Jump to content

  •  

CNers have asked about a donation box for Cloudy Nights over the years, so here you go. Donation is not required by any means, so please enjoy your stay.

Photo

What Hardware to Make PI Really Fast?

  • Please log in to reply
103 replies to this topic

#51 TelescopeGreg

TelescopeGreg

    Apollo

  • -----
  • Posts: 1065
  • Joined: 16 Jul 2018
  • Loc: Auburn, California, USA

Posted 18 September 2019 - 07:35 PM

Yes..but modern NVME and SSD drives will always last far beyond their warranties, which is usually capacity x 50. Anyone who has run a RAID array, or just used a mechanical hard drive over time knows they will not get close to that before they get a drive failure; usually mechanically-related. A  terabyte NVME m,2 drive that has a quarter of its memory partitioned for durability will last for years and years and years.

 

So yes, quarterly backups of your frames to either cloud, or NAS storage is smart, but no, don't lose sleep thinking your new NUC's NVME drive will fail because you put on 20GB a week on it a week in light and flat frames. 

Yes!  But how many folks (besides you and me) will set aside that space for durability purposes?

 

When folks get an SSD (et al) and fill it with file storage and swap is when I get nervous.  Yes, it's fast, until it's not.



#52 miwitte

miwitte

    Apollo

  • *****
  • Posts: 1124
  • Joined: 28 Dec 2016
  • Loc: Charlotte NC

Posted 18 September 2019 - 08:09 PM

Also try different numbers of swap files i found 4 was best for me. I built a 1950X with 64 GB ram 32 GB ramdrive and M2 drive. Wicked fast so cool to watch all 32 cores go to 100% during integration and other stuff. I went from using my laptop which would take 45 minute to drizzle integrate 100 subs and now its under 2 minutes. RAMDrives do work, but you can fill them up too which ive run into and PI will not be happy.. 



#53 Jim Waters

Jim Waters

    Mercury-Atlas

  • *****
  • Posts: 2967
  • Joined: 21 Oct 2007
  • Loc: Phoenix, AZ USA

Posted 18 September 2019 - 09:24 PM

RAMDrives do work, but you can fill them up too which ive run into and PI will not be happy.. 

If I configure two PI Swap Spaces, the first being a 16GByte "RAM Disk" and the second being a "SSD" C:/Users/.../Local/Temp will PI fail over to the SSD when the RAM Disk is full?



#54 miwitte

miwitte

    Apollo

  • *****
  • Posts: 1124
  • Joined: 28 Dec 2016
  • Loc: Charlotte NC

Posted 18 September 2019 - 09:44 PM

Im 99.9%sure the swap space needs to be the same like

 

C:/swaptemp

C:/swaptemp

C:/swaptemp

C:/swaptemp

 

Never tried it with two different disks but it may confuse PI where the swap files are when your goign back and forth making changes.



#55 Chris Ryan

Chris Ryan

    Vostok 1

  • -----
  • Posts: 146
  • Joined: 08 Mar 2014
  • Loc: Brisbane

Posted 18 September 2019 - 09:48 PM

Hey Chris,

 

Nice... but I don't think I am up to tweeking, but then again I'm also not looking to make the fastest PI machine in the world, just something somewhat close. I'll look into the CPU long with looking into many of the others too.

 

 

Miguel   8-)

 

.

Totally understand.  When I built the system I did an initial benchmark and saw how good it was.  I got curious at how far I could push it. :)



#56 james7ca

james7ca

    Fly Me to the Moon

  • *****
  • Posts: 7410
  • Joined: 21 May 2011
  • Loc: San Diego, CA

Posted 18 September 2019 - 11:34 PM

Im 99.9%sure the swap space needs to be the same like

 

C:/swaptemp

C:/swaptemp

C:/swaptemp

C:/swaptemp

 

Never tried it with two different disks but it may confuse PI where the swap files are when your goign back and forth making changes.

 

I use two SSDs that have partitions in a RAID0 arrangement and two of the partitions/volumes are dedicated to PI's swap while another is for the system VM. So, you can definitely use two volumes for PI's swap space (say G: and I:) and then make multiple directories for the swap files and have something like the following (where A, B, C, and D are the swap directories):

 

G:/A

I:/B

G:/C

I:/D

 

This is exactly the setup I've been using for the last two years under Windows 10 Home and it gives me 120GB of space for the PI swap and 50GB for the VM (the latter twice the size of my DRAM).

 

Also, note what the PI help says about swap space (see screen shot below), they recommend a minimum of 60GB which isn't that convenient if you are using a RAM disk. So, benchmarks may be one thing while the actual operation on large data sets may be something else (in term of performance).

 

All this said, I tend to agree with Jon about spending lots of money on new computer hardware. That's fine if you can easily afford it but I wouldn't break my budget just so that I can run PixInsight a little faster.

Attached Thumbnails

  • PI Swap.jpg

Edited by james7ca, 18 September 2019 - 11:35 PM.

  • PirateMike likes this

#57 PirateMike

PirateMike

    Mercury-Atlas

  • -----
  • topic starter
  • Posts: 2634
  • Joined: 27 Sep 2013
  • Loc: A Green Dot On A Blue Sea

Posted 19 September 2019 - 01:12 AM

Jeez, I bought a  6 year old Lenovo I5 with 8 gigs of RAM for $120.00 off Craigslist and it runs PI blazingly fast. Compared to CAD/CAM and video processing, PI does not need very much horsepower to run well. At least that's been my experience

I guess it all depends on what one is doing with it. For me, each sub frame is about 30 meg and the flats, darks and bias frames are of course the same size. I do go overboard (as some have said) with the number of calibration frames that I use for each image. the problem is not so much working with the master frames, the problem is creating the master frames.

 

My computer is a little old but it is not a dinosaur. Just by adding a RamDisk I went from 92 seconds to my newest benchmark of 51 seconds, this is a 40% decrease.

 

One thing I was doing today is to see what benchmark is possible with the latest and greatest PC out there just to see how my machine stacks up and I found was startling.

 

The fastest Windows machines to report their benchmark at 25 seconds and up, so I only need to reduce my time by 50% to reach them.

 

The startling part is that my first reduction of 40% only cost me $9.99 in software and maybe an hour to figure out what to do with it to get a decrease of the next 50% would cost at least $1,500 or more! Never did I think the ROI's would be so drastic to go from good to better and then to the best!

 

 

The fastest I could easily find. Note the processors being used.

 

Benchmark - Fastest.png

 

 

 

Miguel   8-)


Edited by PirateMike, 19 September 2019 - 01:20 AM.

  • Jon Rista likes this

#58 PirateMike

PirateMike

    Mercury-Atlas

  • -----
  • topic starter
  • Posts: 2634
  • Joined: 27 Sep 2013
  • Loc: A Green Dot On A Blue Sea

Posted 19 September 2019 - 02:14 AM

I checked the performance of my machine today and the bottle neck seems to be the processor. It is at 100% usage through nearly the whole bench marking process, the SSD goes up only slightly when the processor goes down slightly, the ram does even less work.

 

 

 

 


Edited by PirateMike, 19 September 2019 - 02:15 AM.


#59 james7ca

james7ca

    Fly Me to the Moon

  • *****
  • Posts: 7410
  • Joined: 21 May 2011
  • Loc: San Diego, CA

Posted 19 September 2019 - 02:14 AM

Miguel have you actually tried to run a large data set using your RAM disk? You MAY find that when using such a "small" swap space that the improvements in your real-world data sets are not as great as shown by the PixInsight benchmark. Note what the PI team said (post #56) about swap space...

a minimum of 60GB is strongly recommended.

I don't know, but 8GB of RAM disk is a long way from a minimum recommendation of 60GB. It's also possible that you will have runtime issues when using that small of a swap space. But, these problems may only show up after long sessions in PixInsight, meaning several hours with large data sets.

 

But, once again I question whether it is economically wise to do hardware upgrades that may only save a few seconds here and there in operation.


  • Jim Waters, PirateMike and Jon Rista like this

#60 PirateMike

PirateMike

    Mercury-Atlas

  • -----
  • topic starter
  • Posts: 2634
  • Joined: 27 Sep 2013
  • Loc: A Green Dot On A Blue Sea

Posted 19 September 2019 - 03:08 AM

Miguel have you actually tried to run a large data set using your RAM disk? You MAY find that when using such a "small" swap space that the improvements in your real-world data sets are not as great as shown by the PixInsight benchmark. Note what the PI team said (post #56) about swap space...

 

 

I don't know, but 8GB of RAM disk is a long way from a minimum recommendation of 60GB. It's also possible that you will have runtime issues when using that small of a swap space. But, these problems may only show up after long sessions in PixInsight, meaning several hours with large data sets.

 

But, once again I question whether it is economically wise to do hardware upgrades that may only save a few seconds here and there in operation.

Yes, it certainly seems like the money that needs to be spent is not worth the gain when one gets closer to the bottom of the barrel. When I started this post I never thought the returns would diminish so quickly. Well I did spend $9.99 on the ramdisk software (and upgraded it again), and I have played with it all day but to no avail.

 

I did move the RamDisk up to 24 gigs of 32 but no improvement. I think I need faster ram, I do have the first very first version of DDR4 that was available, which is obviously the slowest you can get now. 

 

But I think the whole bottleneck is in the CPU, specifically the L1, L2 and L3 caches being to small and slow.

 

I do have an unopened box with a 500gig Samsung 860Evo in it which I will try out for use with the swap files. I'm thinking of putting both PI and the swap files on it and see how that goes with big real world data and a stop watch. I'll use it to try out different combinations and see if I can squeak out an improvement for free.

 

I'm just tired of sitting around waiting to calibrate out my subs to produce my masters. It reminds me of the first computer I built, my lowly and long dead 286DX.

 

Well time for bed, I'm depressed. frown.gif

 

 

Miguel   8-(

 

.


Edited by PirateMike, 19 September 2019 - 03:10 AM.


#61 Jim Waters

Jim Waters

    Mercury-Atlas

  • *****
  • Posts: 2967
  • Joined: 21 Oct 2007
  • Loc: Phoenix, AZ USA

Posted 19 September 2019 - 11:48 AM

FYI - Known Linux releases that PI runs on.  Also pending PI release info. From Juan Conejero.

 

https://pixinsight.c...w;topicseen#new


Edited by Jim Waters, 19 September 2019 - 11:49 AM.

  • PirateMike likes this

#62 jerahian

jerahian

    Viking 1

  • *****
  • Posts: 616
  • Joined: 02 Aug 2018
  • Loc: Maine

Posted 19 September 2019 - 01:09 PM

Miguel have you actually tried to run a large data set using your RAM disk? You MAY find that when using such a "small" swap space that the improvements in your real-world data sets are not as great as shown by the PixInsight benchmark. Note what the PI team said (post #56) about swap space...

 

I don't know, but 8GB of RAM disk is a long way from a minimum recommendation of 60GB. It's also possible that you will have runtime issues when using that small of a swap space. But, these problems may only show up after long sessions in PixInsight, meaning several hours with large data sets.

 

But, once again I question whether it is economically wise to do hardware upgrades that may only save a few seconds here and there in operation.

If you don't have enough total swap space, PI will fail.  The keyword here is "total."  PI allows you to enter multiple directory locations for swap, so one can be the RAM disk and the other SSDs or whatever.

 

The only thing I'm not sure of is how PI selects which swap directory location to use; is it by order of listing, is it by remaining free space, etc.


  • Jim Waters and PirateMike like this

#63 PirateMike

PirateMike

    Mercury-Atlas

  • -----
  • topic starter
  • Posts: 2634
  • Joined: 27 Sep 2013
  • Loc: A Green Dot On A Blue Sea

Posted 19 September 2019 - 01:41 PM

FYI - Known Linux releases that PI runs on.  Also pending PI release info. From Juan Conejero.

 

https://pixinsight.c...w;topicseen#new

Jim,

 

Are you planning on switching to Linux?

 

 

Miguel   8-)



#64 Jim Waters

Jim Waters

    Mercury-Atlas

  • *****
  • Posts: 2967
  • Joined: 21 Oct 2007
  • Loc: Phoenix, AZ USA

Posted 19 September 2019 - 01:53 PM

Looking into it.  Not sure if I should do...

 

- Dual Boot - With sharing a NAS drive or external 3TByte USB 3 drive

- Running Windows 7 Pro virtualization

 

The dual boot would be more efficient from a performance overhead standpoint but more of a pain.  I also need to decide if I will be migrating to Win 10 Pro in the future.

 

EDIT - My main HW concern I have is the video card and Linux driver.  Many video cards don't play well with Linux.


Edited by Jim Waters, 19 September 2019 - 02:03 PM.

  • PirateMike likes this

#65 PirateMike

PirateMike

    Mercury-Atlas

  • -----
  • topic starter
  • Posts: 2634
  • Joined: 27 Sep 2013
  • Loc: A Green Dot On A Blue Sea

Posted 19 September 2019 - 02:09 PM

I'm thinking of at least trying Linux out. I have used Linux before and I really liked it, but it was not compatible with the job that I was doing at the time so I eventually removed it from my computer. I had a lot of fun trying out a lot of distros but Mint really became my favorite. It's too bad that all of the AP software can't be run on Linux, I think that would be the leanest and coolest way to do AP. Windows is just way to overblown to run such simple programs as EQMod, SGP and the such.

 

Let me know when you're going to take the dive. I would be very interested for sure and maybe we can do the switch at the same time. Now that would be fun. waytogo.gif

 

 

Miguel   8-)

 

.


 


Edited by PirateMike, 19 September 2019 - 02:09 PM.


#66 spokeshave

spokeshave

    Vanguard

  • *****
  • Posts: 2091
  • Joined: 08 Apr 2015

Posted 19 September 2019 - 02:36 PM

EDIT - My main HW concern I have is the video card and Linux driver.  Many video cards don't play well with Linux.

Linux had gotten much better about that - as have the video card makers. Most have Linux drivers that work fine. 

 

The Linux dividend in PI is huge. I got at least a 40% performance improvement on the same box when I switched to Linux. Plus, is is extremely simple to set up a permanent ramdisk in Linux and it actually works very well. 

 

Tim


  • Jim Waters likes this

#67 Jon Rista

Jon Rista

    ISS

  • *****
  • Posts: 23721
  • Joined: 10 Jan 2014
  • Loc: Colorado

Posted 19 September 2019 - 03:06 PM

Miguel have you actually tried to run a large data set using your RAM disk? You MAY find that when using such a "small" swap space that the improvements in your real-world data sets are not as great as shown by the PixInsight benchmark. Note what the PI team said (post #56) about swap space...

 

 

I don't know, but 8GB of RAM disk is a long way from a minimum recommendation of 60GB. It's also possible that you will have runtime issues when using that small of a swap space. But, these problems may only show up after long sessions in PixInsight, meaning several hours with large data sets.

 

But, once again I question whether it is economically wise to do hardware upgrades that may only save a few seconds here and there in operation.

We are also talking about a benchmark that processes continuously, one thing to the next without any pauses. So it is bound to spike CPU usage the entire duration. It is like a user moving their fingers across the keyboard and mouse so fast that it's all just a blur.

 

No one can process an image like the benchmark does. So we are talking 30, 60, 90 seconds of TOTAL compute time across the entire workflow, all merged together into one continuous, non-stop process run in the benchmark. That is so unrealistic and doesn't even remotely represent what a real-world experience going through a PI workflow is actually like.

 

The difference between 90 seconds and even 20 seconds, when all that processing time is distributed out across 30, 60, 90 MINUTES of actual human-interactive processing time, the differences are all but meaningless.

 

I think the last time I ran the benchmark I was around 70 seconds or so. My swap speed is poor in the grand scheme of things. So I sit for 15-20 seconds some times while TGV is running. Maybe 30 seconds for a high iteration count deconvolution. These little periods though, only occur once when I apply the final settings to my whole image. Usually, when I am fiddling, I work with tiny previews that process almost instantly. So overall computer performance, particularly the difference between 1000mb/s, 3000mb/s and 5000mb/s just don't actually come up as a meaningful issue when you are actually sitting there processing an image yourself...100x more slowly than the benchmark.


Edited by Jon Rista, 19 September 2019 - 03:08 PM.

  • pfile and PirateMike like this

#68 jdupton

jdupton

    Surveyor 1

  • *****
  • Posts: 1938
  • Joined: 21 Nov 2010
  • Loc: Central Texas, USA

Posted 19 September 2019 - 03:33 PM

If you don't have enough total swap space, PI will fail.  The keyword here is "total."  PI allows you to enter multiple directory locations for swap, so one can be the RAM disk and the other SSDs or whatever.

 

The only thing I'm not sure of is how PI selects which swap directory location to use; is it by order of listing, is it by remaining free space, etc.

 

   I have found that sometimes there is confusion over the term Swap Space when discussing PixInsight. There are actually two type of Swap Space at play:

  1. PixInsight Swap Directories
    This is where PI saves state information and temporary data during a processing run. You can define multiple Swap Directories and PI's multi-threaded nature can use them for saving state data. Putting multiple Swap Directories on a single hardware device helps as does adding additional Swap Directories on multiple hardware devices.
     
  2. Windows Swap Space
    This is the virtual paging space that Windows uses when applications request RAM space for usage. The Swap Space is used by the OS to hold memory data that applications may not need again immediately. You may have only 32 GB of physical DRAM but windows can tell the application that you have 100 GB or more. It simply moves data as needed from DRAM to drive storage and back to DRAM as needed completely invisible to the application. Windows manages this space by default but you can also configure it to provide more as needed.

   The PixInsight Swap Directories can speed up operations but can give an error if you run completely out of space on the underlying drives in question. Having multiple Swap Directories on multiple drives will prevent this (unless they all fill up).

 

   When PI keeps requesting more OS memory (DRAM) and the virtual page space areas fill up, PI can die with another type of  "Out of Memory" error. This often happens when integrating very large numbers of subs. You can use Task Manager to watch the total amount an paged virtual memory in use in Windows as PI runs a large integration. Once the fixed OS Swap Space fills up, it will cause an error in the application requesting more memory space.

 

   I have had to go in several times and manually increase the Windows Virtual Memory Swap Space and spread it across multiple physical drives in order to integrate large sub collections. During one recent event, I watched as my Virtual Memory Swap Space grew to over 250 GB during ImageIntegration. My system only has 32 GB of physical memory in it.

 

   Any time you see an Out Of Memory error from PI, it may be complaining about DRAM / Virtual Memory. Adding PI Swap directories doesn't help in these cases.

 

 

John


  • pfile and jerahian like this

#69 bobzeq25

bobzeq25

    Hubble

  • *****
  • Posts: 16931
  • Joined: 27 Oct 2014

Posted 19 September 2019 - 04:34 PM

  

   The PixInsight Swap Directories can speed up operations but can give an error if you run completely out of space on the underlying drives in question. Having multiple Swap Directories on multiple drives will prevent this (unless they all fill up).

 

 

John

Question.  Do you think there's any possibility that they would fill up a 512GB SSD?  A 970 PRO 512 is $150.

 

I use that and 32 GB of RAM.  Have never seen a sign that I'm running out of space, either in memory or swap files, or having PI slow down noticeably.  Sometimes stack 200 18MB subs.  183C.  Probably going to do more.

 

A bit less than $2000 for me, and PI _flies_.  23 second benchmark.  With the amount of time I spend in front of a PI screen, and the number of 183 subs I need to shoot, it was money well spent for me.

 

Who knows how long the old man has?  <grin>


  • PirateMike likes this

#70 jdupton

jdupton

    Surveyor 1

  • *****
  • Posts: 1938
  • Joined: 21 Nov 2010
  • Loc: Central Texas, USA

Posted 19 September 2019 - 04:45 PM

Bob,

 

   It depends on what else is on the 512 GB SSD. If you have it mostly dedicated to PI Swap Directories then that will probably never be your limiting factor. If you also have that as your boot drive and Windows' Virtual Memory Page Space, you may need to use increase to another drive if PI throw errors.

 

   I was integrating ~850 23 MB subs when I blew out of space on the drive. I fixed that by adding some PI directories on a spinner and then blew out of Windows Virtual memory which had been reserved on my boot SSD. I added additional OS virtual memory page space on the spinner also and then was able to complete the integration.

 

   Windows seems to be a little more conservative these days in regards to allocating page space on the boot drive. You may want to look at what is being used. If PI has never thrown the "Out Of Memory" error, you are likely safe.

 

 

John


Edited by jdupton, 19 September 2019 - 05:05 PM.

  • PirateMike and bobzeq25 like this

#71 pfile

pfile

    Fly Me to the Moon

  • -----
  • Posts: 5178
  • Joined: 14 Jun 2009

Posted 19 September 2019 - 08:49 PM

it also depends on you - the PI swap space is entirely comprised* of the undo history of all the images you have open in your various workspaces. so if you are using a camera like an 8300M and the images are relatively small, even a bunch of f32 representations of that size sensor is not super gigantic.

 

on the other hand if you are using one of these massive DSLRs or are working with a giant mosaic, each processing step writes out a (usually) uncompressed file to the swap space. at 4 bytes per pixel, and perhaps multiple channels, and then at a huge X-Y resolution it adds up fast.

 

you can turn on swap file compression to ameliorate the disk burden, but of course then the CPU has to work each time it reads or writes the swap which will slow things down.

 

rob

 

*well ok, the swap spaces also contain the backup file of all your process icons as well as the swap files. but the process icon files are tiny.



#72 jerahian

jerahian

    Viking 1

  • *****
  • Posts: 616
  • Joined: 02 Aug 2018
  • Loc: Maine

Posted 19 September 2019 - 11:12 PM

   I have found that sometimes there is confusion over the term Swap Space when discussing PixInsight. There are actually two type of Swap Space at play:

  1. PixInsight Swap Directories
    This is where PI saves state information and temporary data during a processing run. You can define multiple Swap Directories and PI's multi-threaded nature can use them for saving state data. Putting multiple Swap Directories on a single hardware device helps as does adding additional Swap Directories on multiple hardware devices.
     
  2. Windows Swap Space
    This is the virtual paging space that Windows uses when applications request RAM space for usage. The Swap Space is used by the OS to hold memory data that applications may not need again immediately. You may have only 32 GB of physical DRAM but windows can tell the application that you have 100 GB or more. It simply moves data as needed from DRAM to drive storage and back to DRAM as needed completely invisible to the application. Windows manages this space by default but you can also configure it to provide more as needed.

   The PixInsight Swap Directories can speed up operations but can give an error if you run completely out of space on the underlying drives in question. Having multiple Swap Directories on multiple drives will prevent this (unless they all fill up).

 

   When PI keeps requesting more OS memory (DRAM) and the virtual page space areas fill up, PI can die with another type of  "Out of Memory" error. This often happens when integrating very large numbers of subs. You can use Task Manager to watch the total amount an paged virtual memory in use in Windows as PI runs a large integration. Once the fixed OS Swap Space fills up, it will cause an error in the application requesting more memory space.

 

   I have had to go in several times and manually increase the Windows Virtual Memory Swap Space and spread it across multiple physical drives in order to integrate large sub collections. During one recent event, I watched as my Virtual Memory Swap Space grew to over 250 GB during ImageIntegration. My system only has 32 GB of physical memory in it.

 

   Any time you see an Out Of Memory error from PI, it may be complaining about DRAM / Virtual Memory. Adding PI Swap directories doesn't help in these cases.

 

 

John

Excellent explanation, John.  Apologies, I should have been more clear in that I was strictly speaking about the PI "swap storage directories" as found in the Global Preferences.



#73 PirateMike

PirateMike

    Mercury-Atlas

  • -----
  • topic starter
  • Posts: 2634
  • Joined: 27 Sep 2013
  • Loc: A Green Dot On A Blue Sea

Posted 19 September 2019 - 11:16 PM

Question.  Do you think there's any possibility that they would fill up a 512GB SSD?  A 970 PRO 512 is $150.

 

I use that and 32 GB of RAM.  Have never seen a sign that I'm running out of space, either in memory or swap files, or having PI slow down noticeably.  Sometimes stack 200 18MB subs.  183C.  Probably going to do more.

 

A bit less than $2000 for me, and PI _flies_.  23 second benchmark.  With the amount of time I spend in front of a PI screen, and the number of 183 subs I need to shoot, it was money well spent for me.

 

Who knows how long the old man has?  <grin>

 

And who know how long this old man has? Hopefully long enough to see if all this stuff will work.

 

 

AMD Threadripper 2920X 3.5 GHz 12-Core Processor

 

ASRock Fatal1ty X399 Professional Gaming ATX TR4 Motherboard

 

Samsung 970 Evo Plus 1 TB M.2-2280 NVME Solid State Drive

 

Corsair Vengeance LPX DDR4-3200 Memory, 128 Gig

 

 

And don't let that guy Robert Zeq fool you. He is, and always has been only 25 years old. grin.gif

 

 

Miguel   8-)

 

.


Edited by PirateMike, 19 September 2019 - 11:26 PM.

  • bobzeq25 likes this

#74 Cfreerksen

Cfreerksen

    Messenger

  • -----
  • Posts: 444
  • Joined: 18 Aug 2018
  • Loc: Tooele, Ut

Posted 20 September 2019 - 08:52 AM

I got a nice boost by creating multiple swap spaces on my fastest ssd. Overclocked the cpu a bit. I will post a link if you don't find it on your own about the multiple swap files. Got to go to work.

Chris

#75 bobzeq25

bobzeq25

    Hubble

  • *****
  • Posts: 16931
  • Joined: 27 Oct 2014

Posted 20 September 2019 - 09:24 AM


 

AMD Threadripper 2920X 3.5 GHz 12-Core Processor

 

ASRock Fatal1ty X399 Professional Gaming ATX TR4 Motherboard

 

Samsung 970 Evo Plus 1 TB M.2-2280 NVME Solid State Drive

 

Corsair Vengeance LPX DDR4-3200 Memory, 128 Gig

 

Good choices, assuming you're planning to do a RAMDisk for swap.  Otherwise you might want two SSD drives, one mostly dedicated to swap.

 

And you'll want something to archive old data with.  I use a 4TB spinning disk, and 4TB is not lavish.

 

Interesting thing, but unimportant.  I wonder if my slower 16 core 1950X would work, compared to your 2920X? 

 

Maybe not so unimportant.  With your budget, I'd get a 2950X with 16 cores.  It will be about 33% faster when processing subs, some unknown (but likely significant) amount faster otherwise.  You'd be processing 32 subs at once, instead of 24.  Seems like it would be cost effective.


Edited by bobzeq25, 20 September 2019 - 09:28 AM.

  • PirateMike likes this


CNers have asked about a donation box for Cloudy Nights over the years, so here you go. Donation is not required by any means, so please enjoy your stay.


Recent Topics






Cloudy Nights LLC
Cloudy Nights Sponsor: Astronomics