HP's Z400 Workstation Runs The Tom's Hardware Gauntlet

The Z400 And Benchmarking Notes

All in all, the Z400 sports a fairly standard X58-based motherboard. The one exception is the memory slot configuration on our first-gen board, which HP has since rectified with a six-slot follow-up, allowing two modules per channel.

While it may seem that this is greatly reduced from your average enthusiast board, remember that these platforms have a different purpose and market segment than those boards. These systems stress reliability and availability as much as performance. There won’t be any detailed conversations about overclocking settings available in the BIOS, because there are no such provisions.

Most workstation motherboards only have a single PCIe x16 slot for graphics. It was only recently that workstation software supported SLI and CrossFire, and thus the systems weren’t equipped for it either. Most server-oriented motherboards have onboard video and lack an x16 slot at all. Instead, they include multiple x8 slots for storage and networking controllers.

The Intel Xeon W3520 processor the machine is equipped with is very similar to Intel's now-EoLed Core i7-920, which shares the same clock speed and most of the same features. There are only two differences to be seen between the two: the Xeon W3520 supports ECC memory and Intel Demand-Based Switching, whereas the Core i7-920 does not. With that said, DBS in the workstation space is what SpeedStep is on the desktop.

Testing Workstations

Testing workstations for Tom’s Hardware is quite a bit different from evaluating the latest desktop CPUs and graphics cards. In approaching the methodology, the benchmarks need to persist for longer than the typical mainstream app, and they need to be intensive enough to demonstrate the performance differences between very similar machines, meaning tests that only take a few seconds to complete are out of the question. Also, the tests must be repeatable and objective. That is, the differences between systems must be measurable in a quantifiable way. “It feels faster” doesn’t cut it.

Over the past several years, SPEC (the Standard Performance Evaluation Corporation) has attempted to create tests that can be used to benchmark systems against each other, including the SPECapc (SPEC application performance characterization) tests. The SPECapc tests are scripted workloads that measure the performance of actual applications. Unfortunately, the SPECapc tests usually only work correctly under the version of the software that they were designed under, and if the test isn’t updated for newer versions of the software, then people trying to evaluate workstations are left trying to size them up using older software. This can lead to disparities between old and new code that go on for five years or more. Also, the SPECapc benchmarks may be enough to weigh down the machine for substantive testing purposes, but those tests, in many cases, do not accurately represent a ‘production workload.’

So, we here at Tom’s Hardware have developed a few workstation tests of our own that we feel are a better approximation of real workloads. In many cases, they should be familiar to people performing production-type tasks. With experience in the digital content creation industry and regular contacts working in the industry, we were able to get a feel for active professionals would like to see in benchmark tests.

Hewlett-Packard supplied the Z400 workstation on extended loan, both as a review unit and as a test case for the development of our new workstation tests. These workloads were developed on the Z400 with the express intent of bringing it to its knees, and we managed to do that after months of work. The suite, as it stands, is not written in stone. And it is our aim to expand them and possibly change some tests to increase their demand--hopefully with some suggestions from you. We tried to cover multiple market segments, but the initial tests are heavy on the DCC (Digital Content Creation) market segment and light on others. This is something we hope to remedy.

Benchmarks And Settings
NewTek LightWave 3D 9.6

Custom workload: High-polygon-count Tom’s Hardware logo

Modeler test: Scripted cloning of surface details across a segment of the logo

Render test: 1920x1080 render of logo with photoreal motion blur, ray-traced shadows, global illumination

OpenGL Test: Generate OpenGL preview of animation for real-time playback on screen

e-on software Vue 8 PLE

Custom workload: Landscape generated in Vue 8 full version and imported into PLE

Autodesk MatchMover 2011

Custom Workload: 720p HD clip tracked in 3D space

Adobe Premiere Pro CS4

Custom Workload: Edit of 59.94 fps 720p DVCProHD footage, with transitions and some color correction, Render To Work Area.

Adobe Media Encoder CS4

Custom Workload: Take above edit and render to H.264 for Blu-Ray

Adobe Photoshop CS4

Custom Workload, Radial Blur, Shape Blur, Median, Polar Coordinates filters

Adobe After Effects CS4

Custom Workload:  SD motion graphics sequence with three picture-in-picture frames sourced from 720p HD

REAPER v.3.63

DAWBench Universal 2010: Test number of simultaneous effects that the system can effectively run

Custom Workload: Render and mix down to .wav custom score project, multiple tracks of audio, VST synthesizers and reverb

Synthetic Benchmarks
Maxon Cinebench 11.5

3D Rendering and OpenGL Benchmarks, built-in benchmarks with default settings

CASE Euler3D

CFD simulation over NACA 445.6 aeroelastic test wing at Mach .5

SiSoft Sandra 2010

Memory Test: Bandwidth Benchmark

This thread is closed for comments
    Your comment
  • bitter
    So JF was right to say that HT should be disabled in some scenarios
  • haplo602
    hmm ... I wonder if you could make a D3D/OGL comparison test in some 3D modeling software ? I don't actualy know if any of them supports both APIs.

    BTW you dual PCIex16 comment is wrong. There were dual PCIex16 slots in HP workstations since xwX400 models (6400,8400,9400).
  • Draven35
    Some of the other models have dual x16 slots... but they are rare in baseline models. At some point we'll look at how well an SLI config works in workstation OpenGL tests.

    3dsMax uses Direct3D (and OpenGL), but prefers Direct3D (it is 'recommended' by Autodesk on my GTX 460-equipped primary machine...), but Maya does not, it requires OpenGL.

    As for whether HT should be disabled? Still questionable. If you're doing animation, and have HT off, you'll gain some interactivity when working, but as soon as you need a test render you'll be losing time. If your 3D application does multithreaded interactiivty tasks, then you're better off with HT on- for instance, the VPR interactive shading mode in Lightwave 10. Its largely dependent on how your software responds to what, and hopefully we'll have more answers on that with the next workstation we look at.
  • ares1214
    Nice cable managment...
  • Draven35
    It came that way! The other models in the z-series have better cable management even, specially with the shrouds.
  • HibyPrime
    Lets see.

    2GB Ram - Not even close to be considered a usable workstation. - Nevermind, I messed up. I didn't see the (x3)

    250GB main/2TB raid 0 - raid 0 setups in workstations are used for scratch disks, and thats it - storage is too risky. 2x 250GB in raid 0 makes much more sense cost wise as you'll never use more than 500GB on a scratch disk, with a 1TB for temporary storage. I say temporary because a workstation is just that, once the work is done (and during) it's backed up and rarely used again.

    Creative Labs X-Fi Titanium PCIe - Any Audio professional needing sound quality is going to be using almost exclusively external hardware. The analog outputs all anyone else is going to need.

    Who ever chose the specs for this machine needs to be shot.
  • MU_Engineer
    This was an interesting review, but the real machine to test would have been the dual-processor Z800. The big advantages of a workstation are better reliability and stability than a standard desktop and the ability to use multiple processors and more RAM than a typical desktop. Perhaps a Z400 vs. Z800 test would have been interesting as there were a mix of poorly-threaded and well-threaded applications in the test.
  • warezme
    This machine is pathetic. Any decently configured OC'ed quadcore i7 will run circles around this thing, ugh.
  • Draven35
    It was agreed to start with a z400... the workstation tests were actually developed on this machine, hence why it has a Gen 1 board. Mayhap we'll look at a z800 later.

    Most video editors use a RAID array for their video files unless they are just editing DV. HD footage shot on a decent quality camera would quickly overwhelm your proposed 500GB array- the raw P2 footage alone for my film project was 350 GB, and that's before the scratch files generated for color correction and transition renders- and the P2 footage is pretty strongly compressed for semi-pro camera footage. (40 Mb/s) If you're editing uncompressed 1080 HD, you're dealing with 124 MB per second of video. With a larger array, a system like this could be set up for editing uncompressed HD with the simple addition of a BlackMagic Decklink in one of its HD flavors, or the HD version of the Avid Mojo if you prefer to work in something like Media Composer. With the transition of TV to HD, more and more projects are being finished for HD in an attempt to future-proof them- which has actually been going on for about five years before the transition. If you want to see something truly horrible, look at the data rates required for 2k or 4k playback...

    HP doesn't sell their machine configured with any kind of professional audio interfaces. They submit their machines for certification by Avid, and the 'full' Pro Tools (Pro Tools HD) comes from Avid on a z800.The audio interface choice for the system was either the Creative card, or the onboard RealTek codec, because that's what HP had and could configure a system with.

    You'll find a surprising number (like, most) audio guys including musicians these days working with a mix of software and hardware, and the mixing is done on a computer, not on external hardware. There are audio interfaces with two, four, six, eight and more inputs... my personal music workstation has eight analog inputs, plus ADAT and S/PDIF in. (Note, I need more like twice that in inputs.) You need all these input to capture multiple instruments on completely independent tracks simultaneously , so that they can be given separate effects and the mix can be adjusted on each track. (That's right, you can capture all those tracks at the same time.) Some of these audio interfaces are an internal card , some are an internal card plus an external breakout box (M-Audio Delta series, MOTU 2408, etc), and most are USB or firewire external (Presonus, TASCAM, Echo, Digidesign, Focusrite, et al) boxes with all the hardware located within it. In addition, many have PCI or PCIe DSP accelerator cards from UAD or TC electronic for effects processing and synthesis.
  • cadder
    The big advantages of a workstation are better reliability and stability than a standard desktop

    I'm not sure I understand this. A machine built with the best enthusiast-class hardware should be basically 100% reliable. The most unreliable part is the hard drive. This machine uses Seagate hard drives, probably the most unreliable brand of drives. This machine also uses an HP motherboard, and HP is below average in reliability in their retail products. Should we expect the workstation components to be better? As for stability, it's using the same OS that everybody else has access to.

    All of our CAD machines here use i5/750 CPU's, aftermarket coolers, 8GB, Gigabyte motherboards, a Velociraptor or SSD for the boot drive, a WD Black for the storage drive, ATI FireGL video cards, Win7 Pro 64bit. The machines run at 3.5GHz 24/7 with no problems. I add 2 cooling fans to each case and that in conjunction with the aftermarket cooler keeps things running real cool.
  • Draven35
    Before the 7200.11 problem, Seagates were widely regarded as the fastest and most reliable drives, and were the drives most often used in RAID arrays for high-end workstation work. (Look at the drives in an old Medea, SGI, Avid, etc RAID for media use- a large percentage of the time, they are Seagate drives.) Western Digitals were largely viewed as 'acceptable' and Maxtors were 'unreliable'. IBM drives were considered to be very reliable until the Deskstar problem. I've had several Maxtor drives fail on me, and a few Western Digitals, but my Seagate drives often have ended up removed and put on a shelf when they got to small for practical use rather than failing.

    I'd be interested in seeing how the I5-750 compares with the Xeon used in this review (as well as the i7 that Chris used in the prior workstation comparison...)
  • JohnA
    They needed a benchmark. They have one. You have to start somewhere, and the price was right.

    I'm most interested in how the video needs have changed over the past few years. I do a lot of CAD/CAM, but the video card doesn't make much difference in most of the stuff I do, as long as it is a work station class.

    As always, throwing in a standard rig with Non-FX video cards for comparison is a must.
  • ecrenshaw
    I just got a Z800 with single Xeon 5620 processor, 8GB ram, and an ATI FirePro 7800 to replace my aging xw4400 that has 4GB ram and a Fire GL7200.

    Autodesk Inventor screams on the new box. Load time is pretty quick...parts/assemblies load a bit quicker unless they have to be pulled of the network first (via Vault Manufacturing).

    Had some year end money to replace the wx440 which is still a very solid machine. Plan to use the 'hardware at home' policy. HP has been making some really solid performers lately.
  • tsnorquist
    For $3298.00... you'd think HP would hit the local Lowes up and get some zip ties & shrink tubing for that birds nest inside.

    Just saying....
  • casey_souder
    Great Article!

    If you are looking to add more scientific programs to the benchmark suite, I would look at LANDIS-II. It is a forest dynamics model developed by the University of Wisconsin and the US Forest Service. It can be downloaded from www.landis-ii.org. If you decide to add it to your suite and need help creating a benchmark test I would be more then happy to help.
  • mc tanza
    I think the DAW bench can't be taken seriously. It's so flawed and limited, it doesn't represent real world usage. your custom test is probably more realistic, altough it tests a different kind of workload.

    Anyway, using asio4all and a Creative card is hilarious. I do understand that's what you had on hand though.

    On a different note, I must agree that using a RAID 0 setup on a workstation class machine is a big no-no for almost any usage. It's kind of suicidal...
  • digitalrazoe
    To use a car analogy and were traveling over a rough road while pulling a trailer... would you take a Corvette or Ferrari ? No you would take something with clearance and built tough. a good ol Truck preferably a 4x4 3/4 ton or jeep that has the muscle to get the job done. Will it dust a sports car of the line in a 1/4 mile ? Nope but that's just it you want reliability when you are crunching many polygons or massive tracks of audio. The last thing you want is a data error that could render your hard work useless. hence things like ECC REG memory or just ECC will make data more reliable. Yes you can build workstation class performance with enthusiast parts. But it will take one good kernel panic for you to wish you still had that old dual PII Kayak of yesteryear. as for cable management; as long as the airflow is good I could care less. Just get my mix, movie creation or new way of air travel designed with out wondering if my overclock is going to cook my goose in the morning.
  • dcc
    I am really happy to see a benchmark for workstations start to appear. For my company, I chose to have our 3D artists on Dell 9000's with i7-920's and NVIDIA 260's. The machines were $1k and have been stable and productive using Maya, Zbrush and Photoshop. I just couldn't see the price/performance ratio on Xeon CPU's and the Quadro cards really don't do anything that a 3D animation program needs - anyone know what it is about a quadro that is unique?

    I would really like to see consumer chips vs. their workstation counterparts, hyperthreading on vs. off, ECC vs. non-ECC, quadro vs. gforce in comparisons of rendering and image processing using industry standard applications - maya, max, mentalray, vray, renderman, Photoshop, After Effects, Nuke. That benchmark would help me and many other studios determine where to invest our hardware dollars.
  • Draven35
    DAWBench is what it is, and works reasonably well, though it definitely isn't any kind of realistic workload, and takes awhile to run the test. The problem with any given DAW test is twofold: 1: something quantifiable and repeatable (mind you, running a decent CPU tach instead of the one built into the software may be good enough for our purposes) and 2: licensing issues with VSTs. I have alot of VSTs that i could have used but the demo versions of many of them do a whole 'random bursts of noise' thing- which on several of them causes a drop in CPU usage that basically invalidates its use in any cpu tach based testing.

    I'm also looking into getting an external (likely USB) audio interface that is a known good performer to use on testing workstations.

    And yes, likely the RAID array would be a RAID 5... but i didnt have the drive bays for it. When i started the tests, i was doing them under WinXP x64 and Vista x64, then added Win7 x64. Hence why, ig you look in the photos, there are labelled system drives- i had three boot drives.
  • Draven35
    DCC, i couldn't find any tests that would show the difference of ECC memory.
  • Cadder - this is a commerical machine, not a retail machine... Show me a retail HP product that offers a 3/3/3 warranty on it... The comparision between a retail and commerical product, the relability, is like the difference between night and day.
  • dEAne
    Yes this is for gaming PC. The specs are good for Workstation.