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

AMD ThreadRipper 3990x Review/Thoughts

  • Please log in to reply
27 replies to this topic

#1 stobiewankenobi

stobiewankenobi

    Mariner 2

  • -----
  • topic starter
  • Posts: 272
  • Joined: 25 Apr 2019
  • Loc: Laguna Beach, CA

Posted 25 February 2021 - 02:30 PM

Hello all,

 

I recently upgraded my PC hardware in a few ways.

 

1: AMD RYZEN 9 3900X to a ThreadRipper 3990x

2: SATA SSDs to NVME

 

I wanted to post my experience with PixInsight benchmarks comparison between them, as well as real world feel for performance.

 

Benchmarks

Ryzen + RamDisk/SATA PixInsight Benchmark

Ryzen Benchmark

Total: 7488

CPU: 15629

SWAP: 2383

 

 

Threadripper + NVME PixInsight Benchmark

ThreadRipper Benchmark

Total: 17014

CPU: 26696

Swap: 6857

 

Threadripper + NVME with Swap dirs added (10 on NVME) PixInsight Benchmark

Threadripper With Swap Improvements
Total: 23304
CPU: 26440
SWAP: 15847

 

 

The results are pretty impressive. Almost 3x the performance with the Ryzen + NVME disks.

 

 

Real World Feel

This is obviously what is most important, benchmarks are not really a great representation of how long things take for various processes. I did not do a super specific timing test between the two, but I have processed enough to know generally what feels much improved after the upgrade. Below are a few examples of common things I do in my flow and what feels faster or the same.

 

Image: Andromeda

Camera: QHY 247c

Drizzle: No

Image Size: 142mb

 

This is all with NVME + ThreadRipper (no RamDisk)

I am unable to get any ramdisk utility to actually start (GUI just opens then crashes with no error) at the moment, so I'm not able to test yet with the Threadripper + RamDisk. Will update this thread if I can get that working with the new hardware.

 

ABE with Division == Same (surprisingly)

ABE with Subtraction == Faster (definitely)

EZProcessingSuite DeNoise ==  Faster, definitely faster but didn't blow me away (as I was hoping it would). I'd estimate it is something like 15% faster

Starnet++ == Much faster, but about the same when you're using the GPU settings to do GPU star removal. So overall I'd say GPU and Threadripper are the same, but Threadripper vs Ryzen is much faster on Threadripper

HDRMultiScale == Slightly Faster (sad face)

LocalHistogramEquilization == Much faster (surprisingly)

MultiScaleLinearTransform == Insanely faster (very very insanely faster)

 

Overall I am definitely noticing my workflow runs faster, even just small things like curves stretch, auto transform are just instant vs a few seconds. Definitely feels much snappier to me. Worth $4,000 for just the CPU? I'm not sure yet to be honest. I think you can probably get pretty comparable results with the cheaper Threadrippers. 

 

 

-----------------------------------------------------------------------------------------------------------------

 

If anyone here has any ideas on other tests to do, or benchmarks we can run to see how this performs (vs other CPU's) I am happy to run anything anyone has!


Edited by stobiewankenobi, 25 February 2021 - 03:02 PM.


#2 khursh

khursh

    Viking 1

  • ****-
  • Posts: 919
  • Joined: 23 Jan 2018

Posted 25 February 2021 - 02:50 PM

You can probably get better swap performance. Not sure how this affects real world performance though. I have a lowly Ryzen 3700 which is way worse than yours in terms of CPU performance, but my swap numbers are triple to quadruple with a single 500Gb NVME. My 'total performance' is not much worse than yours at ~16000. Creating multiple swap files (8 if I remember correctly) does the trick.



#3 stobiewankenobi

stobiewankenobi

    Mariner 2

  • -----
  • topic starter
  • Posts: 272
  • Joined: 25 Apr 2019
  • Loc: Laguna Beach, CA

Posted 25 February 2021 - 02:52 PM

You can probably get better swap performance. Not sure how this affects real world performance though. I have a lowly Ryzen 3700 which is way worse than yours in terms of CPU performance, but my swap numbers are triple to quadruple with a single 500Gb NVME. My 'total performance' is not much worse than yours at ~16000. Creating multiple swap files (8 if I remember correctly) does the trick.

Are you creating swap files just on your NVME? I will add that and test. 



#4 stobiewankenobi

stobiewankenobi

    Mariner 2

  • -----
  • topic starter
  • Posts: 272
  • Joined: 25 Apr 2019
  • Loc: Laguna Beach, CA

Posted 25 February 2021 - 02:58 PM

You can probably get better swap performance. Not sure how this affects real world performance though. I have a lowly Ryzen 3700 which is way worse than yours in terms of CPU performance, but my swap numbers are triple to quadruple with a single 500Gb NVME. My 'total performance' is not much worse than yours at ~16000. Creating multiple swap files (8 if I remember correctly) does the trick.

Yeah sure enough just creating more swap directories on the NVME massively improved performance, I'll update original post. 

 

Threadripper With Swap Improvements


#5 MikiSJ

MikiSJ

    Surveyor 1

  • *****
  • Posts: 1,818
  • Joined: 24 Feb 2006
  • Loc: San Jose, CA

Posted 25 February 2021 - 03:25 PM

There was a thread a couple of days ago regarding tuning PixInsight. I followed the OP instructions and got a modest increase in overall speed. I added 3 additional swap files but I did it on Drive D which is a SATA drive. My laptop has a SSD Drive C so if I had added the swap files to the SSD drive the increase would likely have been greater.

 

Late last year I went from an Intel i3-7100U @ 2.40GHz processor (SATA drive) total score of 1,390 to a AMD Ryzen 7 3700U with Radeon Vega Mobile Gfxtotal (SSD drive) total score of 5,880 and was very pleased with the fourfold plus increase in throughput.

 

The FITS file sizes I process are 22,846kB and it takes seconds to process 20 files so I am not too eager to pay a lot more money for a minimal decrease in waiting time for what I use in my workflow.

 

The AMD Ryzen is definitely the way to go if your application takes advantage of multi-threading. The Ryzen 7 3700U has 8 logical processors and the Intel i3-7100 has 4 logical processors with similar clock speeds (2.3mHz vs 2.4mHz).

 

I don't do games so I don't know if the number of actual/logical processors affects gaming as much as it does with PixInsight. 



#6 stobiewankenobi

stobiewankenobi

    Mariner 2

  • -----
  • topic starter
  • Posts: 272
  • Joined: 25 Apr 2019
  • Loc: Laguna Beach, CA

Posted 25 February 2021 - 04:46 PM

There was a thread a couple of days ago regarding tuning PixInsight. I followed the OP instructions and got a modest increase in overall speed. I added 3 additional swap files but I did it on Drive D which is a SATA drive. My laptop has a SSD Drive C so if I had added the swap files to the SSD drive the increase would likely have been greater.

 

Late last year I went from an Intel i3-7100U @ 2.40GHz processor (SATA drive) total score of 1,390 to a AMD Ryzen 7 3700U with Radeon Vega Mobile Gfxtotal (SSD drive) total score of 5,880 and was very pleased with the fourfold plus increase in throughput.

 

The FITS file sizes I process are 22,846kB and it takes seconds to process 20 files so I am not too eager to pay a lot more money for a minimal decrease in waiting time for what I use in my workflow.

 

The AMD Ryzen is definitely the way to go if your application takes advantage of multi-threading. The Ryzen 7 3700U has 8 logical processors and the Intel i3-7100 has 4 logical processors with similar clock speeds (2.3mHz vs 2.4mHz).

 

I don't do games so I don't know if the number of actual/logical processors affects gaming as much as it does with PixInsight. 

Yeah PI does handle threading fairly well per their eng team, I have read a few things about it (which is what lead me to purchase the threadripper). You could theoretically run crysis just on a threadripper on high settings (although that'd probably be a bad idea due to the bad heat management with the CPU today). This computer is solely for image processing, so not really any gaming going on on it. 



#7 KTAZ

KTAZ

    Surveyor 1

  • *****
  • Posts: 1,930
  • Joined: 09 Apr 2020
  • Loc: Scottsdale, AZ

Posted 25 February 2021 - 05:19 PM

Yeah sure enough just creating more swap directories on the NVME massively improved performance, I'll update original post. 

 

Can you give some more guidance on the creation of the additional swap files? I will certainly try this as a performance boost; and how large are the swap files that you created?



#8 stobiewankenobi

stobiewankenobi

    Mariner 2

  • -----
  • topic starter
  • Posts: 272
  • Joined: 25 Apr 2019
  • Loc: Laguna Beach, CA

Posted 25 February 2021 - 05:22 PM

Can you give some more guidance on the creation of the additional swap files? I will certainly try this as a performance boost; and how large are the swap files that you created?

Hey there, I think my messaging was slightly misleading. By "creating more swap directories" I am specifically referencing something in PixInsight. To do this you do the following

 

Click edit -> Global Preferences -> Directories and Network -> Add 

 

I click add around 10 times and specify an empty dir on my NVME drive, this just gives PixInsight more space to create swap files across the NVME thus improving performance. A lot of folks do with with RAMDISKs as well which would create the directory directly in memory, which is faster even than NVME, but I can't seem to get RAMDISK working on my new build, so am doing this on NVME which is still definitely an improvement. 


  • KTAZ likes this

#9 WoodlandsAstronomer

WoodlandsAstronomer

    Ranger 4

  • -----
  • Posts: 338
  • Joined: 26 Aug 2019
  • Loc: Texas

Posted 25 February 2021 - 05:23 PM

I run the thread ripper and it is insane for all manner of stacking and post processing
  • stobiewankenobi likes this

#10 LauraMS

LauraMS

    Ranger 4

  • -----
  • Posts: 365
  • Joined: 29 Mar 2011
  • Loc: Germany

Posted 26 February 2021 - 01:45 AM

Your 3990x should be able to do more. Can you describe a little more detail. Ram size, type of Nvme-ssd: PCie3.0 or 4.0.

 

Without much tweaking I got this benchmark on my Threadripper 3970x, 128GB RAM, MP600 2TB Pcie4.0 SSD: https://pixinsight.c...9E7JAQFIR392MFV

 

Total score around 33000, runtime 14sec

 

This was on Windows 10Pro with 8 swap directories on the SSD. More Swap directories (up to 64 tested) didn't improve things.

 

Key thing is Ram size (at least 2 GB/core) and quick and large Pcie4.0 SSD.

 

If you want quicker, go Linux: same system but Ubuntu 20: https://pixinsight.c...IJ2P8LD5SJCUX8U

 

Total score around 40000, total runtime 12sec.

 

Total runtime is decreasing exponentially with speed, i.e. diminishing returns. 

 

Btw.,  on the Linux benchmarks I entered the wrong ssd type: it was a Samsung EVOPlus 4TB (I think 970). My system has 2TB Pcie4.0 (MP600) and 4TB Samsung 970 EvoPlus PCie3.0 for storage. 

 

On your threadripper 3990x you should be able to get higher benchmarks. Can you describe RAM and SSD? 

 

BTW, does PI's Starnet use your GPU? I haven't been able to use my 3070  with Starnet last time when I tested in January. 

 

Best, Laura 



#11 LauraMS

LauraMS

    Ranger 4

  • -----
  • Posts: 365
  • Joined: 29 Mar 2011
  • Loc: Germany

Posted 26 February 2021 - 02:19 AM

@StobiWankenObi,

 

looking at your benchmarks, and comparing mine (Windows 10):

 

Total performance 32773
CPU performance 36157
Swap performance 24071

 

In particular your swap performance appears a bit on the low side. There is something slowing it down (i.e. slow disk). If you have enough Ram (128GB or more) you could try swap disks (try 8 or 10) on ram disk. BTW., 'slow disk' is relative here - we are talking high end. But for a HEDT (high end desktop system) like your dream Threadripper 3990x you need an environment supporting it. And maybe some tuning.

 

 

Your CPU scores should be higher as well - do you observe thermal throttling? Or maybe more RAM needed.

 

 

Looking from a different perspective: If I compare your and my benchmarks with my old laptop computer with total scores of a few thousand and benchmark runtimes of 5min and more, our systems are a very fine improvement, I tend to think smile.gif

 

 

 

Best wishes and have much fun 

Laura 


Edited by LauraMS, 26 February 2021 - 02:34 AM.


#12 MikiSJ

MikiSJ

    Surveyor 1

  • *****
  • Posts: 1,818
  • Joined: 24 Feb 2006
  • Loc: San Jose, CA

Posted 26 February 2021 - 02:32 AM

One of the issues I found with creating more swap directories is that PI doesn't cleanup the directories, leaving a lot of fairly large file to be manually cleaned up.



#13 LauraMS

LauraMS

    Ranger 4

  • -----
  • Posts: 365
  • Joined: 29 Mar 2011
  • Loc: Germany

Posted 26 February 2021 - 02:42 AM

One of the issues I found with creating more swap directories is that PI doesn't cleanup the directories, leaving a lot of fairly large file to be manually cleaned up.

I checked this when doing the benchmarks but afterwards directories were empty. 

 

Never checked after that - maybe I should after having worked with PI ... thanks for the info! 



#14 KTAZ

KTAZ

    Surveyor 1

  • *****
  • Posts: 1,930
  • Joined: 09 Apr 2020
  • Loc: Scottsdale, AZ

Posted 26 February 2021 - 08:43 AM

I increased my swaps to 10 and my swap score increased from around 3900 to over 21000. Looking forward to playing with it today to see if I got a performance increase.



#15 stobiewankenobi

stobiewankenobi

    Mariner 2

  • -----
  • topic starter
  • Posts: 272
  • Joined: 25 Apr 2019
  • Loc: Laguna Beach, CA

Posted 26 February 2021 - 10:27 AM

@StobiWankenObi,

 

looking at your benchmarks, and comparing mine (Windows 10):

 

Total performance 32773
CPU performance 36157
Swap performance 24071

 

In particular your swap performance appears a bit on the low side. There is something slowing it down (i.e. slow disk). If you have enough Ram (128GB or more) you could try swap disks (try 8 or 10) on ram disk. BTW., 'slow disk' is relative here - we are talking high end. But for a HEDT (high end desktop system) like your dream Threadripper 3990x you need an environment supporting it. And maybe some tuning.

 

 

Your CPU scores should be higher as well - do you observe thermal throttling? Or maybe more RAM needed.

 

 

Looking from a different perspective: If I compare your and my benchmarks with my old laptop computer with total scores of a few thousand and benchmark runtimes of 5min and more, our systems are a very fine improvement, I tend to think smile.gif

 

 

 

Best wishes and have much fun 

Laura 

Yeah as I mentioned I can't get SWAP disk to work on this new machine. Even Dataram (the guys who make some of the bigger ramdisk softwares) were stumped. I'm not a windows guy and generally use Linux at work so this is a bit new to me in general, but cant' really debug someone elses closed source software. So I am using NVME with 10 swap directories created in PixInsight. I am only running 32gb of memory, but have some in the main to upgrade to 128gb, so we'll see how that improves things to get more memory per core for processing. 

 

 

Your 3990x should be able to do more. Can you describe a little more detail. Ram size, type of Nvme-ssd: PCie3.0 or 4.0.

 

Without much tweaking I got this benchmark on my Threadripper 3970x, 128GB RAM, MP600 2TB Pcie4.0 SSD: https://pixinsight.c...9E7JAQFIR392MFV

 

Total score around 33000, runtime 14sec

 

This was on Windows 10Pro with 8 swap directories on the SSD. More Swap directories (up to 64 tested) didn't improve things.

 

Key thing is Ram size (at least 2 GB/core) and quick and large Pcie4.0 SSD.

 

If you want quicker, go Linux: same system but Ubuntu 20: https://pixinsight.c...IJ2P8LD5SJCUX8U

 

Total score around 40000, total runtime 12sec.

 

Total runtime is decreasing exponentially with speed, i.e. diminishing returns. 

 

Btw.,  on the Linux benchmarks I entered the wrong ssd type: it was a Samsung EVOPlus 4TB (I think 970). My system has 2TB Pcie4.0 (MP600) and 4TB Samsung 970 EvoPlus PCie3.0 for storage. 

 

On your threadripper 3990x you should be able to get higher benchmarks. Can you describe RAM and SSD? 

 

BTW, does PI's Starnet use your GPU? I haven't been able to use my 3070  with Starnet last time when I tested in January. 

 

Best, Laura 

Yeah I am not using Linux on this machine at the moment, other software I use (NINA) doesn't support linux unfortunately or I'd be running Arch or Fedora for sure. 

 

Ram == 32GB

SSD == NVME x 2 2tb, I have swap created under one of the NVME as I called out, and that did massively improve performance. Reading through this thread though it seems I'm still not getting as much performance as I can be. I am wondering if it's RAM (only 32GB). I have some more in the mail so we'll see if that improves things significantly or not. 

 

 

I can say though that memory is barely even touched (at least during the benchmark) with still around 20gb available and not cached. 


Edited by stobiewankenobi, 26 February 2021 - 10:28 AM.


#16 jdupton

jdupton

    Gemini

  • *****
  • Posts: 3,449
  • Joined: 21 Nov 2010
  • Loc: Central Texas, USA

Posted 26 February 2021 - 11:25 AM

stobiewankenobi,

 

   What form does the current 32 GB of RAM take? Is it a single DIMM, two DIMM set, or four DIMM set?

 

   This can make a big difference with your CPU. Your Threadripper CPU has a NUMA-like architecture. (NUMA = Non-Uniform Memory Access. Each CPU block can only access certain DIMM sockets directly. To get to other RAM, a CPU request may need to be routed across other chips in the CPU package making the memory access slower.)

 

   If the total system memory is all connected to a single memory controller inside the CPU package, then your overall system speed will suffer. Your motherboard should have multiple sets of memory DIMM sockets. Once you get more memory and properly plug it in so that each controller has direct access to some portion of the total local RAM, you will see your CPU Benchmark speed improve quite a bit.

 

 

John


Edited by jdupton, 26 February 2021 - 11:32 AM.


#17 stobiewankenobi

stobiewankenobi

    Mariner 2

  • -----
  • topic starter
  • Posts: 272
  • Joined: 25 Apr 2019
  • Loc: Laguna Beach, CA

Posted 26 February 2021 - 11:31 AM

stobiewankenobi,

 

   What form does the current 32 GB of RAM take? Is it a single DIMM, two DIMM set, or four DIMM set?

 

   This can make a big difference with your CPU. Your Threadripper CPU has a NUMA-like architecture. (NUMA = Non-Uniform Memory Access. Each CPU block can only access certain DIMM sockets directly. To get to other RAM, a CPU request may need to be routed across other chips in the CPU package making the memory access slower.)

 

   If the total system memory is all connected to a single die inside the CPU package, then your overall system speed will suffer. Your motherboard should have multiple sets of memory DIMM sockets. Once you get more memory and properly plug it in so that each die has direct access to some portion of the total local RAM, you will see your CPU Benchmark speed improve quite a bit.

 

 

John

Right now I have 4x8gb, but am reviewing and the speed is only 2800 on these things. It's older memory from a previous build. I have around 64gb in the mail coming that are 4400 and will be spread across 8 sockets, so that'll be interesting to test. 

 

This is helpful! Thanks Jdupton. 



#18 SteveL42

SteveL42

    Vostok 1

  • *****
  • Posts: 108
  • Joined: 18 Sep 2020
  • Loc: Colorado

Posted 26 February 2021 - 05:37 PM

What version of Windows are you running?  Pro has issues with that many threads, Enterprise might be what you need.

 

https://www.anandtec...-3990x-review/3


Edited by SteveL42, 26 February 2021 - 10:41 PM.


#19 LauraMS

LauraMS

    Ranger 4

  • -----
  • Posts: 365
  • Joined: 29 Mar 2011
  • Loc: Germany

Posted 26 February 2021 - 05:38 PM

I don't think the 2800 DIMM specification is the major issue. 32 GB is probably on the low side, in particular for a 64 core/128 thread CPU. It needs data food quickly :-) And since it is dual  quad channel (four memory channels which have direct access to the processor dies simultaneously) , install/upgrade always in steps of four memory chips.

 

What NVMe drives do you have? 


Edited by LauraMS, 26 February 2021 - 05:45 PM.

  • jdupton likes this

#20 jdupton

jdupton

    Gemini

  • *****
  • Posts: 3,449
  • Joined: 21 Nov 2010
  • Loc: Central Texas, USA

Posted 26 February 2021 - 07:19 PM

stobiewankenobi,

 

   As LauraMS indicates above, the "2800 MHz" memory speed is not a big issue. You may run into some configuration issues if plan to add the new 64 GB to the existing DRAM. Mixing speeds of memory can be very tricky.

 

   Regarding the 4400 MHz memory speed of your ordered DRAM, be aware that the AMD Ryzen Threadripper 3990X memory interface is only rated to run at 3200 MHz. See the AMD specifications page for verification. You can still run the RAM at its full rated speed but will not be able to simply tell the system to run at 4400 MHz. There is more to it than that. It is actually the access speeds of the DRAM that matter and not just the frequency of the interface.

 

   If you choose to leave the original DIMMs in place, you need to run the memory at no faster than 2800 MHz. Otherwise you will be plagued by system crashes and errors in trying to use the system. If you use only the 64 GB on order, you can run the memory interface at 3200 MHz while still taking full advantage of the memory speed but will need to make some changes to the memory timing settings in BIOS in order to do it.

 

   (Depending on BIOS, you may be able to simply select a 3200 XMP profile setting and have everything work. Many CPU chips can be frequency overclocked to near 3600 MHz but AMD does not guarantee that to always work. Attempting to go above 3600 MHz will be likely to slow everything down as the Infinity Fabric Bus inside the chip will drop back into a ratio'ed mode rather than running 1:1. In general, I always stay away from voltage overclocking. I always feel it is better to keep all settings within the specifications recommended by AMD. That can be done and still give the full performance of the 4400 MHz DRAM while running it at lower frequency but with fewer cycle wait states.)

 

 

John


Edited by jdupton, 26 February 2021 - 07:26 PM.

  • LauraMS likes this

#21 stobiewankenobi

stobiewankenobi

    Mariner 2

  • -----
  • topic starter
  • Posts: 272
  • Joined: 25 Apr 2019
  • Loc: Laguna Beach, CA

Posted 26 February 2021 - 07:41 PM

stobiewankenobi,

 

   As LauraMS indicates above, the "2800 MHz" memory speed is not a big issue. You may run into some configuration issues if plan to add the new 64 GB to the existing DRAM. Mixing speeds of memory can be very tricky.

 

   Regarding the 4400 MHz memory speed of your ordered DRAM, be aware that the AMD Ryzen Threadripper 3990X memory interface is only rated to run at 3200 MHz. See the AMD specifications page for verification. You can still run the RAM at its full rated speed but will not be able to simply tell the system to run at 4400 MHz. There is more to it than that. It is actually the access speeds of the DRAM that matter and not just the frequency of the interface.

 

   If you choose to leave the original DIMMs in place, you need to run the memory at no faster than 2800 MHz. Otherwise you will be plagued by system crashes and errors in trying to use the system. If you use only the 64 GB on order, you can run the memory interface at 3200 MHz while still taking full advantage of the memory speed but will need to make some changes to the memory timing settings in BIOS in order to do it.

 

   (Depending on BIOS, you may be able to simply select a 3200 XMP profile setting and have everything work. Many CPU chips can be frequency overclocked to near 3600 MHz but AMD does not guarantee that to always work. Attempting to go above 3600 MHz will be likely to slow everything down as the Infinity Fabric Bus inside the chip will drop back into a ratio'ed mode rather than running 1:1. In general, I always stay away from voltage overclocking. I always feel it is better to keep all settings within the specifications recommended by AMD. That can be done and still give the full performance of the 4400 MHz DRAM while running it at lower frequency but with fewer cycle wait states.)

 

 

John

My understanding has always been that mixing ram of the same type (DDR4 with DDR4) would be fine even if the voltage is off as it will just clock down the speed to whatever the lowest value is automatically. I haven't had issues in the past. 

Either way, I wasn't planning on keeping the old RAM in the system, and will just be using the new 64GB. 

 

 

Thanks for the input here. Once I get my new memory in here, as well as my other NVME disk that is one of the WD blacks that has like 7000/5000 drives (currently just using my older nvme drive) it'll be fun to test and see where we are at. 

 

 



#22 LauraMS

LauraMS

    Ranger 4

  • -----
  • Posts: 365
  • Joined: 29 Mar 2011
  • Loc: Germany

Posted 26 February 2021 - 11:52 PM

With your new 8 memory chips coming you probably won't have a chance to mix them mit the old one since Threadripper boards have only 8 sockets anyway. In my system I use DIMMs of 3200 specification because that is what Threadrippers are specified for. I suggest you try running your new memory with that speed at the beginning because of the Threadripper specs.

Personally I would never play around with only  5-10% improvement (I think that is what you can expect from increasing memory speed beyond specs) if there would be the slightest risk of introducing any instability into my system. So I never went with overclocking, neither CPU nor memmory.

 

If the numbers you give for your new drives are transfer speed that sounds great!

 

Looking forward to hearing from your experiences with your upgrades. 


Edited by LauraMS, 26 February 2021 - 11:54 PM.

  • psandelle likes this

#23 stobiewankenobi

stobiewankenobi

    Mariner 2

  • -----
  • topic starter
  • Posts: 272
  • Joined: 25 Apr 2019
  • Loc: Laguna Beach, CA

Posted 27 February 2021 - 10:09 AM

With your new 8 memory chips coming you probably won't have a chance to mix them mit the old one since Threadripper boards have only 8 sockets anyway. In my system I use DIMMs of 3200 specification because that is what Threadrippers are specified for. I suggest you try running your new memory with that speed at the beginning because of the Threadripper specs.

Personally I would never play around with only  5-10% improvement (I think that is what you can expect from increasing memory speed beyond specs) if there would be the slightest risk of introducing any instability into my system. So I never went with overclocking, neither CPU nor memmory.

 

If the numbers you give for your new drives are transfer speed that sounds great!

 

Looking forward to hearing from your experiences with your upgrades. 

Ah yeah good point Laura, thank you. Yeah I will post updates. Thanks!



#24 MikiSJ

MikiSJ

    Surveyor 1

  • *****
  • Posts: 1,818
  • Joined: 24 Feb 2006
  • Loc: San Jose, CA

Posted 27 February 2021 - 06:06 PM

In a post above I had added 3 additional swap files on a SATA drive in my laptop. There was a modest increased in the benchmark numbers.

 

Performance Indices    1 Swap file  4 Swap file  10 Swap file 
Total performance ......    5191          5880          6538
CPU performance ........   5684           6643          6610
Swap performance .......  3848           4009          6317

 

Two conclusions: 1) more swap files increase performance, and 2) having swap files on a SSD disk increases swap performance significantly.



#25 stobiewankenobi

stobiewankenobi

    Mariner 2

  • -----
  • topic starter
  • Posts: 272
  • Joined: 25 Apr 2019
  • Loc: Laguna Beach, CA

Posted 27 February 2021 - 09:59 PM

Pretty interesting results here. 

 

I got my new NVME which is the WD Black that supports 7,000 mb/s read, got 96GB of the new RAM in . Everything hooked up and looking good. Results are not that far off from what I was seeing before. SWAP Directories are created on the new NVME of course. But am only seeing a tiny amount of performance increase. 

I am considering just dual booting and being done with Windows, I hate windows, and I can always swap if I need to hop into NINA, I am wondering how much of this issue is related to Windows itself. By "issue" I of course just mean seeing reduced performance benchmarks on a system that should see superior benchmarks. (obviously still very thrilled with current benchmarks). 

 

Latest Benchmark Run

 

Total: 21,777

CPU: 23,326

SWAP: 17,302

 

I will boot up Fedora or Ubuntu really quick and see what's up over on the Linux side of things, and maybe I can just emulate a VM for NINA if I need to (although idk how ASCOM plays with emulation or VM's). 




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