building a new pc for pixinsight + ram disk

[yobbo89]

wasn't too sure where to put this,thought image processing might be an ok place since it's on topic with astrophotography . it's ok to move it.

so i've been over at the pc forums for awhile ,learning about current cpu,s and what not as my last system was built about 5 years ago and a little clueless on whats the top end products and what works with what .

i've been running some benchmarks and comparing them over there as one of the users is familiar with pixinsight and doing astro processing builds .

It currently takes me about 18mins 50 sec to deconvolve a 9 panel image .. ZZzZZzz

it turns out linux is faster than windows in pixinsight, i've also learnt that using a ram disk as a swap file really dominates in swap performance and it might be a good choice for you . i used 8 gb of my 32gb ram for a swap file drive but i'm sure you can use it with a smaller amount of ram .

ram disk info.(i use primo ramdisk 30 day trail) .

http://dslr-astrophotography.com/improv ... e-storage/

some performance charts, all the ones on top are linux.
https://pixinsight.com/benchmark/


a post from the forums in pi from the pixinsight team.

(Please give a read to this post:

https://pixinsight.com/forum/index.php?topic=10588.0

where I describe the state of parallelization in current versions of PixInsight, and provide important information on the repercussion of I/O wait states for data and processor intensive tasks.

As you see in the article above, all image preprocessing tasks are already using high-level parallelization in current versions of PixInsight. Take into account that there are very important performance differences among operating systems. In descending order of performance: Linux, FreeBSD, macOS and Windows.

Instead of BPP, which is now obsolete and maintained only for compatibility with existing projects, you should use the new WeightedBatchPreprocessing script (WBPP). The image weighting task in this script requires high-level parallelization, just as implemented in the SubframeSelector tool. I am working on it, time permitting.)

windows is the least favourable for cpu utilization in pi at the moment ...

anyhow this is my first bench without ram disk

937mib/s swap speed on raid 0, older nvme drives, 5194 swap score

3290mib/s on a ramdisk 2400mhz ram ddr4 18222 swap score ( triple the speed)

anyway to sum it up there's a cpu performance limitation with windows and pi due to parallelization efficiency ,my system processor has a lot of spike with the cpu doing nothing then 100%,anyway one small way to make up performance is with better io,a ram disk or faster nvme drives , and linux or mac as the prefered choice .

anyway here is my current and new future build ,i will be reusing parts,i plan to dual boot with linux ,i haven't got the parts yet but still considering swapping around stuff .

current build
I7 6700K -1151 LGA
Gigabyte z170x-gaming 7
sli evga gtx 1080 ti with EK Full Cover VGA Block EK-FC1080
32 gb ddr4 ,4 sticks of 8 gb (Kingston HyperX Fury HX424C15FBK2/16 16GB (2x8GB)
Seagate ST4000DX001 Desktop SSHD 4TB
Seagate ST2000DX001 Desktop SSHD 2TB
Western Digital WD Black 3.5" 2TB SATA 6.0Gb/s 7200RPM 64MB Hard Drive
2x Intel 600P Series 512GB M.2 SSD in raid 0
corsair hx1000 psu

upgrade build

cpu : AMD Ryzen 9 3950X
mobo: MSI X570 Ace Motherboard $ 619
ram: Kingston HyperX Fury 32gb (4x8gb)
gpu: sli evga gtx 1080 ti
sdd: ADATA XPG SX8200 Pro M.2 NVMe SSD 2TB $539
hdd: Seagate ST4000DX001 Desktop SSHD 4TB
hdd: Seagate ST2000DX001 Desktop SSHD 2TB
psu: corsair hx1000


edit: i also forgot to mention i think there is a performance problem with fits and apparently xsif. runs faster in pi

[UlteriorModem]

Hrm this is very interesting. I will have to take a look at it! Thanks!

[UlteriorModem]

Well I added the swap file directories, 8 of them, and crash.

[DEnc]

upgrade build

cpu : AMD Ryzen 9 3950X
mobo: MSI X570 Ace Motherboard $ 619
ram: Kingston HyperX Fury 32gb (4x8gb)
gpu: sli evga gtx 1080 ti
sdd: ADATA XPG SX8200 Pro M.2 NVMe SSD 2TB $539
hdd: Seagate ST4000DX001 Desktop SSHD 4TB
hdd: Seagate ST2000DX001 Desktop SSHD 2TB
psu: corsair hx1000

You're going to like that CPU! Recently I got a system with the same and it's very very fast.

[yobbo89]

did you try making a ramdisk tom ? , i had better luck disabling the nvme ssd swap file (operating system drive) and just running on the ram disk with 4 directories .

[yobbo89]

upgrade build

cpu : AMD Ryzen 9 3950X
mobo: MSI X570 Ace Motherboard $ 619
ram: Kingston HyperX Fury 32gb (4x8gb)
gpu: sli evga gtx 1080 ti
sdd: ADATA XPG SX8200 Pro M.2 NVMe SSD 2TB $539
hdd: Seagate ST4000DX001 Desktop SSHD 4TB
hdd: Seagate ST2000DX001 Desktop SSHD 2TB
psu: corsair hx1000

You're going to like that CPU! Recently I got a system with the same and it's very very fast.

yes i see the top score for the 3950 in pi got around 16 seconds on the bench,a heap faster then my 50-55 sec ! , i think even the 3900 x is pretty close . i still haven't put the order in the basket yet, there's news slowly rolling out on the new amd cpu set that's coming out soon , plus i might just consider the 3900x and save some $ and put the rest on faster memory

[DEnc]

[quote=yobbo89 post_id=84365 time=1589141289 user_id=90
....i might just consider the 3900x and save some $ and put the rest on faster memory
[/quote]

I was under the impression that threads were the key to speed with PI. Would speedier memory make up for the reduced number of cores with the 3900X?

[yobbo89]

[quote=yobbo89 post_id=84365 time=1589141289 user_id=90
....i might just consider the 3900x and save some $ and put the rest on faster memory

I was under the impression that threads were the key to speed with PI. Would speedier memory make up for the reduced number of cores with the 3900X?
[/quote]

From the benchmark database, the 3900x was 0.7 sec slower then the 3950x not much difference, I would like to see these users full system specs. My understanding by looking at the benchmarks and tinkering with swap files that alot of pi benifits with quick access low latency and having a faster swap file speed. It dosn't look like CPU power scales well, As when you run a bench with a CPU performance monitor up you'll see alot of CPU power unutilised or spiking in performance. Trying to work out the bottleneck in pi is a tricky one, I hate to break out the negative but yeah it looks like the programing side could do with an improvement...

[Juno16]

Dang, yobbo, that is an awesome build!

I build an economy system last year with the Ryzen 7 2700x (8 core, 16 thread) and 32g of DDR4 ram with a nvme gen 3 boot drive. Blasts through Startools. I know that PI is a different animal. Good luck and post back what you decide on.

[yobbo89]

with the new pc i cut my 9 panel mosaic 50 iterations deconvolution from 18mins 54 sec to 8 mines 52 sec over the previous build , havn't tested it with a ram swap file yet.

Here is the pc

[Juno16]

Very nice looking pc Yobbo!

Definitely not an economy build! Looking forward to your results with the ram drive.

[STEVE333]

with the new pc i cut my 9 panel mosaic 50 iterations deconvolution from 18mins 54 sec to 8 mines 52 sec over the previous build , havn't tested it with a ram swap file yet.

Here is the pc

Looks like something out of a sci-fi movie. Impressive and impressive results so far.

Steve