Results from plotting with just 1 thread

For those interested in experiments, here’s what plotting 2 plots with single threads (in parallel) looks like vs 1 plot with 2 threads on the same machine (single threads were given bit more memory though).

single

1 Like

I’m interested in hearing about plotting with “1 thread”. Thanks for giving the total time in hours and minutes. Some other tests given the total time in seconds (had to bring up calculators).
Your post does not answer me though…:
“added more RAM after test or just added without test?”
CPU GHs speed?
AMD / Intel?
What if "you do not give more RAM, what would be the result?

Yes, sorry for not controlling all parameters. Still thought I’d report it as very few ever seem to ahve attempted single thread. I also need to mention this is not a fully controlled environment as this is just running on a PC that is used for some light office work during the day.

I moved to plotting with 4608 MiB as I can’t have more than 2 parallel plots running without being impacted too much. 4608 MiB was the “old” requirement, before the 1.04 release. I’m still experimenting with different memory allocations which so far I have not found to be easy to answer as in there does not seem to be just a simple correlation as in 6790 MiB seems slightly better than 3390 MiB, but 8192 MiB is worse than both)

Specs:
Intel Core i7 6700 (3.4 Ghz) 4 cores, 8 threads
16 GB DDR4 1066Mhz
Samsung 870 EVO 1TB Sata (temp)
Synology DS920+ connected via 1GB/s (final)

So, with an i9-9900K at 5GHz, even with a single thread, those figures you have given should change drastically, am I right?
Your RAM speed sucks, :slight_smile:

Not sure as I was assuming 1 thread would be mostly IO blocked.

As for the ram speed, I know, it is an old machine that got for free from someone that thought it was beyond rescue :slight_smile:

This is very good info, thank you for sharing it, but can you post as text rather than images? Text in images can’t be indexed by google…

Ofc, no problem.

Phase1Seconds Phase2Seconds Phase3Seconds Phase4Seconds CopyTimeSeconds TotalSeconds Threads Buffer Buckets
04h 15m 47s 01h 29m 01s 03h 34m 10s 26m 55s 15m 37s 09h 45m 55s 2 3390 MB 128
07h 13m 38s 02h 09m 16s 04h 57m 04s 33m 16s 38m 39s 14h 53m 15s 1 4608 MB 128
07h 13m 41s 02h 09m 22s 04h 57m 07s 34m 05s 38m 24s 14h 54m 16s 1 4608 MB 128
2 Likes

Was just wondering how this would scale @Yae so this is really valuable.

Given my recent experiment, I think I found some overhead for +2 additional plots I could run, but that would cause my thread count to go over the 16 on the chip - so I was thinking of making the SATA-based plots single-threaded.

As long as the plot time increase is < 100% increase, it makes sense!

1 Like

On your 5950x you have 32 threads. Admittedly these are 16 full threads and 16 supporting hyper-threads. A hyper thread is fine to fill in some gaps while the main thread is blocked.

You want to leave at least 2 threads for things going on besides plotting. That leaves you 15 thread pairs (ideally spread 2 per core, which you could try by affinity assignment) to plot. If you use a plotting manager that controls entry into phase 1, you could further optimize for this.

Trying to go 30x single thread is ill advised as this would result in hyper-threads being loaded with full workloads, which will not be ideal IMHO.

Maybe you are thinking of @codinghorror - but I’m on an i7-11700k (8 cores / 16 threads)

Then you don’t want to use more than 14 threads, pref 7x2.