Optimal PC for Startools

[maxchess]

Hi,
I am thinking about getting a new PC (laptop or Desktop) and wonder if there is an optimal configuration, beyond which it is not worth going for running Startools>

I currently run Startools on a Dell Inspiron 620 : Intel(R) Core(TM) i3-2100 CPU @ 3.10GHz, 3100 Mhz, 2 Core(s), 4 Logical Processor(s) 8gb ram with an SSD
I also have a Dell XPS laptop : I7-7560U @ 2.4Gz 16GB memory also SSD which based on benchmarks is only 33% faster.

What sort of balance between processor, cores, memory and SSD would best suit Startools? At present I have plenty of time to make a cup of coffee and a sandwich between processing steps.
According to benchmarks a I9-9900K would be 2.3 x faster but thats just raw processing power.

Regards
Max

[admin]

Hi Max,

Desktops, by far, are the preferred option for heavy tasks like StarTools.
Most (not all) laptops are built for power efficiency, rather than grunt. The 3rd gen Ryzen series laptop chips are now finally beginning to come close to the performance of the 2nd gen i7 processors from almost 9 years ago.

So if portability is not not an issue, I'd strongly recommend a desktop.

In order of importance, StarTools prefers;

• Higher number of cores
• IPC (Instructions-Per-Clock); how fast can a single core process a task.
• Memory speed; how fast can the system read and write to memory
• Storage speed; how fast can the system transfer data back and forth to storage memory.

All the above is assuming you have sufficient RAM (16GB+ recommended).
If you have 24GB+, you could start using your excess RAM as a RAM drive, which will speed up StarTools' Tracking somewhat. I've been working hard over the past few days to make StarTools more efficient and faster when it comes to storage. This enhancements should become available in the 1.6 alpha.

StarTools does not use your GPU and will not do so for the foreseeable future, as the Tracking engine's signal path is just too complex in the way it branches and pulls data from multiple sources for a GPU to efficiently process (they hate branching!).
EDIT 2020: As of version 1.7 supports GPU acceleration in all modules, with many parts of the signal path rewritten for GPU execution (reduced branching, fewer memory transfers/access).

Also not unimportant, a Desktop will let you choose (or upgrade) the screen you are working with.

As of the time of this writing, as far as Desktops go, the 6-core, 12-thread 3rd generation Ryzen 5 series CPUs offer exceptional value. You could also consider a 2nd generation Ryzen 7 (8-core, 16-threads) series if offered at the same price. Intel CPUs are currently overpriced for what they offer.

I hope this helps!

[maxchess]

Ivo,
Yes that is really helpful
Max

[maxchess]

Ivo
You mentioned upgraded screen in your reply, is there any particular screen or set of characteristics I should look for.

Regards

Max

[admin]

Ivo
You mentioned upgraded screen in your reply, is there any particular screen or set of characteristics I should look for.

Regards

Max

Color faithfulness is one. That said, you can get decent results from most monitors by getting a cheap color calibrator (for example the ColorMunki Smile).

[maxchess]

Sorry, one more question, does M.2 SSD Drive improve speed compared to SSD drive, if so how much is worth having??

Max

[admin]

Hi Max,

If you choose an NVMe-based M.2 SSD, this should indeed be faster than a SATA3 based SSD.
That said, in practice the real-world difference will not be that noticeable for most applications. You may notice some small speed difference in between modules (when StarTools is tracking).
Whether you benefit from the higher data throughput depends a little on what other tasks you plan on using you PC for.

SSDs have come down in price a fair bit, and 960GB (or something else in the 1TB range) is now just as good value in terms of $ (or euro or pound ) per GB as smaller drives.

You can also keep a bigger, good old fashioned mechanical drive as a secondary drive for long term data storage (e.g. datasets, sub frames, etc.) once you no longer need them.

[maxchess]

Ok, so I decided to go for it and got a PC with:
AMD Ryzen 9 3900X 12 Core CPU (3.8GHz-4.6GHz/70MB CACHE/AM4)
RAM: 64GB DDR4 3200MHz
M.2 SSD Drive 1TB CORSAIR MP600 GEN 4 PCIe NVMe (up to 4950MB/R, 4250MB/W)
Storage: 6TB SEAGATE 7200 RPM 256MB CACHE
CPU (3.8GHz-4.6GHz/70MB CACHE/AM4) , RAM: 64GB DDR4 3200MHz
M.2 SSD Drive 1TB CORSAIR MP600 GEN 4 PCIe NVMe (up to 4950MB/R, 4250MB/W); Storage: 6TB SEAGATE 7200 RPM 256MB CACHE

Just got delivered and I will run a few comparisons with my old PC in the next few days.
And I am now running the latest StarTools-Windows64-AVX2 and it seems pretty fast, but not instantaneouson on an unbinned image 4072x2747 after cropping.

Are there any system tweeks you would recommend to get the most out of this PC/Starttools combo?

Max

[admin]

Wow, that's a monster of s system!

And I am now running the latest StarTools-Windows64-AVX2 and it seems pretty fast, but not instantaneouson on an unbinned image 4072x2747 after cropping.
Are there any system tweeks you would recommend to get the most out of this PC/Starttools combo?

That's excellent. Running StarTools from a RAM drive is the fastest possible way of running StarTools. Given you have 64GB of RAM, you should be able to allocate some to a RAM drive.
Unfortunately, I have limited experience with using Windows for this, but on Linux it is as easy as setting up your /tmp folder to use RAM using tmpfs.
Doing so, you will only notice a difference when StarTools is swapping Tracking information to and from disk, which really mostly only happens during module init and when 'keep'ing the result.
You could also overclock your CPU and/or memory if you are knowledgeable enough. StarTools is exceptionally good for testing your system's stability.