AVC/AAC MP4 rendering (without the use of Intel QSV or NVENC).

I traced the issue down to thermal throttling of the CPU via the free tool HWiNFO64.

Now I know you are all saying not enough cooling, right? I do have one of the best air coolers on the market and have a stable high-end PC I build myself. Demanding games play without an issue. So why I did not have this issue before? I realized I keep my PC up to date; then I remembered I installed a recent new BIOS for my Gigabyte Z690 Aorus Master motherboard (F27). I know my processor (i9 13900K) is a notorious hot-head and can draw insane amounts of current. To my horror I discovered that the default motherboard setting in the BIOS for TDPL1 (sustained power limit) and TDPL2 (turbo power limit) was both set to “AUTO” (like most settings), but I found out that AUTO setting meant “indefinite power”. Reading it out in HWiNFO it said that power limit to something like 4095 Watts. Also, the multicore enhancement in default “AUTO” was active.

So, what does this mean in practice: if one or more cores get to 100 degrees Celsius, they throttle back (another setting in BIOS) to protect the cores. This is what happened on my PC, and when this happens, Vegas crashes almost always. Very reproducible. The thing is, however, other programs when stressing the CPU causing thermal throttling like Cinebench 2024, do not crash.

Conclusion, Vegas cannot handle thermal throttling of the CPU cores, something the developers should investigate... @VEGASDerek . It would help to avoid many Vegas crashes on modern systems, not to mention the useless crash reports send to the developers wasting their time to investigate.

As the title suggests, I do have a solution if you have the same issue: go into BIOS of your motherboard and set the power limits of your CPU to what your CPU cooler can handle:

I give an example for my case: the default on my motherboard was set to AUTO and meant indefinite (shame on you Gigabyte), for starters this is a breach to Intel’s own maximum power limits, the Intel specs for i9 13900K states 2 possible scenarios:

normal config: PL1=125W, PL2=253W (with max duration of 56 seconds)
extreme config: PL1= 253W, PL2=253W (with max duration of 56 seconds)

By experimenting I found (on my system) that running Vegas CPU only rendering at 100% multi core load, I can keep the core temperatures around 89 degrees Celsius when manually imposing a PL1=200W and PL2=200W power limit in BIOS with my cooling solution (Noctua NH-D15s), thus avoiding thermal throttling at 100 degree Celsius. You must establish your own maximum power limit based on experiment as your cooling solution (including ambient and case fans) and CPU type differ from me, and could be better or worse than the CPU’s own spec.

There are other tricks you may use, such as disabling multicore enhancement (MCE), and “Undervolting”, but I do not recommend this. Especially “Undervolting the CPU” praised on the internet as the new holy grail to combat CPU temperature, is tricky and very quickly leads to an unstable system, especially when running idle (the graph required voltage/frequency and margin for tuning is not linear, not to mention the silicon lottery - not everyone has a golden sample).

And before someone would suggest replacing one of the best CPU Air coolers by water cooling, it will not help if there is no power limit imposed, the small surface area of INTEL CPU’s can only transfer so much heat, water cooling will just be marginally better and needs to be a triple fan and big radiator to even beat the Noctua. I could add a second fan to the Noctua, but I predict it will only nock of 3-4 degrees extra, so this will not help.

So, is this a Gigabyte and INTEL only problem? No, they all do it, ASUS, name it, and the issue is not limited to INTEL alone, AMD has similar issues. Recently, AMD processers were melting due to aggressive motherboard settings, they had to bring out new BIOS to prevent this.

Why they do it? Competition, benchmark reviews, they all count on thermal throttling to the rescue.

The victim is Vegas who rarely survives thermal throttling. The bad news is that most people do not even know that their system is thermal throttling due to overzealous motherboard BIOS defaults (driven in turn by competition) and wonder why Vegas crashes…

Comments

andyrpsmith wrote on 12/25/2023, 4:30 AM

Yes everyone needs to be aware of the Bios Auto default setting for managing CPU power. I set mine to Intel default max power for the 13900K, 253W I think, otherwise it ran to 352W. Also remember that XMP settings for the memory is an overclock as far as the CPU is concerned and can cause instability with some applications. (Starfield game hated it and need it lowered from 6800MHz to 6400MHz to be stable.). I do run water cooling with my CPU and have controlled any thermal throttling, no crashes with V20 yet. Many thanks for the info. I also found under volt-ing my 1080Ti GPU from 1.2V to around 1.0V helped stability (in starfield) and reduced temps.

john_dennis wrote on 12/25/2023, 10:12 AM

@bitman

Some bios screenshots might be useful for people that don't do this stuff often.

I haven't done a bios update since I built my system so ASUS hasn't had a chance to destroy my machine to get a few more cycles in a benchmark, yet.

My long-term power limit was 253 Watt on water. System Upgrade 2023

I throttle on individual core temp. Originally, I set the value to 65 degrees C for all cores, (peak performance be damned, I want it running when I get back from my walk). I've since raised the number to 70 degrees C.

Are you asserting that Vegas fails when the machine throttles on PMax and/or Power Limit?

Note

It's probably going to be three years before I go back into my bios.

bitman wrote on 12/25/2023, 3:19 PM

@john_dennis I leave the core temperature limit the default to the TJ max of 100 degrees Celsius. Limiting the core temperatures like you do to 65 degrees without setting proper power limits will lead to throttling if, and only if your cooler cannot dissipate the heat beyond 65 degrees C when all cores are 100% busy all of the time. This can happen during CPU multicore benchmarking or if the GPU is not involved in the rendering or perhaps edit preview. Also do not forget there are always background programs running as well.

I may not mentioned it in the post above, but I see the thermal throttling mainly on the higher clocked P-cores, not on the E-cores.

Are you asserting that Vegas fails when the machine throttles on PMax and/or Power Limit?

No, Vegas does not crash if the set power limits are reached and the CPU throttles down because of that. Vegas crashes only (pretty much always) if the CPU throttles because of the core maximum temperature is reached.

For example on my PC with my cooling, Vegas crashes when I put the long term power limit to 210W (and the turbo to 210), as it creeps to 94-100 degree C under 100% CPU load. Setting the power limit to 200W long term and turbo to 200W I can keep the temperature around 89 degrees C under 100% full load, keeping well off 100 degrees, and no thermal throttling occurs. Obviously it is different for everyone.

bitman wrote on 12/25/2023, 3:29 PM

@john_dennis looking at your CPU, you could raise your cpu max temp (TJ max) to the official max of 100 degrees C to avoid thermal throttling happening to soon (rumor is that processor junctions will only melt at 115 degrees), and maybe keep both your power limits to 253W, next test via HWiNFO if your cooler can keep it under 90 degrees under 100% CPU load. if not, lower by steps of 10W until reaching the sweet spot for optimal balance between stability, economy and performance.

bitman wrote on 12/25/2023, 4:14 PM

Regarding setting the power limits, for modern CPU's it is best to keep the long term power limit and turbo short term power limit close. For the i9-13900K TAU can be set to 56 second of turbo, but it can also be left alone as it becomes irrelevant if both power limits are set to the same value.

bitman wrote on 12/25/2023, 4:52 PM

@john_dennis You seem to get away with under-volting, but I would be very cautious about it.

When I was investigating the issue I also tried under-volting (if you can get away with it, it does lower the temperature), but the CPU got unstable and unresponsive, even with an offset of -0.050V, especially when running idle. With the current knives-edge aggressive default BIOS settings, I would not be surprised it already got an optimized hidden voltage offset leaving no margin. In fact in the end I did the opposite of under-volting, I increased the offset to + 0.050V to improve stability. There is also the risk of ghost throttling with under-volting, difficult to detect, I recall reading it somewhere, but I forgot the details.

Stability is key, and performance comes second to me, anything above 200 W power draw and the 13900K CPU becomes very fast inefficient, with marginal performance gain against massive rise in heat and energy cost. But it has a very efficient sweet spot below it.

Howard-Vigorita wrote on 12/26/2023, 12:55 AM

I always prove my bios settings by running prime95 for 12 hours without a crash. Its small ffts are usually the back breakers. If particular memory and cpu settings in bios can handle that, they can handle anything. It might take a while to figure out by successive approximation what the limits are but I've never experienced a throttling crash with any software or benchmark. For those that don't use that approach to determine the limits of their system, Asus bioses also have what they call an AI feature that dynamically makes adjustments to bios settings based on observed temps over time. I've tried it in the past but found it overly conservative.

bitman wrote on 12/26/2023, 4:27 AM

@Howard-Vigorita indeed, prime95 is a good tool; I have used it many times in the past. Just to satisfy my curiosity, I downloaded the latest version, and let it run with the 200w power limit for my system. Core temperatures are en par with running Vegas 21 cpu cores 100% busy - cpu only render (around 85 degrees C.). These temperatures are stable, and no thermal throttling occurs.

It is odd that Vegas can survive a CPU throttle induced by reaching the imposed power limits, but not a thermal throttle. it must be that Vegas gets unstable by the throttling at the core temperature limits towards 100 degree C rather than the throttling at lower temperatures when power limits are imposed and reached. Another odd thing is that Cinebench 2024 does survive thermal throttling at 100 degree C. and does not crash...

@john_dennis here is the BIOS screenshot of the 200W setting I am satisfied with:

Mohammed_Anis wrote on 12/26/2023, 7:44 AM

Wow. I feel illiterate!

The last time I interacted with Bios was to optimize it for Virtual Machines (long story)

Can anyone comment if this is worth investigation for AMD? I have a Ryzen 5900X

Howard-Vigorita wrote on 12/26/2023, 12:23 PM

@bitman Haven't looked at prime95 lately myself... not since I built my 11900k system about 3 years ago. Things have changed since then but I don't see much difference in the app. Would have expected it to at least distinguish p- from e-cores and perhaps have a checkbox to torture test only the p's . Here's my usual arrangement of the latest version runnng on my laptop which is 12th gen. Core Temp is running along side of it. Looks like the 1st 6 are p-cores. When I see one core (like #3 here) running hotter than the others I usually go into the bios and lower that core's affinity so it gets assigned work less than the others. Because the cores are not asynchronous. So thermal throttling can only be done across the board on all cores when any one of them gets too hot. I personally think dynamic affinity adjustment would be more effective. Don't know if the bios's for motherboards for 12th gen and later Intel cpus still provide for even manual affinity adjustments since the p/e core scheme is billed as a way to minimize the concern.

Reyfox wrote on 12/26/2023, 2:11 PM

The AMD CPU's that had issues were the 7000 series X3D, which is fixed with a BIOS update to include a cap on SOC voltage at 1.3V, which is bypassed by XMP or even using AMD's Precision Boost Overdrive. I have a AMD 5950X (previous gen) and never an issue, even with all cores/threads running at 100% for extended periods of time. I do have XMP for the RAM enabled and sometimes do run PBO, and still no throttling issues. I've monitored CPU temps with HWiNFO64, and the temps are well below thermal throttling.

In addition, my case (Phanteks P400a) is known for it's airflow and being a "cool" case to put hardware in. Also using air cooling using one of the least expensive air coolers, the Thermalright Peerless Assassin.

john_dennis wrote on 12/27/2023, 3:39 AM

I'm unfamiliar with the UEFI system used by Gigabyte, but the one used by ASUS has a screen capture function that allows one to save a screen to a USB flash drive.

@bitman said: "No, Vegas does not crash if the set power limits are reached and the CPU throttles down because of that. Vegas crashes only (pretty much always) if the CPU throttles because of the core maximum temperature is reached."

Setting maximum core temperature limits individually to less than 100 degrees C will ensure that the core maximum temperature is not reached.

I took a 6.5 % render performance hit to limit core temp to 65 degrees C. System Upgrade 2023

I got some of that back when I raised the core temperature limit to 70 degrees C. I have zero interest in being number 1 on the forum benchmark list.

My HWInfo64 Summary

An Aside

I have a twenty-year-old truck and a thirteen-year-old car. I don't run either of them at the maximum temperature, either.

Former user wrote on 12/27/2023, 4:40 AM

Can anyone comment if this is worth investigation for AMD? I have a Ryzen 5900X

@Mohammed_Anis I have a mild undervolt to reduce fan noise and it works well for that. In the past when I flash new motherboard bios's everything is reset to stock, and I notice the CPU fan being louder at boot, load the saved bios settings and boot is noticeably quieter.

I did not want to mess about a lot and have crashes on boot so I worked on the assumption that the highest performing cores need the most power, with the others requiring less power, Highest performing cores have gold and silver icons in AMD Ryzen Master. I set these to only -5 undervolt, while the other cores set to -15

You can use values up to -25 or even more, but it's more messing about and can experience crashing and clock stretching. Clock stretching is when the CPU understands it's not getting enough power but instead of the core crashing it effectively reduces the frequency of the core making it slower and use less power while Windows and most reporting software still believes the CPU is operating at full frequency. That's the purpose of running benchmarks after every voltage change, you may think the CPU is both running faster and using less power but benchmark will tell you the truth.

Reyfox wrote on 12/27/2023, 4:52 AM

@Mohammed_Anis right now, I am using Precision Boost Overdrive (PBO). They have a "Creator Mode", that you can try and experiment with. You can leave it at "default" see how your CPU is being used (Task Manager/HWiNFO64), what heat it's producing, making sure your RAM is running at it's right speed. With the 5900X I believe the max RAM speed supported is 3200, which I had already set in the BIOS with XMP enabled.

There are many messages and tutorials on using PBO out there if you are interested. Doing "nothing" but using Firefox at the moment, only one core is being used, while the others are "sleep". Temps are in the mid 30's C.

PBO also depends on what cooler you have and well it can cool.

bitman wrote on 12/27/2023, 5:13 AM

@john_dennis Thanks for screenshot method, I have to see if the method is available on the new Gigabyte BIOS. I used the old school method of taking a monitor photo!

Both our methods are different, but purpose a same goal, keeping the core temperatures in check.

I prefer to limit the power rather than limiting the core temperature. It has the advantage that it avoids thermal throttling all together. The default TJ max at 100 degree C. is then just a "safety valve" if ever the ambient temperature (summer heat wave?) or a cooling fan breakdown occurs. I too have zero interest in being number 1 on the forum benchmark list as I stated before as key is stability on finding the sweet spot.

By the way, a decade or more back, temperatures like 65 degrees C. were considered (too) hot and official TJ max were lower in those days. Technology has advanced and allows for higher temperatures.