Motherboards Memory Storage Cases/Cooling/PSUs IT Computing Displays Mobile Mac CPUs & Chipsets Video Digital Cameras Linux Gadgets Systems Trade Shows Guides Home Increase Font Size Decrease Font Size Change Page Size
Supermicro 5046A-XB arrives, Workstation Review Opinions Requested...
Supermicro 5046A-XB arrives, Workstation Review Opinions Requested...
Date: January 21st, 2009
Author: Gary Key
 
 

We received numerous emails and forum messages after our last X58 articles requesting that we take a different look at this platform. One that is not consumer/gaming oriented and instead focuses on the workstation capabilities of Intel's latest platform featuring the i7/X58. With that in mind we have been working diligently on a new test suite oriented towards the workstation crowd. The problem we discovered is that one could end trying to procure and test so many various programs that the review never gets done.

Believe me, that is one bad habit of mine after reviewing my initial rough draft for our first user experience article. After melding a few spreadsheets, pasting together all of the test notes, and looking at the results, it hit me that we had tested 83 different components, 22 games, and 37 different applications, not too mention a dizzying combination of hardware combinations for the memory and overclock results. The outcome is that this article is now under the editor's knife for obvious reasons. Probably the primary reason is to keep the reader awake and focused on the actual motherboard being reviewed, which happens to be the Gigabyte GA-EP45-UD3P before we move on the 790GX/GF9300 products.

So, for those readers who are passionate about viewing workstation results, we would like to hear from you again. Mainly, what are your top three to five programs that you would like to see tested on this platform. While we have procured several video/audio content creation applications along with other business centric programs, we fully realize there is diversity in the workstation market. With that in mind, we want to focus our efforts on providing relevant coverage and results for the top applications where possible. We use the word possible, as procuring a $40K seat license for a particular CAD/CAM package will probably be outside our current scope as one example. Also, we want to tailor the test suite to the hardware received for review.

That said, the wizards over at Super Micro Computer, Inc. sent us their new 5046A-XB bare bones workstation. This kit features the C7X58 motherboard, a high quality 865W power supply, custom designed cooling system, pre-configured Hot-Swappable drive setup, and a Tower case that feels as if it were built out of granite. The base bare bone 5046A-XB MSRP is around $900 and will vary depending upon the options chosen. Based on current test results, this Supermicro solution is a bargain to us.

We are still wrestling with our first 24GB memory kits (not a board problem), but all of our initial tests indicate that Supermicro has done a wonderful job with this platform. We have not needed or required multiple BIOS releases for stable operation, the custom cooling system is very quiet, the case is easy to work with and the internal wiring is impeccable. As an added bonus, the hot-swappable bays are a breeze to use, especially considering the number of times we installed and removed some firmware challenged Seagate drives. We are still working on 24GB memory results and some additional digital content creation tests, but performance has been flawless so far.

To say we are pleased at this point would be a serious understatement. Besides a meticulous design, customer service and technical support has been superb to date. We will be back with a full review, but this product already has our blessing. In the meantime, drop us a note and let us know your opinions on workstation benchmarks.


58 Comments
Username:
Password:
Thank you by turner01, 383 days ago
Thank you for reviewing this equipment. Kudos

I look forward to this review very much. I hope you were able to get some graphics benchmarks for us CAD users. I would like to see some day a comparison of professional vs. gaming cards on CAD programs. I wonder how much difference the professional cards make.

Another point of interest for me on this type of machine is the Raid setup, a Raid card comparison is also on my wish list. I have Areca and Highpoint cards.


Reply
RE: Thank you by kyleb2112, 383 days ago
Yes, a comparison of gaming cards vs professional is long overdue. Graphics professionals are always having to make that call with next to zero direct comparison data--and just a lot of opinions on message boards.

Reply
RE: Thank you by ssj4Gogeta, 383 days ago
I don't know if this is relevant here:

http://www.techarp.com/showarticle.aspx?artno=539&pgno=0


It's a guide to soft mod your GeForce into a Quadro. I've never used a professional gfx card. I came across this guide so I though I'll post the link.

The guide says that the only difference between professional and gaming cards is that the drivers are better optimized in professional ones for professional software. But you noramlly can't install a Quadro driver on a GeForce card. Following this guide apparently lets you do that and get the benefits of Quadro. They've also provided improvement numbers.

Reply
RE: Thank you by MilesT, 383 days ago
I've done the bootstrap mod to convert an 8800GTS 320MB card into the equivalent Quadro, and the difference isn't noticible unless you have a seriously big model on screen in CAD. But with a big model, it sure does make a difference. I would estimate i had over a million verticies on screen for my testing. with the unmodded card, i was averaging 8-9 fps, but with the softmod, it was managing over 20. So yes, they do make a difference. so if you're going to be using cad in a non-professional environment (ie, use in small business or home), go for the equivalent non-quadro card and do the softmod to save yourself a good few hundred dollars.

Reply
GIS benchmarks by cjcoats, 383 days ago
One thing that comes to mind is large GIS
applications. The
GRASS GIS is open source (so the price
is right) and originally came from US Army Corps of Engineers work; see
http://grass.ibiblio.org/

Windows support is quite new, but 64-bit versions are easily available under Linux. For some simpler examples, see
http://www.grassbook.org/;
if you'd like some larger ones (Continental US at 30 meters resolution), contact me...

Also under Linux,
I have meteorology
modeling and air quality modeling
benchmarks in-house.

And it will be interesting to see
how performance is
affected by large-RAM configurations vs
huge-RAM configurations; some met benchmarks we've done show a 20% speedup when going from 8GB to 32GB for
an executable with
a working-set size of only 1.5 GB (i.e., how much does
OS-level I/O caching
help you?).

Reply
RE: GIS benchmarks by Patrese, 380 days ago
I would also love some GIS benchmarks, specially the ones analysing some huge raster files. There's a free software called Spring that does it.
Also, perhaps you could grab a free license from Bentley of ESRI for some MicroStation Descartes, ArcGIS tests... It's kinda a free ad for them, can't see why they wouldn't give it to you. Thanks a lot!

Reply
RE: Thank you by Lehm, 379 days ago
+1 on the request for Gaming vs Professional cards. This would be invaluable. I'm surprised more haven't done it already.

Reply
Number crunching tests by gomakeit, 383 days ago
I'm interested in some number crunching tests. I work on scientific simulations and raw computation power is important to me. The famous folding-at-home would be a easy test. Just measure simulation duration per wall-clock and cpu-clock hour. If possible I'd also like to see some quantum mechanical calculations based on the venerable Gaussian 03 package (relatively cheap) using its test suite.

Reply
RE: Number crunching tests by trainey, 382 days ago
I agree: Gaussian 03 tests would be fantastic

Reply
Virtual Machines by MGSsancho, 383 days ago
I know this would be difficult to test but at work I test everything in VMs. i might have an XP VM machine or so open that's for training other in tech support how to help others (teach noobs how to set up web cames etc), then ill have a VM of oracle on solaris, then an ubuntu vm with a mirror of current web site, another vm to run on said solaris vm. I understand this would be hard but I do think many of up use VMs a lot for testing. for me they do not use too much CPU just lots of ram. oracle requires minimum 512mb. Then I minimize all that an have photoshop open with dreamweaver in the mix. so for me Its just lots of apps that have varying degrees of useage. sometimes its opening a 4gb tarball, with some VMs in the background testing software while im using photoshop. Then someone comes in and wants a DVD we made into a format to be put on the site so that needs encoding. again it would be awsome if you could find a test for people who just runs lots of stuff that eats up ram.

Anyways looking forward to a review. you already answered 1 question Gary, RAS. nothing like opening my box and replacing an optical drive or disc drive in seconds

Reply
Program Suggestions Three by fri2219, 383 days ago
1) BLAST
2) Maya
3) MATLAB Black-Scholes pricing implementation



Reply
Wow, that workstation looks perfect by Calin, 383 days ago
As for software for testing workstations... I'd say some kind of CAD/CAM package, some compilation tests, some image generating programs (POVRay), graphic work (Photoshop, filters), and probably some video-related tests. Maybe some syntetic performance tests running on the main system and as a virtual machine.

Reply
Photography test suite by BiscuitMonster, 383 days ago
It would be great if you could test some photo-related apps as well. Things like exporting large numbers of raw files from Lightroom 2 into high-quality JPG, or stitching panoramas / merging into HDR from multiple high-res files in Photoshop CS4.

Reply
not sure by ElAngelo, 383 days ago
Not sure if this fits in a workstation benchmark... but I would like to a minimal testing of virtual machine capability...
Right now i'm looking to set up a lab with some virtualized machine and i really no longer know what to choose... i7? phenomII? quad core 2? I'm tempted to go for phenomII as i think it will have better performance in virtualization then the quad core 2 while still being considerably cheaper than an i7 (which would prolly shine in virt tests as well)
but maybe this is something for it.anandtech.com

Reply
RE: not sure by ElAngelo, 383 days ago
maybe just a vmmark? it's seems very hard to find any vmmarks on the i7 while imho it should shine in that benchmark?

Reply
3d software! Compositing software! by Draven31, 383 days ago
Don't bother with POVRay. Get real apps. Either use eval licenses, or talk to the vendors. Use scenes that come on the software CD, or use scenes you create or have created for you, but have those scenes available on the site. CineBench, and the usual suspects of SPEC ViewPerf benchmarks...

Reply
RE: 3d software! Compositing software! by Amiga500, 383 days ago
+1 on that

Contact the vendors (ANSYS, MSc or Dassault Simulia to name a few). Let them know of your situation, and why you want to benchmark - improve the effectiveness of their software for their customers through using more appropriate hardware.

They may give you academic licenses as your work is completely non-proprietary.

It also is some free advertising (of sorts) for their wares on a website with a large and technically astute audience.

Reply
RE: 3d software! Compositing software! by Amiga500, 383 days ago
Oh, and software areas I would like to see evaluated:

FEA (i.e. one of MSc Nastran/ANSYS Mechanical/DS Simulia Abaqus)
[tests mainly FPU performance and I/O performance)

CFD (ANSYS CFX/ANSYS Fluent/Star-CD)
[tests mainly FPU and memory bandwidth]



Reply
CUDA by nnunn, 383 days ago
Could you try a pair of GTX 295 boards in those 2 (x16) PCI-Express 2.0 slots? How does CUDA handle the 4 devices on the two slots. The interesting comparison would be these two games cards vs. a Tesla S1070.

Reply
Supermicro has always been a premium-grade "whitebox" company by James5mith, 383 days ago
I've been a fanboy of Supermicro since I used my first 2U rackmount server from them about 12 years ago. Back then they were one of the only options for something that compact that wasn't from a major brandname that sold preconfigured solutions.

Excellent build quality, attention to detail, and the ability to continue to provide excellent stable solutions that appear to market shortly after the technology is released is the reason I keep coming back.

To this day, they are one of only a handful of companies that offer hotswap backplanes that incorporate SAS expander chips. Running a single cable to a controller vs 24 can considerably ease airflow issues. (Check out an SC846E1-R900B)

Better yet, want a robust blade enclosure and solution? They've got you covered. How many non-Dell, HP, IBM, etc. companies can say that?

As I said in the beginning, I'm a fanboy. But I'm a fanboy because Supermicro is a company that continues to impress. There is no other company out there that matches their ability to offer whitebox solutions that rival the custom, and often proprietary ones that are offered by the brand names.

Reply
Benchmarks by brausekopf, 383 days ago
I would like to see numbers for the STREAM benchmark, especially the scaling from 1 to 4 cores w/o HT and from 1 to 8 cores with HT. That's where the integrated memory controller should have a big impact.

DGEMM (doouble precision matrix multiplication) numbers using Intel's latest MKL library, sequential and parallel might also help many people working in science in their procurement decisions.

Reply
How about Virtualization by Ratman6161, 383 days ago
I'm not sure if this is what you had in mind, but I'd really like to see Virtualization tested. A lot of us IT types who are living in a VMWare world are using VMWare Workstation for a lot of different purposes. Some examples: When I create a new server I often build and test it on my workstation first then migrate it to ESX server. We also use VMWare Workstation for isolated test environments where we need to run several virtual servers at the same time. I also have virtual machine copies of our several different end user workstations available on my workstation so that I can run an environment that looks like what the end user has (I'm on 64 bit Vista while my end users are still on XP 32 bit). The list goes on and on but for one reason or another I've pretty much always got one or more virtual machines running while I'm also doing my other work.

Reply
Compilation by BrightCandle, 383 days ago
Some compilation tests would be nice, Java and C++ are probably the most widely wanted (although C# is another possibility here which is likely more often used than C++ by professionals).

For C++ most reviews do a compile of the Linux kernel.

For Java the compilation of a moderate sized open source package like Maven or Geronimo from Apache.org would likely serve well.

In particular I am would be interested in the variations of CPUs and especially additional cores and differing hard drives.

Reply
RE: Compilation by brausekopf, 383 days ago
Nowadays these compilation tests are more of a harddrive test, aren't they?

Reply
RE: Compilation by RagingDragon, 382 days ago
Nitpick: the Linux kernel is C, not C++.

Compiler tests would be of interest to me too, especially the Sun Java compiler.

Reply
Having built a number of systems for work on a chassis related to that one... by Demon-Xanth, 383 days ago
We built about a dozen systems using the CSE-733T chassis and I found them quite easy to work with. Doing HALT testing the systems ran well upto 80C, at 90C things like the face plate started melting, but if your room is that hot you got other issues. Placing things inside was easy, the drive bays worked well, and everything fit together nicely. This was with the X6DAL-XG motherboard and dual Xeons. We did end up going into production with a Chenbro chassis however because they fit our application better.

Reply
Test suggestions by Jorgisven, 383 days ago
I think a pervasively meaningful test might be to time how long it takes to open Word 2007. I know it's not computer intensive, and might require honing to thousandths of seconds, but it's probably one of the most frequently used programs in any given work place.

Reply
ESX Review by MilesT, 383 days ago
I'd like to see it loaded with ESX and put through its paces with some test VMs running various corporate software packages..

Reply
RE: ESX Review by MilesT, 383 days ago
Sorry, one more addition:

It looks like there's PCI-X PCB mounts on that motherboard.. Is the motherboard available with PCI-X slots for older server cards or RILO/ILO?

Reply
RE: ESX Review and PCI-X Slots by GhostWriter, 383 days ago
Yes, there is another version of this board, the X8SAX, that populates the two PCI-X slots.

The primary difference between X8SAX and C7X58 is that C7X58 has the nVidia SLI technology supported, and the PCI-X slots de-populated to reduce costs.

nVidia charges a license fee for SLI support on X58, and many vendors assume the customer usage of SLI will not be 100%, so they don't want to pass on the extra cost to customers.

Reply
Workstation programs by EE1, 383 days ago
SPICE (circuit simulation)
PCB (printed circuit board) autorouting


Reply
RE: Workstation programs by danger22, 382 days ago
i second the spice suggestion - also the software is free

Reply
Software Builds by kolotyluk, 383 days ago
Yes, I agree some compilation and build benchmarks would be good, if possible running under some IDEs like Eclipse, NetBeans, Visual Studio.

Reply
Workstation I/O Benches by spookware, 383 days ago
I for one would like to see some heavy duty I/O benches. Something like filesystem activity + heavy network activity + heavy CPU activity.

Something like copying loads of files over SMB, while untaring a gigantic archive, while copying a 8GB ISO, while compiling the linux kernel, while verifying the digital sig of a truly huge file.

That or test for maximum latency under load. What really differentiates a workstation load from a server load is that I want to keep using it while it is under load massive load. So system responsiveness while under a taxing load is very important as a workstation user.

you could measure such things with Dtrace on solaris, latencytop on linux, or a mix of tools on Windows (kernrate comes to mind). Or on windows measuring the DPC queue lengths and maximum service time under massive load would be a decent proxy for true latency numbers.

Reply
FPGA Compiles by k3y0n4, 383 days ago
Large Altera's Quartus or Xilinx's ISE design compiles would be nice.

Reply
Another vote for VMs and some hardware thoughts... by TechDicky, 383 days ago
I am dying to get some good info on Virtualization. I would really like to know if money is better spent on a Core i7, Dual Xeons, Phenom II, Dual Opterons... Also performance differences with various hardware RAID configurations.

Reply
RE: Another vote for VMs and some hardware thoughts... by adicro, 382 days ago
I agree, virtualization benchmarks would be great.

Reply
supermicro case by fredsky, 383 days ago
just to say a word about workstation case :
the supermicro is the best for the buck you can get.

especially the "old" sc742 with 7 hdd hot swap. you can also add supermicro 5in1 enclosure to this.

you can use almost any motherboard with this. works great with promise/areca card.

the newest chassis version also include a SAS expander to use with areca 1680 series or adaptec 3xxx/5xxx.



Reply
Very good idea by Michae, 383 days ago
It is the best idea i read online since month: the computation power for scientific computation is the need, and therefore i appreciate the plan you made... Scientific calculation must be included like Spec-CPU for multi processor purposes simulation with FEM! Thanks in advance!

Reply
Blender, Blender , Blender Benchmarks ! by greylica, 383 days ago
I want to see Blender Benchmarks !


Reply
Benchmarks I'd like by Randybob, 383 days ago
Maya
Lightwave
Photoshop

64-bit versions, of course.

Reply
relevant computational kernels & SPEC by rmlarsen, 383 days ago
I would like to see benchmarks include relevant computational kernels. As a developer of HPC code these tell me more than the performance of pre-existing applications (not that these shouldn't be included also).

Off the top of my head, I'd like to see numbers for

1. Fast Fourier transforms (1d, 2d, and 3d) using the latest version of Intel's MKL library (and possibly FFTW).
2. DGEMM and SGEMM (matrix-marix multiply in double and single precision).
3. STREAM benchmark.

Benchmarks should include a range of array sizes to reveal the effects of the memory-hierarchy. I would also like to see speed-up curves.

SPEC CPU scores (speed and throughput) usually correlate well with the performance I end up seeing in my own code. If you choose not to run SPEC yourself, I would appreciate if you could simply quote the numbers submitted to SPEC for the CPUs you are reviewing.

Reply
Thanks by adicro, 383 days ago
Thanks for the article!

How about some compiler tests? Some GCC and Visual C++/Visual C# benchmarks would be great, as well as light database testing.

Reply
RE: Thanks by icrf, 381 days ago
Agreed on some testing for developer tools. It seems these days, I don't spend as much time compiling code, as I am waiting for Visual Studio Team System to do something else, like open or validate a large database project. I suspect parsing/compiling in general is a close enough analog.

Reply
3D CAD Software, please. With professional vs. desktop video comparing by Hudly, 383 days ago
This is in desperate need.

I suggest 3 initial programs:

1. Catia
2. Realflow
3. Solidworks

Catia gives you a high-end idea of structural 3D

Realflow gives you a high-end idea of 3D simulation and representation (including physics calculations)

Solidworks gives you a medium-end perspective on structural 3D

And I agree with others, go through the vendors for licenses and support. They will be much more willing to help you get a hold of the software.

Reply
scientific apps - floating point performance in real apps by danger22, 382 days ago
matlab is expensive why not use scilab or octave for your benchmarks? this would be great for those of us in the scientific community.

also some fem (finite element method) apps would be a great benchmark - want to know how good cpus are at matrix algebra? FEM will give an idea of real world performance. simply download the student version of maxwell 2d or use FEMM - i would be happy to supply some example files for testing

fem apps need lots of memory, memory bandwidth and good floating point performance so what could be a better benchmark for a workstation?

Reply
RE: scientific apps - floating point performance in real apps by danger22, 382 days ago
scilab and octave are free by the way

Reply
RE: scientific apps - floating point performance in real apps by danger22, 382 days ago
im glad to see so many people are interested in FEM/FEA sims from in particular from ANSYS

Reply
VMWare / Hyper-V by VooDooAddict, 382 days ago
As other's have said, I will echo. We need some form of VMWare capacity tests. There's going to be may people out there loading the i7 up with VMWare Workstation, VMWare Server(not ESX), and Microsoft's Hyper-V for Dev/Test. I don't really have enough free time to discuss testing methods.

Reply
Microsoft Visual Studio 2008 by Maverick, 382 days ago
The new version of Visual Studio 2008 includes the ability to do parallel builds that take advantage of multi-core systems.

I'd love to some benchmarks around it. Send me an email/PM and I can provide some guidelines on how to benchmark this app.

Reply
Comments Page 1 of 2





AnandTech.com Blog Categories
All categories
Anand's Macdates
Anand's Theater Construction
Anand's Updates
Cases and Power Supplies
CeBIT 2008
CES 2008
Computex 2009
Derek Decanted
Eddie's Got Game
Gary's First Looks
IT Computing general
Jarred's Musings
Kris's Corner
Raja's Ramblings
Rob's Experiences...
Ryan's Ramblings
Virtualization
What's New with Wes
Blank
Blank

Blank

Latest news by
DailyTech

 February 9, 2010

Blank
Blank
Blank
Blank
Blank
Blank
Blank
Blank
Blank
Blank
Blank
Blank
Blank
Blank
Blank
Blank

 February 8, 2010

Blank


more Blogs Discussions



pipeboost
Copyright © 1997-2010 AnandTech, Inc. All rights reserved. Terms, Conditions and Privacy Information.
Click Here for Advertising Information