Re: Subject: Computer for image analysis

[Oliver Biehlmaier-2]

Dear Arvydas,
I equipped an entire image analysis room with new Image analysis machines about 1.5 years ago. During the evaluation, our main focus was on the system's performance using software such as Imaris, Volocity, Huygens, Fiji, etc.
As already posted in other replies to your email it turns out that GPU is important, but bottlenecks are CPU, RAM, and the speed of the HDD.
As our institute's IT asked us to go for a Dell-solution, we evaluated several possibilities from Dell. We ended up buying 2 Dell Precision with 3GB-GPU, XEON-processors and between 24 to 48GB of RAM, and many "pimped" Optiplex systems where we installed 3GB-GPU, the max. RAM (16GB), an SSD for the OS and swapping and a fast 500GB-HDD for saving the data.
Price wise the Optiplex systems sum up to a third of the price of the precision.
The main reason for the Optiplex was the i7 processor which is capable to do overclocking which is not possible on the XEON systems. We expected this to be a key advantage in comparison to our expensive Precision systems.
Now, after 1,5 years of usage I can confirm that this fully worked out. As many programs (especially Imaris) are still mainly relying on only one but definitely not on all cores, the overclocking feature of the i7 system usually keeps them at the same level or even outperforms the Precision systems. Only the 48GB-RAM system is a bit faster on the rare occasions when it can fully profit from the large RAM (large time lapse or stitching tasks). But even then the fast swapping onto the SDDs on the Optiplex keeps them almost at the same level of performance.
Only recently we ran into some minor problems with our ATI graphics cards which could have been prevented by using NVIDIA cards, thus I would recommend the latter. There is definitely no need to go for Quadra cards, they are super expensive and receive less updates and patches than the gaming cards.
I hope this helps you in your decision for your new systems.
Best,
Oliver


----------------------------------------------------------------
Oliver Biehlmaier, PhD
Head of Imaging Core Facility
Biozentrum, University of Basel
Klingelbergstrasse 50/70
4056 Basel
Switzerland

Tel:        [phone]
Email:     [hidden email]<mailto:[hidden email]>
http://www.biozentrum.unibas.ch/imcf
----------------------------------------------------------------

_________________
From: Arvydas Matiukas <[hidden email]<mailto:[hidden email]>>
To: [hidden email]<mailto:[hidden email]>=20
Sent: Friday, March 8, 2013 12:24 PM
Subject: Computer for image analysis

*****
To join, leave or search the confocal microscopy listserv, go to:
http://lists.umn.edu/cgi-bin/wa?A0=3Dconfocalmicroscopy
*****

Dear listers/microscopists,

I assume there is good time to update new trends in
image analysis hardware. The last discussions on image
analysis computer were in 2006-8. Though the basic
principles of CPU, RAM, hard drive, video card, monitor
selection still hold some new types of hardware became
popular/available, e.g. SSD drives, APU, water cooling.
Now a decent gaming computer (~$1k) has the processing power
of a 2006 expensive workstation (~$20K). I was suprised that
I was able to completely overhaul my 8 year old ATX case
to a quad core 2GHz APU, 8GB 1600MHz RAM, 160GB SATA-2
SSD, water cooling, USB3 and SATA3 Gigabyte motherboard,
and 4 monitor 1GB video card.
for under $300 (online, after rebates).

Now I am wiling to upgrade/overhaul my work computer which
is used to run ImageJ, Fiji, Deconvolution (Autoquant, Huygens),
Matlab, PV-Vawe, Labview, Origin. Please advice/share you thoughts
what best configuration is possible to buy for $2-3k (monitor
excluded).
My first choice would be  to go with a fast gaming computer, e.g.
Dell-Alienware Aurora=20
Windows* 7 Ultimate, 64Bit, English
2nd Generation Intel* Core* i7-3820 (10M Cache, Overclocked up to 4.1
GHz)
16GB (4 X 4GB) Quad Channel DDR3 at 1600MHz
NVIDIA* GeForce* GTX 660 1.5GB GDDR5
1TB RAID 0 (2x 500GB SATA 6Gb/s) Solid State Hybrid
19-in-1 Media Card Reader
No Monitor
Integrated 7.1 Channel Audio

The second  choice would be to buy all components online and
build a computer myself (I have done this about 50 times over
25 years). This option typically saves money or buys better
components,
and provides you full specs of the hardware. The con of this
approach is that it wastes some of your time to debug/make all
the hardware work together and with your software. However,
as the computer is for me not just a box but a tool I am ready
to make this sacrifice.

BTW, is there any solid preference towards CPU Type (Intel ix/AMD/Intel
Xeon)

Thanks for your input/advice/thoughts,
Arvydas
--------------------

[Watkins, Simon C]

*****
To join, leave or search the confocal microscopy listserv, go to:
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
*****

So Oliver, what you are saying is that the ultimate bottleneck is the CPU
speed, followed by RAM, followed by CPU core count and finally graphics
card capabilities?

Simon Watkins Ph.D

Professor and Vice Chair Cell Biology
Professor Immunology
Director Center for Biologic Imaging
University of Pittsburgh
Bsts 225 3550 terrace st
Pittsburgh PA 15261
Www.cbi.pitt.edu <http://Www.cbi.pitt.edu/>
412-352-2277






On 3/9/13 3:39 AM, "Oliver Biehlmaier" <[hidden email]> wrote:

>Dear Arvydas,
>I equipped an entire image analysis room with new Image analysis machines
>about 1.5 years ago. During the evaluation, our main focus was on the
>system's performance using software such as Imaris, Volocity, Huygens,
>Fiji, etc.
>As already posted in other replies to your email it turns out that GPU is
>important, but bottlenecks are CPU, RAM, and the speed of the HDD.
>As our institute's IT asked us to go for a Dell-solution, we evaluated
>several possibilities from Dell. We ended up buying 2 Dell Precision with
>3GB-GPU, XEON-processors and between 24 to 48GB of RAM, and many "pimped"
>Optiplex systems where we installed 3GB-GPU, the max. RAM (16GB), an SSD
>for the OS and swapping and a fast 500GB-HDD for saving the data.
>Price wise the Optiplex systems sum up to a third of the price of the
>precision.
>The main reason for the Optiplex was the i7 processor which is capable to
>do overclocking which is not possible on the XEON systems. We expected
>this to be a key advantage in comparison to our expensive Precision
>systems.
>Now, after 1,5 years of usage I can confirm that this fully worked out.
>As many programs (especially Imaris) are still mainly relying on only one
>but definitely not on all cores, the overclocking feature of the i7
>system usually keeps them at the same level or even outperforms the
>Precision systems. Only the 48GB-RAM system is a bit faster on the rare
>occasions when it can fully profit from the large RAM (large time lapse
>or stitching tasks). But even then the fast swapping onto the SDDs on the
>Optiplex keeps them almost at the same level of performance.
>Only recently we ran into some minor problems with our ATI graphics cards
>which could have been prevented by using NVIDIA cards, thus I would
>recommend the latter. There is definitely no need to go for Quadra cards,
>they are super expensive and receive less updates and patches than the
>gaming cards.
>I hope this helps you in your decision for your new systems.
>Best,
>Oliver
>
>
>----------------------------------------------------------------
>Oliver Biehlmaier, PhD
>Head of Imaging Core Facility
>Biozentrum, University of Basel
>Klingelbergstrasse 50/70
>4056 Basel
>Switzerland
>
>Tel:        [phone]
>Email:     [hidden email]<mailto:[hidden email]>
>http://www.biozentrum.unibas.ch/imcf
>----------------------------------------------------------------
>
>_________________
>From: Arvydas Matiukas <[hidden email]<mailto:[hidden email]>>
>To:
>[hidden email]<mailto:[hidden email]>=
>20
>Sent: Friday, March 8, 2013 12:24 PM
>Subject: Computer for image analysis
>
>*****
>To join, leave or search the confocal microscopy listserv, go to:
>http://lists.umn.edu/cgi-bin/wa?A0=3Dconfocalmicroscopy
>*****
>
>Dear listers/microscopists,
>
>I assume there is good time to update new trends in
>image analysis hardware. The last discussions on image
>analysis computer were in 2006-8. Though the basic
>principles of CPU, RAM, hard drive, video card, monitor
>selection still hold some new types of hardware became
>popular/available, e.g. SSD drives, APU, water cooling.
>Now a decent gaming computer (~$1k) has the processing power
>of a 2006 expensive workstation (~$20K). I was suprised that
>I was able to completely overhaul my 8 year old ATX case
>to a quad core 2GHz APU, 8GB 1600MHz RAM, 160GB SATA-2
>SSD, water cooling, USB3 and SATA3 Gigabyte motherboard,
>and 4 monitor 1GB video card.
>for under $300 (online, after rebates).
>
>Now I am wiling to upgrade/overhaul my work computer which
>is used to run ImageJ, Fiji, Deconvolution (Autoquant, Huygens),
>Matlab, PV-Vawe, Labview, Origin. Please advice/share you thoughts
>what best configuration is possible to buy for $2-3k (monitor
>excluded).
>My first choice would be  to go with a fast gaming computer, e.g.
>Dell-Alienware Aurora=20
>Windows* 7 Ultimate, 64Bit, English
>2nd Generation Intel* Core* i7-3820 (10M Cache, Overclocked up to 4.1
>GHz)
>16GB (4 X 4GB) Quad Channel DDR3 at 1600MHz
>NVIDIA* GeForce* GTX 660 1.5GB GDDR5
>1TB RAID 0 (2x 500GB SATA 6Gb/s) Solid State Hybrid
>19-in-1 Media Card Reader
>No Monitor
>Integrated 7.1 Channel Audio
>
>The second  choice would be to buy all components online and
>build a computer myself (I have done this about 50 times over
>25 years). This option typically saves money or buys better
>components,
>and provides you full specs of the hardware. The con of this
>approach is that it wastes some of your time to debug/make all
>the hardware work together and with your software. However,
>as the computer is for me not just a box but a tool I am ready
>to make this sacrifice.
>
>BTW, is there any solid preference towards CPU Type (Intel ix/AMD/Intel
>Xeon)
>
>Thanks for your input/advice/thoughts,
>Arvydas
>--------------------
>