Advice on which platform will work bast with IMatch (AMD vs Intel)

[Damit]

I am about to build a new system.  I am leaning toward Intel 13th gen Rocket Lake, going with an i7-13700K or an i5-13500KF. IMatch will be heavily used along with Photoshop, Lightroom, Davinci Resolve, and premiere. Which processor would be best used by IMatch?  Would there be a big difference in performance between the-13700K i5 and i7-13700K?  I know that IMatch has used my processors more than any other program used for DAM, so it is obviously using all my cores, which I love. At times, things are processing and the CPU shows very little movement.  I assume this is where memory and the disk speed take over.  I am using a 4TB WD SN850X for my database, an EVO 970+ for my main C drive, and an EXOS seagate drive for my storage.  I really want to go SSD all the way, but I am running out of room on my 8TB EXOS, so it would be cost prohibitive for a year or two, until 10-12 TB SSDs come down to sub $500.I plan on going with 128 GB of DDR4 or DDR5 RAM.

So what is the advice? Where is the money best spent to get the quickest IMatch performance?

[Mario]

The only part of IMatch where CPU is the main factor is basically the IMatch AI (face recognition). This is also where IMatch can easily max out all CPUs. Assigning persons to faces is again mostly bound by how fast the database system can read data from the disk.

In fact, most operations in IMatch are I/O-bound, which means that the disk containing the databases (or the images) is the main performance factor. When IMatch has to read data from the database, read images, calculate checksums, read and write metadata, 90% of the time is spent "in" file system routines. IMatch does overlapping I/Os, reading/writing multiple files simultaneously etc. but there are limits of what can be done. That's also the part where I often tinker to improve speed.

Searching, data-driven category updates which can work on data already cached in memory are super-fast and mostly use one or a few threads. So no benefit from more cores, single-core is what would make a difference here. Of a few milliseconds, mind!

My impression is that modern processors are all "good enough" for IMatch. You can run IMatch just fine with small to medium-sized databases on modern Notebooks. I run IMatch very smoothly on a 6 year old PC with 6 cores and on my current development PC with 12 cores. This of course depends on whether your database has 100,000 or 500,000 files. My largest test database has one million files.

[Damit]

Thank you for the insight. Maybe the money is best spent on a SSD for my image files and not just the database.  I do notice my image drive always being used. Maybe the second factor would be memory speed?

[Mario]

The speed of the disk containing the images is not really that important.
IMatch reads images only once, unless you write metadata back.
Thanks to SSD storage (for the database) things are very fast these days.

I use a SSD for Windows and another for my development stuff.
For mass storage (images, videos) I use spinning RAID disks and NAS and my own cloud.

[jch2103]

One thing that will make a real difference in system performance for some programs is the choice of graphics card. Not so much in IM, but definitely for graphics/raw development programs that use 'AI' processing (e.g., noise reduction, etc.). 

[Mario]

One thing that will make a real difference in system performance for some programs is the choice of graphics card. Not so much in IM, but definitely for graphics/raw development programs that use 'AI' processing (e.g., noise reduction, etc.).

Yes. RAW processors with some sort of "AI" processing can benefit from fast GPUs.
IMatch too, to a certain extent, since it uses Windows DirectX for super-fast rendering and color management in the Viewer, Quick View panel and Slide Show.

I could also speed up the IMatch AI (or future versions of it) even further by requiring a certain level of graphic card.
But that's a real can of worms, because it adds two additional points of trouble: the graphic card and the driver.

Remember when some users had problems with IMatch crashing when the Viewer was used?
All caused by a buggy NVIDIA driver. Despite I'm providing NVIDIA with two crash logs pointing at the exact point in their driver that crashed, it took them about six months to get it fixed.

Using NVIDIA's CUDA or the AMD version of it (and maybe whatever Intel cooks up for their new line of graphic cards) is too much of a burden for a product like IMatch. To many variables. To little benefit.
It was hard enough to require AVX for IMatch 2020 and later - and that's a technology than is over a decade on the market!

I use CUDA with higher-end graphic cards on my development systems to speed up "AI experiments". It's a PITA to get it running, but it can make certain operations 10 times as fast. As in "needs one hour to complete instead of 10 hours". Which is really helpful!

[Damit]

Yes, I have been writing a bunch of metadata as I play with my keywords, sometimes it causes a lot of write-backs.  I have seem my pictures hard drive at 100% for long periods of time.  With a spinner, I get seek response times in the teens.  I did a test with an ssd.  The drive with the pics was now at 50% and rose to 80% but the read and write times were almost triple. The response times were in the low single digits. So it seems having the actual photos in an SSD will increase speed especially when writing metadata changes. 

I have a NVidia Founder's Edition 2070 super in my system and I am impressed at how fast IMatch renders files with no lag at all even though at times I am dealing with very LARGE .tiffs.