Review: HP Z800

With hyper-threading technology enabled on the HP Z800, you get 16 cores on a dual-processor, quad-core Intel Nehalem system. Rhozet Carbon Coder got them all working, too.

Since the launch of the Core 2 Duo line of processors in mid-2006, new workstations have been more about evolution than revolution, with solid incremental but uninspiring performance gains. This is no longer. Sporting a completely redesigned case and Intel''s new Nehalem processor, the HP Z800 knocks the socks off HP''s existing workstation line—especially for video editors and streaming producers.

Is “knocks the socks off” a bit too vague for you? Try this: In a real-world trial, I compressed a 90-minute HD-recorded ballet performance to DVD format. The previous performance champ in my office—a 3.3GHz dual-processor, quad-core system (64-bit Windows Vista, 16GB of RAM)—finished in two hours and two minutes. The Z800, equipped with two 3.2GHz quad-core Nehalem processors (64-bit Vista, 18GB of RAM), finished in 59 minutes and 30 seconds—more than 50 percent faster.

Before you jump over to hp.com to place your order, you should know that your mileage will definitely vary, not only by application but even by project type within the same application. To understand why, we need to take a quick look at Nehalem.

Chances are that you don''t lie awake at night thinking about how to make faster CPUs, but if you did, you would probably figure out that there are two main approaches: You can make the CPU itself faster, which is a rising tide that lifts all boats in the harbor. Or you can improve the data flow to the processor, which primarily speeds applications that were previously bottlenecked because the CPU sat idle while waiting for data to arrive for processing.

Related Links

href="http://blog.digitalcontentproducer.com/briefingroom/2009/03/31/hp-reinvents-the-workstation/" target="_blank">HP Reinvents the Workstation Rosie

Tackles the New HP Z800

Edit

Expertise: Test Drive: HP xw8400

For example, a 100K data set in a 3D design program such as Autodesk 3ds Max can necessitate hours of rendering time. However, the problem there isn''t getting the data to the CPU, so enhancing the bandwidth to and from the CPU will produce only a marginal benefit. On the other hand, in a video-editing application that throws gigabytes per second at the CPU, improving data throughput can deliver profound speed improvements.

On what did Intel focus with Nehalem? Intel primarily focused on the latter category, for which the new chip delivers several new innovations to the Intel processor line, most notably an integrated memory controller (IMC) and QuickPath Interconnect (QPI). If you follow CPUs closely, you''ll remember that AMD debuted an IMC several generations ago. Intel has improved on AMD''s 2-channel design with a triple-channel memory controller that should deliver significantly higher bandwidth to and from the CPU than AMD does. The other major bandwidth-related innovation, QPI, is a much faster replacement for the front-side bus that transmits data directly from CPU to CPU and from CPU to the chipset that integrates the CPU with other system components.

In regards to making the CPU itself faster, Intel did add one familiar technology back to Nehalem: Hyper-threading Technology (HTT), which adds components of a second processor to each Nehalem core. When you open the Performance tab of Windows Task Manager, you''ll see 16 CPUs running, which is an impressive sight.

While more sounds better, HTT helps only when applications are efficiently multithreaded. When it doesn''t help, it usually hurts. For example, in the ballet benchmark, the Z800 finished in 59 minutes and 30 seconds with HT enabled. Without HTT, performance slowed to 71 minutes. That''s 19 percent faster with HTT enabled. On the other hand, the Z800 ran through my 48-file encoding benchmark with Rhozet Carbon Coder in 36 minutes and 22 seconds with HTT enabled and in 26 minutes and 53 seconds with HTT disabled. That''s 26 percent slower with HTT enabled.

Why the difference? Because it''s harder for the operating system to manage 16 cores than eight cores. And when processor utilization is poor, the extra overhead degrades performance. Though Rhozet is a fantastically multithreaded encoder, apparently the VC-1 codec isn''t (at least when it''s working within Carbon Coder), and simultaneously encoding 16 files to VC-1 produced an overall CPU utilization of around 18 percent. This jumped to 50 percent with HTT disabled, reducing encoding time by 33 percent.

Another enhancement to the CPU''s pure processing capabilities is Turbo Boost, which over clocks the processor when it''s operating below power, current, and temperature specifications—which can increase performance by about 10 percent. All Nehalem-based systems will benefit from Intel''s Turbo Boost feature, while HP added an additional level of turbo-boost to their own line of Nehalem-based workstations.

HP Z800''s toolless interior.

I''ve been fortunate to review many HP workstations, dating back to the last century. As I scan through the online reviews, I see that the xw4100—which I reviewed in August 2003—looks pretty much the same as the xw8600 that I reviewed last year. That makes five years without a physical makeover.

That long dry spell is over. The Z800 features brushed-aluminum side panels; integrated handles in the front and back; and a cableless, toolless interior with innovations such as a modular power supply that you can replace by pulling it out with a handle. Hard drives work the same way; you click a green tab, a handle drops, and you pull it out. I asked my third grader to try it, and she removed and replaced a drive with no problem (and then demanded $5, but that''s a different story). Go to digitalcontentproducer.com/dcptv/rosie_hp_z800 to see this video demonstration.

Two easily removable panels cover the expansion slots and memory compartments. Another panel, complete with fans, covers the memory. Insert your thumbs and forefingers in the prominently marked slots, and that comes out too—again, with no tools. As before, diagrams on the case detail the memory load order and identify the major components within the case.

The 12 dual in-line memory module (DIMM) slots can hold up to 192GB of memory, though the necessary 16GB DIMMs won''t be available until late 2009. The motherboard has integrated SATA and SAS controllers, and the system can house up to five internal drives, with a 7.5TB capacity. I didn''t see any noise-level ratings, but the workstation is the quietest that I''ve ever worked with, which makes it great for recording narration and similar activities. The unit also comes with a limited three-year, next-business-day warranty.

So the Z800 is big, bold, and highly functional, but how does it compare to what you''re using today? Let''s have a look.

Table 1. Autodesk 3ds Max benchmarks.

I tested the Z800 against three other computers: a dual-processor (DP), quad-core (QC) 2.83GHz HP xw6600 workstation running 32-bit Windows XP with 3GB of RAM and a Nvidia Quadro FX1700 graphics card with 512MB of memory; a DP, QC 3.33GHz HP xw8600 workstation running 64-bit Windows Vista with 16GB of RAM and a Nvidia Quadro CX with 1.5GB of memory; and a new single-processor, QC Z400 workstation running 64-bit Windows XP with a 3.2GHz Xeon W3570 processor, 6GB of memory, and a Nvidia Quadro FX4800 with 1.5GB of memory. To repeat, the Z800 came with two QC Nehalem processors running at 3.2GHz, with 18GB of memory running 64-bit Vista and the same Quadro FX4800 with 1.5GB as the Z400.

I started my tests with Autodesk 3ds Max 2009, hoping to run the Standard Performance Evaluation Corporation (SPEC) benchmark, but it didn''t appear to be updated for the latest version of 3ds Max. And the benchmark crashed on each workstation. So I went freelance and loaded some individual projects from the SPEC benchmark and started previewing and rendering. Table 1 shows the results.

In general, the harder the task, the more the Z800 excelled—especially against the single-processor Z400. The Z800 didn''t impress in the real­time preview or in the first single-frame rendering trial, but it more than held its own otherwise—with an impressive 32 percent edge over the xw8600 in the longest-running test.

To be honest, I''m not a 3ds Max user, and I wasn''t really counting on the SPEC benchmark to present a useful result. Still, even in the limited tests that I performed, the Z800 proved to be a very worthy upgrade from the xw8600 workstation that it replaces.

Table 2: Adobe Premiere Pro benchmarks.

I am a heavy Adobe Creative Suite 4 (CS4)
user, and I was happy to have a couple of real-world projects to throw at the Z800. First, there was the aforementioned ballet performance, which was a two-camera HDV shoot that I mixed using Premiere Pro''s multicam feature and delivered on SD DVD. I also shot a single-camera show in progressive widescreen DV for a local group auditioning for America''s Got Talent. The edited clips were less than 10 minutes long, which I rendered to H.264 for uploading to the show and to YouTube. Table 2 contains those results.

As mentioned earlier, the Z800 really excelled in the ballet test. Though the 31 percent performance boost over the xw8600 in the DV concert clip is also nothing to sneeze at. Why was there so much less of a performance disparity in the DV test than in the first test? It probably relates to the fact that SD has less than a quarter of the pixels of an HDV stream, making throughput to the processor—Nehalem''s greatest strength—much less of an issue. H.264 also encodes more slowly than MPEG-2, which again favors the older-style Xeons, because getting data to the CPU is much less important.

The Red Digital Cinema Red One-related tests are not real-world tests (I wish); rather, they''re Red clips I downloaded from the Web and produced into a short 90-second project with two picture-in-picture effects and one greenscreen effect. The first was provided via Adobe After Effects through Dynamic Link; I applied the second using Premiere Pro''s native chroma-key filter.

When using After Effects and Dynamic Link, the test is much more computationally intense and consumes more memory, forcing the 32-bit xw6600 computer to store data to and from the hard disk (called “paging”), which destroyed its performance. The native greenscreen test used the same files and was the same duration, but this test was relatively simpler to compute because Premiere Pro''s chroma-key filter is not as complex (nor as capable) as the excellent After Effects Keylight filter. Obviously, the xw6600''s time comes back to earth in the second test because After Effects and Dynamic Link are no longer required, reducing the overall memory requirements to less than 2GB.

It''s impossible to tell exactly what''s going on in any particular test, but one interpretation of the Red-related results is that as projects decrease in complexity, data throughput becomes more important—which would explain why the Z800''s advantage over the xw8600 and especially the Z400 becomes more pronounced in the second Red test. When combined with the results of the first Red test, I take this to indicate that if you''re a video editor mushing through long, relatively lightly edited productions—such as event, training, and other videos—the Z800 should deliver very impressive performance gains. On the other hand, if you''re producing short, heavily edited 30-second to 60-second clips that involve lots of effects and processing, the decrease in rendering time will be sizeable, but not nearly as significant.

Table 3. Rhozet Carbon Coder benchmarks.

I tested with Rhozet''s Carbon Coder because it''s the most efficient multithreaded program I''ve ever worked with. That generally translates to excellent performance with multiple-core processors. After a brief hiccup at the start of the testing, the Z800 didn''t disappoint.

Specifically, the first test involved 16 1-minute files encoded to VP6, H.264, and VC-1 formats. As you can see, though the Z800 was comfortably the best performer, the advantage wasn''t particularly earth-shaking. Note that, as with all tests, I tried the Z800 and Z400 with HTT enabled and disabled and used the fastest score. As mentioned, in the first test, the Z800 was about 9:30 (min:sec) slower with HTT enabled.

During this first test, I noticed that all computers got bogged down while encoding into VC-1 format, meaning slow encoding and particularly inefficient CPU utilization. So I encoded all 16 files to a single H.264 preset in the second test and to a single VP6 preset in the third. As you can see, in these latter tests, the Z800 really started to shine. So if you''re producing exclusively in these two formats, you could really see significant benefits from moving over to the Nehalem-based Z800.

What''s it going to cost you? As tested, the Z800 costs $10,787, while the Z400 would cost $5,873. I haven''t done the comparisons, but HP tells me that these prices reflect only a small premium over prices for systems based on the older-style Xeon technology. Any way you look at it, with the Z800, you''ll be getting a whole lot more for your money than with any other computer that I''ve reviewed to date.

The other point is that if you''re a content producer who''s still working in 32-bit Windows, you really are making it hard on yourself—particularly with CS4. If memory limitations on your current system prevent you from upgrading to 64-bit Windows, you should strongly consider the Z800.

Company: HP
www.hp.com/workstations
Product: Z800
Assets: Intel Nehalem; easily accessible toolless chassis.
Caveats: Performance benefits from the new Nehalem CPU are application-dependent, but they should be considerable for most video editing and encoding tasks.
Price: Starting at $1,999