Originally posted by ProfessorBuzz That clock speed is a big yellow flag right there. What make/model laptop is it?The i7-8850 has a few variants, but mostly they run at 2.6GHz to 4.3GHz peak. The low-power frequency is 800MHz
DELL laptop, i7-8850U, peak clock speed can go up to 4GHz. Power management was set to the default "balanced", but the GPU was set to max power saving + medium power saving on the PCI bus + some mixed CPU cooling. CPU cooling can be manually set to passive or active. I figured "Passive" means the temperature is regulated via clock speed to save energy, and "Active" means temperature is lowered via fan, if the temperature continues to rise when the fan is at full speed the clock speed decreases. Changing the GPU power setting immediately made a big difference, I supposed it was a first bottleneck. Changing the CPU cooling mode avoids the CPU to slow down after processing a couple of images, but CPU clock still gets lowered to about 2.25GHz because the cooling air flow is insufficient to get rid of the heat. I didn't notice a speed difference by switching off the PCI bus power saving. The CPU now runs 100% and the clock is lowered a bit, I think that's the bottleneck , coming from the cooling. Batch export speed improved ~ 3x compared to before, but the most important is that working on raw images with sliders and local adjustments is now almost real time, that's the big improvement.
---------- Post added 06-01-22 at 08:15 ----------
Originally posted by jgnfld Sorry to say this but this is not probably not a machine with the best specs for large batch image processing jobs on 100 full frame RAW files even if you fully max it out, put it in a very cold room, and leave it plugged in unless you just set up a job and leave it for a period. Even at 3 min/file the machine will complete the job easily overnight, for example.
Yes you are right. When I purchased the notebook, my priority was the compact small size for traveling and using it as a tablet. On paper the specifications were quite good at the time (2017), I looked at good battery life, but I didn't pay attention to the GPU spec, and notebook mother board designs are optimized for being power efficient, not highest performance.
---------- Post added 06-01-22 at 08:17 ----------
Originally posted by steephill It is an i7-8850U processor, not a great choice but it should be capable of much better performance. The 800Mhz mode is used when saving power.
You are right.
---------- Post added 06-01-22 at 08:19 ----------
Originally posted by Medex Your 3 minutes for 100 jpg files (Full size?) are very good
Arrhhh , no, it was 3 minutes for one file export, 30 minutes for 10 files...
---------- Post added 06-01-22 at 08:28 ----------
Originally posted by marcusBMG Your i7-8850U is a relatively low performing cpu by contemporary standards, only nominally about 3 times the speed of one of those old ones on passmark benchmark scores. I suspect that similar stipulations apply.
Normally, process can use the GPU to offload some of the work, but my GPU is probably not up to the task to help complex image processing. What makes me think that way is that Topaz Sharpen AI performs a "calibration" to decide how to load the machine, and in my case Topaz decides to use the CPU and not to use the GPU. So, not only my CPU isn't strong but it also take the full work load and the GPU does nothing. I've seen a GPU test done in 2021, my model of GPU gets 2% of the max performing GPU model in the list and the top 20 GPU (NVIDIA) deliver at least 60% of the max performance.
---------- Post added 06-01-22 at 08:31 ----------
Originally posted by GUB In Darktable profiled denoise seems to slow things down a bit.
Exactly. The cleanliness of the output file impacts processing time. I processed /exported the same raw files with RT, the processing was much faster, it took less than a minute for 10 files, but the output image quality isn't as smooth and nice.
---------- Post added 06-01-22 at 08:41 ----------
Originally posted by GUB Is there anything to learn by a batch export versus a single export?
Interesting, I haven't tried that. Also, I used Adobe DNG converter to compress some RAW files. The load time is 2 to 4 times quicker depending on what's inside the DNG file. In this process , I understand that the image processing software reads the DNG data and replicate a proxy of image data in memory, cache, and also in the file system before it can process the image through the UI. The amount of image proxy data determines real time processing speed. Export speed is different because it take the full amount of image data processed according to parameters set on the proxy image. So if I take a RAW file, go through Adobe DNG converter and modify setting such as changing embedded JPEG preview from full size to 8Mpixel (~4K) + data compression, the output DNG now loads two times faster in image processing software.
---------- Post added 06-01-22 at 08:46 ----------
Originally posted by turbo_bird According to a computer guru friend of mine, it was overheating and throttling when doing batch exports or trying to render 4k timelapse videos, and the real bottleneck was the slow buss speed of the motherboard and the slow ram it used, both only running at 100mhz I think he said.
Yes, the specifications is only part of the equation for the processing speed. There is also how fast is the bus to transfer data between HDD/SSD and RAM, how fast the data can transit between RAM and CPU, between CPU and GPU. The slowest component in the chain slows every else.
---------- Post added 06-01-22 at 08:51 ----------
Originally posted by turbo_bird He put together a new system for me using an AMD Ryzen with the same amount of ram and not a lot faster processor speed. It's way faster than my old computer, cost about 4 times as much, and takes up twice as much room on my desk, but it's amazing how fast it is.
It's probably got a much fast RAM access, faster data bus and much better cooling. I'm now thinking my next computer purchase will be a desktop.
Last edited by biz-engineer; 01-06-2022 at 12:47 AM.