Graphics card?

[Replytoken]

Thanks for that link. However I cannot see in his methodology where he clears the cache between each test and where his LR catalogue is stored. It is possible all he is doing is using cached data on his internal SSD for each "test". But he is the guru so....

I did not ask Jim, but he is usually pretty careful and thorough. But, yes, it could be possible.

--Ken

 

[Replytoken]

Thanks for that link. However I cannot see in his methodology where he clears the cache between each test and where his LR catalogue is stored. It is possible all he is doing is using cached data on his internal SSD for each "test". But he is the guru so....

I did reach out to Jim and he had some useful information to share with me. He did not recall clearing the cache based on his past experiences. But, Victoria (aka "the LR Queen" also ran tests where she cleared the cache and had similar results. Jim also ran one of the series again for me last night/this morning while clearing the cache and here are his results:

1. Files on Internal SSD: 2.20 secs per image (2.28 secs on the previous test).
2. Files on External SSD: 2.25 secs per image (2.28 secs on the previous test).
3. Files on External USB3 hard drive: 2.23 secs per image (2.24 secs on the previous test).

There was minor improvement, but not enough to really make a significant difference. To sum up the findings in Jim's words: "the file read speed is just not a factor in the overall preview-building process. Frankly, the drive could be read at Warp speed and it still wouldn't make much, if any, difference. Preview building is almost all CPU driven, so using the GPU instead of the CPU might help, as we've just seen with AI Denoise (and with export performance), but until then I can't see how the user can improve the preview building performance."

Hope this helps,

--Ken

 

[Alistair]

I did reach out to Jim and he had some useful information to share with me. He did not recall clearing the cache based on his past experiences. But, Victoria (aka "the LR Queen" also ran tests where she cleared the cache and had similar results. Jim also ran one of the series again for me last night/this morning while clearing the cache and here are his results:

1. Files on Internal SSD: 2.20 secs per image (2.28 secs on the previous test).
2. Files on External SSD: 2.25 secs per image (2.28 secs on the previous test).
3. Files on External USB3 hard drive: 2.23 secs per image (2.24 secs on the previous test).

There was minor improvement, but not enough to really make a significant difference. To sum up the findings in Jim's words: "the file read speed is just not a factor in the overall preview-building process. Frankly, the drive could be read at Warp speed and it still wouldn't make much, if any, difference. Preview building is almost all CPU driven, so using the GPU instead of the CPU might help, as we've just seen with AI Denoise (and with export performance), but until then I can't see how the user can improve the preview building performance."

Hope this helps,

--Ken

Thanks Ken, I must confess to being a little puzzled by these results!
Regards
Alistair

 

[Replytoken]

Thanks Ken, I must confess to being a little puzzled by these results!
Regards
Alistair

I think that Jim's quote in my final paragraph is the key. No matter how fast you can deliver the file information, the CPU is more or less the bottleneck for this action.

--Ken

 

[Alistair]

I think that Jim's quote in my final paragraph is the key. No matter how fast you can deliver the file information, the CPU is more or less the bottleneck for this action.

--Ken

I had a little spare time and did a quick test of my own. I simply don't understand your friend's findings. Here is my computer building standard previews for a around 3,000 images stored on a spinning SATA disk. As you can see, the disk is fully maxed and the CPU is idling. If I move the images to a fast NVME drive, the CPU maxes out and the drive idles. It takes a fraction of the time that the slow disk does. I don't really understand how your friend is reaching the conclusion that disk speed makes no difference. It does. An enormous difference.

 

[Replytoken]

I had a little spare time and did a quick test of my own. I simply don't understand your friend's findings. Here is my computer building standard previews for a around 3,000 images stored on a spinning SATA disk. As you can see, the disk is fully maxed and the CPU is idling. If I move the images to a fast NVME drive, the CPU maxes out and the drive idles. It takes a fraction of the time that the slow disk does. I don't really understand how your friend is reaching the conclusion that disk speed makes no difference. It does. An enormous difference.

View attachment 62044

I appreciate you running a test and do not doubt what you are reporting, but I am wondering if you timed this with each drive, as that was how Jim reported his findings, and I believe that is how Victoria did as well. Perhaps I may have misspoken when I said the CPU was the bottleneck. It could just be that the LRC software code is the bottleneck. I have not run any tests of my own, mostly because I do not feel confident enough to say that I have eliminated any variables that could impact the results. And, my system is quite old, so I am not sure how that would impact any test.

--Ken