The workstation market has always been a consistent seller. The dream of offloading to an on or off-site VM and a cluster for work processing still lies more in the realm of mass production over local quick-to-compute work, and because of that low latency for quick interaction, workstations are here to stay. While the consumer is going smaller, business aims to scale and on the back of our P300 review, Lenovo sent their 34kg dual Xeon E5 v3 behemoth version of the ThinkStation P900 for review.

When designing a workstation, the teams have to think of the two main people that will interact with it. On one side is the user who will want quick response times, a specific feature set and the ability to complete intense compute tasks quickly. A company such as Lenovo can enable this through motherboard design, component pricing and adapting the preinstalled software. This includes getting official ISV certification for many of the professional level applications that someone might want to use a workstation for.

The other side of the equation is that the system has to be suitable for the person who is repairing or upgrading it over the lifetime of the unit. The ability to easily remove, diagnose or replace components on the hardware side helps avoid downtime and improves overall work efficiency. So far in our foray of reviewing workstations some of them have been built to be proprietary and difficult, whereas others adhere to standard base unit design principles.

This is where Lenovo's new ThinkStation design is planned to attack. We reported on the official announcement at the end of last year, with the P500, P700 and P900 range getting the full overhaul in terms of ease of use along with industrial design.

At the time, Lenovo promised air baffles for directional flow and a Flex system to allow hybridization of specific functionality depending on the customer need. Ryan was lucky enough to see a demo unit in the flesh around the same time as the announcement, and I received numerous emails and tweets about the design. Well I am glad to say that Lenovo has now sent us one to review. We were sent the P900 no less, featuring the top of the line in design and weighing in at almost $17000. The only way this could get more expensive would be through denser memory or more PCIe coprocessors. The P900 can hold three NVIDIA K6000s (or Xeon Phi based derivatives) along with a single slot PCIe card.

The full specification of our sample gives two Intel Xeon E5-2687W v3 processors - these are the high TDP 160W models with 12 cores at 3.1 GHz (3.5 GHz Turbo) each that we reviewed at the end of last year that retail at almost $2140 apiece. These are paired with the NVIDIA Quadro K5200, an 8GB Kepler based GK108GL with 2304 CUDA cores running at 650 MHz. For the DRAM, our sample came with eight modules of 8GB DDR4-2133 ECC from SK Hynix.

Lenovo ThinkStation P900
2 x Intel Xeon E5-2687W v3 (160W)10 Cores, 20 Threads, 3.1 GHz (3.5 GHz Turbo)
Lenovo Custom
Aluminium Heatsink with Copper Heatpipes and Fan
Power Supply
Delta 1300W 80 PLUS Platinum
SK Hynix 128GB (16x8) DDR4-2133 ECC RDIMM
Memory Settings
2133 15-15-15
Video Cards
NVIDIA Quadro K5200 8GB2304 CUDA Cores, 650 MHz core, 256-bit
Hard Drive
Samsung XP941 512GB PCIe 2.0 x4 M.2 SSD
Optical Drive
Realtek ALC662
Intel I218-LMIntel I210
Lenovo ThinkStation 2P
Operating System
Windows 7 64-bit SP1 with Downgrade Rights from Windows 8
Front Panel
Power ButtonSD Card Reader4 x USB 3.0
Rear Panel
Audio JacksPS/2 MousePS/2 KeyboardCOM Port2 x Intel NIC (I218-LM and I210)4 x USB 2.04 x USB 3.0DVI-I (Quadro)DVI-D (Quadro)2 x DisplayPort (Quadro)
3 Years On-Site
Starts at $1830 with 1x E5-2609 v3, Quadro K620Review system at $16,797.60
The P900 sits atop the ThinkStation range with dual Xeon E5-2600 v3 processors, starting (from the website) at $1830 with a single E5-2609 v3 (six cores, no HT, 1.9 GHz) with one module of DDR4-2133 memory and an NVIDIA Quadro K620 (Maxwell, 2GB, 384 CUDA cores at 1000 MHz) discrete card. This is in comparison to the P700 which focuses on a compact 2P Xeon build and the P500 which is a 1P Xeon system but still based on the LGA2011-3 socket.

Clearly the first port of call for criticism off the bat is going to be the list price, and I agree, for the specifications Lenovo's pricing seems way off, similar to that of the P300 we reviewed last year. Lenovo's answer stems from the fact that they get almost zero sales from the website, which acts as a business-to-consumer frontage, but the ThinkStation range is more often sold as a business-to-business unit either directly to the vertical industries (financial, medical, defense, oil and gas) which often get volume discounts or don't care about pricing, or as an OEM to resellers who buy closer to list and can add their own components. Lenovo's attitude in this final case is that they don't want to compete with their resellers, who may ultimately be able to offer a better deal/support package to individual consumers buying single units.

Despite the individual pricing, as a unit and by extension the company, the P900 and Lenovo are lucky. This is the first true high end halo workstation product we've reviewed at AnandTech for a long while, and Lenovo has pulled out the bells and whistles on the high end designed to dazzle and amaze. We are in talks to get similar systems from HP and Dell, however my lack of experience with their designs means that for now, Lenovo has my undivided attention.

We asked Lenovo for the Chipset diagram, so users can see how each of the CPUs are routed with regards to the PCIe lanes:

Given the level of functionality that Lenovo is going for with the P900, a lot of the pathways here make sense in order to maximize bandwidth and location. Both of the Flex connectors are PCIe 3.0 x8, allowing for the FLEX modules to use whatever hardware they need. The front USB 3.0 ports are provided by the VIA VL812 hub, with one of the ports acting as an always-on charger port. Perhaps the only downside to the specifications here is the Realtek ALC662 audio codec, which is best described as a poor laptop audio solution. Most motherboard and system manufacturers at this point use an ALC892 as the bare minimum, or an ALC898/ALC1150 which sit at the top of the stack for less than a few cents over the cost of the ALC662. Now ultimately users can add their own audio cards, and the P900 isn’t particularly designed for audio, but the laptop-class codec shouldn’t be here.

The base system is just over two-feet long at 24.4 inches, while being 17.5-in high and 7.87-in wide. Moving outside ecosystems where 1.5 liter chassis exist, the 55 liter workstation is a behemoth by comparison. 7.87 inches high converts to 4.5U for racks, perhaps indicating at what level Lenovo is positioning the P900. Carrying our workstation sample, even with only an M.2 SSD and one GPU was still an effort, with the chassis being made from a combination of steel for strength and aluminum for looks.

Opening the side panel via the latch shows of the internals. Similar to when we saw the initial renders of the P900, we get the three level split airflow in order to keep the system cool. For the top and bottom, air comes in the front via the fans in the middle of the case, over the storage drives, and out the rear past the PCIe coprocessors. In the middle, the CPUs use a rear case fan to draw air in, split the airflow independently across the CPUs and DRAM, and then out the front of the case around the sides of the ODD bays.

The air baffle for the CPUs is easy to take out via the handle provided.

On the side of the handle, here shows the split for the top and bottom, with the air from the top going to the furthest CPU from the rear of the case.

It is a little difficult to characterize in pictures, but the principle is sound.

For the central airflow, where the air from the CPUs moves around the ODD bays, we have the required space:

With the baffle removed the custom motherboard shows both CPUs in a linear design with four banks of DDR4 either side of each socket. Normally we see CPUs staggered to aid in design of the motherboard, but this application suits the Lenovo airflow management.

Each CPU is served by a six-phase server-grade power design to ensure efficiency and longevity. These are Cooper-Bussmann chokes, similar to those used on some of the consumer grade motherboards tested which focus on server-level longevity. As there is no overclocking on this platform, there is little need to over-engineer this aspect of the design. There’s a hint of cable management coming through between the sockets due to extra cabling required at the bottom of the motherboard. Lenovo has this fixed down as to not be effected by other components – the CPU fans are also equipped with fan guards so if the cable does come loose, it doesn’t start grating against the fan blades.

Installed in this system is the Quadro K5200 mentioned in the introduction. This is a standard NVIDIA blower design with no backplate and requiring double slot width. This card is using the lower down PCIe slot, but there is space just below the DRAM slots for another. If this was installed in that position, the position of the PCIe cables would affect the FLEX module installed on the right.

The FLEX module sits in a PCIe x8 slot alongside the main PCIe x16 slots, and is designed to be easily removable with the latch next to it. This card is in the right way round, although the server admin will have to remove it to see the bounty underneath:

It splits into two PCIe 3.0 x4 M.2 slots, although Lenovo can provide a range of FLEX modules to suit the customer requirements. To be honest, this is a nice way of putting in storage, especially for OS and fixed drives, into a high end workstation like system.

The drive supplied in our system was the 512GB model of the XP941 from Samsung, rebranded slightly with Lenovo product codes in case of replacement.

Where the FLEX module sits in the motherboard actually covers some of the cabling underneath, but putting the module back into place was easy due to the chassis design.

Next to this is one of the system fans. Rather than be placed right in the front of the chassis blowing air from the front to rear, these sit in the middle between the storage drives and the PCIe coprocessors and can be removed very easily by pulling the handle at the red stripe.

The fans are guarded on the open side and use Lenovo’s interesting fan connector in order to slot in. No screws are required. The fan used here is an AVC DBTA0938B2U, a 92x92x38mm fan which can use a fair amount of power at full load. Even with the side panel off I never heard the fan move above a low hum, even when hammering the system. That being said, this fan has to be able to cope with a fully laden system in a hot climate which might initiate a higher speed profile.

Similarly, the top strip has an identical fan as well as space for another FLEX module.

At the top we find the 1300W power supply, removable in one action by lifting the lever with the red strip.

Before we get to the power supply itself, this is how it connects directly into the system. We have a direct line in on the motherboard which looks like (but isn’t) a PCIe slot as well as a series of headers screwed into the case for external connectors. Through these two inputs/outputs, the system power needs of the components (motherboard, CPUs, drives, PCI coprocessors) are handled.

The connectors on the power supply also look deceptively like a 20+4 pin power connector and a PCIe port.

To the user on the outside it looks very simple, with a single 3-pin lead requirement. The power supply also supports a power self-test mode, presumably indicated by the green and amber lights beside the power socket.

The two fans at the rear of the power supply in the case ensure that airflow is from front to back based on the units orientation.

The sticker on the top indicates that this power supply is a Delta Electronics unit that satisfies the 80 PLUS Platinum standard, giving 90-92-89 percent efficiency at 115V and 90-94-91 percent at 230V (based on 20-50-100 percent loading). Interestingly the way this sticker is written seems that the power supply has a total of nine voltage rails, each limited to 18A.

Moving back inside the case, this is the side fan for the processors that interacts with the air baffle. It is easily removed by pulling on the red strip:

But the reason to take it out is to get a better look at the connector Lenovo is using for these fans.

The standard 4-pin fan header is moved into a bulkier 2x2 arrangement. On the motherboard, this looks like this:

Which is rather bulky and we won’t see it on anything designed to be commercial any time soon. As Lenovo is designing the whole system, they can almost afford to be liberal with PCB area. It is also worth noting that the power delivery cooling heatsink has a small cut-away in order to fit the fan.

The rear fan supplying the processors with cooler air is an AVC DV12038B12M, a 12cm ball bearing design to which there is very little information online.

For the PCIe slots under this fan, Lenovo uses a latch mechanism to keep the PCIe coprocessors in place. Pulling on the red strip opens the part up and the card is easily removed:

Similarly at the top, although only three PCIe slots are present here allowing for a big co-processor and a smaller connectivity card. The PCIe 2.0 x4 slot is open ended, allowing for bigger cards at reduced bandwidth.

On the other side of the chassis are the ODD bays. There are three possible, with one of our sample loaded with a DVD/RW drive. Rather than going for slim bays, these are full on allowing for other devices that require the room. The drives can be removed by pulling on the tag with the red arrow below the drive.

For people who want to add more drives in this area, the SATA ports are found beside the FLEX module on the bottom of the board, though it should be noted that only a few are available:

For storage drives, the P900 chassis comes beautifully equipped on the far right hand side. There is even a handy diagram to follow:

This allows for a 3.5-inch, a 2.5-inch or a combination to be used per bracket. These brackets have their own SATA and power connections already pre-attached, allowing them to be installed quite easily.

The brackets themselves are plastic, requiring the built in push-pins to put the drives in:

That almost sums up the design of the system, with a couple of final images confirming the SK Hynix DRAM:

As well as the side panel containing an easy to use diagram of what goes where in the motherboard:

The box contents for the P900 match those for the P300 – at a basic level we get a bundled mouse, keyboard, power cord and operating system/recover disks. Because Lenovo provides the systems as ‘Windows 8.1 downgraded to Windows 7’, it means the price of our sample comes with 8.1 should the user want it, but it ships with Windows 7 as the operating system at boot. Going back up to Windows 8.1, or a fresh install, requires these disks.

For users that are used to consumer grade motherboards, the BIOS design is basic at best which is unfortunate given Lenovo's prowess in chassis design. Rather than follow the route for a graphical interface, because these workstations are designed to be for software use rather than self-build configurables, we get a black on grey implementation showing basic lists of features.

It is also worth noting that Lenovo’s BIOS does not have a ‘save screenshot’ feature like many other modern UEFI setups. As a result, I apologize for the series of photographs.

The main screen of the BIOS gives us the brand name (ThinkStation P900), MAC addresses and a BIOS date, but in order to get meatier info on the system, pressing Enter on the ‘System Summary’ option tells us more of what is under the hood:

The E5-2687W CPU is found, although only one is mentioned here rather than the two installed. Total DRAM at 128GB running at 2133 MHz is also indicated, with the fans and audio/Ethernet. What is interesting is that due to the FLEX Module being PCIe based, we don’t see the XP941 drive shown here.

The Devices tab at the top gives a direct link to many of the controller and PCH based options:

Most of these are enable/disable, and the Thunderbolt menu requiring a Thunderbolt card to be installed. The Haswell-E chipsets technically have two SATA controllers, one that can hold six and is RAID 0/1/5/10 capable (‘SATA’) and another that holds four but is not RAID capable (‘sSATA’). Users can select between IDE, AHCI in these sub-menus and RAID also in the ‘PCH SATA’ option – it is worth noting that Lenovo only uses four from the SATA controller in the PCH.

The Advanced top tab focuses more on the intricate options of the system, most of which will mean nothing to most users.

The most important options, in order, are as follows.

Above 4G Decoding, for Xeon Phi type co-processors, is by default disabled and found in the PCI Subsystem settings:

The CPU Configuration menu gives more details on the processors in each socket:

Here we can enable/disable HyperThreading, as well as look into the power and control states of the system in order to adjust efficiency.

The Memory Configuration option allows the administrator to select ECC support, the memory frequency, the memory type as well as subtimings. Normally in this sort of BIOS we do not get this level of memory detail, however the administrator will have to know exactly what is under the hood to make sense of the options.

For example, in the memory sub-timings override menu, all the values are set to 0 rather than showing what the training/pre-testing did on POST:

One wrong setting could cause the system not to boot, requiring a BIOS reset. Lenovo’s BIOS team might benefit from looking at how this is done on more commercial products in order to develop this section of the BIOS.

The Memory Topology option is helpful in case the user notices the total memory of the system decrease. If one module is not working as it should, this screen should show it.

VT-d options for virtualization are enabled by default as shown above, however Intel TXT is disabled by default:

Our system came equipped with an Intel I218-LM and Intel I210 network interfaces which get their own screens in the BIOS. The I218-LM is designed for vPro capable products, whereas the I210 is a more general interface that allows virtual MAC addresses:

Security features include support for TCG, hard disk passwords, a fingerprint sensor (if one is installed) as well as a system event log.

The Hard Disk Password screen also provides an option for ‘Security Erase HDD Data’, which is presumably passing the Secure Erase function over the IDE protocol.

Users who want to use the P900 as an IO-less machine will have to navigate to the Startup tab and adjust the ‘Keyboardless Operation’ option to let the system boot without a keyboard attached.

Aside from this is the regular Save and Exit features. Unfortunately Lenovo doesn’t include a Quick Boot option, allowing for a one-time boot from a different device. This option is usually useful for users playing around with LiveUSB installs.

Though reviewing different types of systems and components, the software packages from manufacturers broadly fall into three categories. On the low-end we have the zero-software implementations, usually found on server motherboards or third-tier manufacturers. At the high-end is a plethora of testing options and settings that allow for the system to be configured and manipulated, as well as additional tools that some users will find useful. In the middle is where we found the ThinkStation P300, and in a similar situation the P900 is as well. It comes with a couple of useful tools for administrators, but nothing mind-blowing. Nevertheless, users rarely buy this level of system for the software – professional packages could easily cost twice as much as this system for a single license.

The main frontage is the Lenovo Tools suite, giving access to the Password Vault, Backup and Restore, an update system as well as diagnostics and recovery packages.

The Password Vault (or manager) allows users to create a username and password list for their browsers which are auto entered but not held by the browser itself.

There was no indication that the passwords were encrypted by the software but one would assume so given the nature of what the software holds.

Lenovo’s Solution Center, which we saw back on the P300, is a rather nice tool that shows the user when something needs critical attention. In this case, as shown above, the main drive has less than 10% capacity and there is no virus protection installed. (The first issue is my fault; I split the drive into two partitions for benchmarking.) We also get non-critical alerts, such as registration and regular scanning.

Alternatively the dashboard shows the issues from a separate system/security/checkup standpoint.

Individual elements of the solution center will show the issues at hand, as well as potential solutions. For example, here we get Disk Cleanup as an option, or a link to order more storage.

Whereas with the DRAM section we are told we have ‘enough installed memory’, and the tip is to ‘try closing programs’ to optimize performance.

The system also gives a full information report which can be saved and uploaded as part of a debugging procedure should the need arise.

The System Update tool is also identical to the one we saw in the P300, whereby the tool synchronizes with Lenovo’s server and checks for updated versions compared to those on the system. These updates are for both software and drivers.

As before, an update schedule can also be selected.

Lenovo also includes two storage based programs – one on cloud storage and one for sharing files over a network.

ShareIt acts as a WiFi peer-to-peer arrangement when a DHCP server isn’t available.

ReachIt combines multiple cloud storage accounts into one single interface, allowing users to search through them all from any device to obtain their files.

Video Conversion – Handbrake v0.9.9: link

Handbrake is a media conversion tool that was initially designed to help DVD ISOs and Video CDs into more common video formats. The principle today is still the same, primarily as an output for H.264 + AAC/MP3 audio within an MKV container. In our test we use the same videos as in the Xilisoft test, and results are given in frames per second.

The single thread frequency on a high core-count processor is usually lower than a similar TDP low-core count one, and as a result the low quality conversion performance is down on a good Haswell quad core. On the other hand, moving to double UHD sized frames allows the P900 to stretch its legs. Performance isn't exactly 3x the 5820K here based on core count, but it does take an extra chunk of performance.

Point Calculations – 3D Movement Algorithm Test: link

3DPM is a self-penned benchmark, taking basic 3D movement algorithms used in Brownian Motion simulations and testing them for speed. High floating point performance, MHz and IPC wins in the single thread version, whereas the multithread version has to handle the threads and loves more cores.

Similar results to HandBrake here, but the multi-threaded version of the benchmark is memory independent, allowing the 20 cores / 40 threads to stretch its legs.

Compression – WinRAR 5.0.1: link

Our WinRAR test from 2013 is updated to the latest version of WinRAR at the start of 2014. We compress a set of 2867 files across 320 folders totaling 1.52 GB in size – 95% of these files are small typical website files, and the rest (90% of the size) are small 30 second 720p videos.

Despite the semi-threaded nature of WinRAR, having 40 threads helps in compression here.

Image Manipulation – FastStone Image Viewer 4.9: link

Similarly to WinRAR, the FastStone test us updated for 2014 to the latest version. FastStone is the program I use to perform quick or bulk actions on images, such as resizing, adjusting for color and cropping. In our test we take a series of 170 images in various sizes and formats and convert them all into 640x480 .gif files, maintaining the aspect ratio. FastStone does not use multithreading for this test, and thus single threaded performance is often the winner.

FastStone relies on single thread performance, so we see it further down the charts.

Rendering – PovRay 3.7: link

The Persistence of Vision RayTracer, or PovRay, is a freeware package for as the name suggests, ray tracing. It is a pure renderer, rather than modeling software, but the latest beta version contains a handy benchmark for stressing all processing threads on a platform. We have been using this test in motherboard reviews to test memory stability at various CPU speeds to good effect – if it passes the test, the IMC in the CPU is stable for a given CPU speed. As a CPU test, it runs for approximately 2-3 minutes on high end platforms.

POV-Ray is also heavily multithreaded, allowing any big cores to push through.

Power Consumption

Power consumption was tested on the system at the wall. This method of power reading allows us to compare the power management of the UEFI and the board to supply components with power under load, and includes typical PSU losses due to efficiency. These are the real world values that consumers may expect from a typical system (minus the monitor).

Idle power consumption, compared to some of the other big systems, is actually rather good due to the 1300 W Platinum power supply. Load consumption for a 2P system was also lower than expected, giving 231W for both CPUs between load and idle power results.

Windows 7 POST Time

Different motherboards have different POST sequences before an operating system is initialized. A lot of this is dependent on the board itself, and POST boot time is determined by the controllers on board (and the sequence of how those extras are organized). As part of our testing, we look at the POST Boot Time using a stopwatch. This is the time from pressing the ON button on the computer to when Windows 7 starts loading. (We discount Windows loading as it is highly variable given Windows specific features.)

As with most 2P systems, POST time is rather long. There is no management IC on this system, but DDR4 training for each socket easily accounts for 20-30 seconds of the POST time.

Rightmark Audio Analyzer 6.2.5

Rightmark:AA indicates how well the sound system is built and isolated from electrical interference (either internally or externally). For this test we connect the Line Out to the Line In using a short six inch 3.5mm to 3.5mm high-quality jack, turn the OS speaker volume to 100%, and run the Rightmark default test suite at 192 kHz, 24-bit. The OS is tuned to 192 kHz/24-bit input and output, and the Line-In volume is adjusted until we have the best RMAA value in the mini-pretest. We look specifically at the Dynamic Range of the audio codec used on board, as well as the Total Harmonic Distortion + Noise.

The ALC662 is a laptop audio codec at best, and as a result the P900 joins the P300 down at the bottom of our charts. The THD+N result is actually not that bad in retrospect.

USB Backup

For this benchmark, we transfer a set size of files from the SSD to the USB drive using DiskBench, which monitors the time taken to transfer. The files transferred are a 1.52 GB set of 2867 files across 320 folders – 95% of these files are small typical website files, and the rest (90% of the size) are small 30 second HD videos. In an update to pre-Z87 testing, we also run MaxCPU to load up one of the threads during the test which improves general performance up to 15% by causing all the internal pathways to run at full speed.

USB 2.0 perfromance follows that of our other 9-series testing, while USB 3.0 perfomance for the P900 is similar to the P300.

DPC Latency

Deferred Procedure Call latency is a way in which Windows handles interrupt servicing. In order to wait for a processor to acknowledge the request, the system will queue all interrupt requests by priority. Critical interrupts will be handled as soon as possible, whereas lesser priority requests such as audio will be further down the line. If the audio device requires data, it will have to wait until the request is processed before the buffer is filled.

If the device drivers of higher priority components in a system are poorly implemented, this can cause delays in request scheduling and process time. This can lead to an empty audio buffer and characteristic audible pauses, pops and clicks. The DPC latency checker measures how much time is taken processing DPCs from driver invocation. The lower the value will result in better audio transfer at smaller buffer sizes. Results are measured in microseconds.

Normally 2P systems have horrendous DPC Latency numbers, but the P900 does rather well and matches some of our Z87 counterparts. Normally we prefer under 100 microseconds, but ultimately under 200 is a good result also.

Due to the market positioning of the ThinkStation P900 as an high level workstation with a Quadro professional graphics card, it makes sense to compare the system with industry-standard benchmarks and the other entry level workstations we have previously tested where possible. Benchmarks in this instance come from SPECviewPerf 11, a well-known multi-software test, SYSMark from BAPCo that is used by several industries for comparison, and Linux-Bench for some more scientific and synthetic analysis.

SPECviewPerf 11

Despite having 20 cores and 40 threads in the mix, as well as the highest Quadro we've tested with SPECviewperf, the P900 has a few issues in these benchmarks. Not only does the software package have to be able to cope with a split memory configuration between the two CPUs (a non-unified memory architecture), but the single threaded performance of the system isn't winning any awards. That being said, the ensight section of the benchmark shows good numbers, as well as a few others. Lightwave does take a hit though.

SYSMark 2014

SYSmark focuses on several tests, with the office tests being more single threaded compared to the data and financial analysis. As a result the P900 falls behind on the office section, but steams ahead with multi-threaded tests. The overall score puts the top spot with the P900.

Linux-Bench: link

Built around several freely available benchmarks for Linux, Linux-Bench is a project spearheaded by Patrick at ServeTheHome to streamline about a dozen of these tests in a single neat package run via a set of three commands using an Ubuntu 11.04 LiveCD. These tests include fluid dynamics used by NASA, ray-tracing, OpenSSL, molecular modeling, and a scalable data structure server for web deployments. We run Linux-Bench and have chosen to report a select few of the tests that rely on CPU and DRAM speed.

C-Ray: link

C-Ray is a simple ray-tracing program that focuses almost exclusively on processor performance rather than DRAM access. The test in Linux-Bench renders a heavy complex scene offering a large scalable scenario.

C-Ray doesn't have to deal with inter-CPU transfers or DRAM snooping, resulting in a good score.

NAMD, Scalable Molecular Dynamics: link

Developed by the Theoretical and Computational Biophysics Group at the University of Illinois at Urbana-Champaign, NAMD is a set of parallel molecular dynamics codes for extreme parallelization up to and beyond 200,000 cores. The reference paper detailing NAMD has over 4000 citations, and our testing runs a small simulation where the calculation steps per unit time is the output vector.

The NAMD libraries are designed to minimise any cross-CPU talking in order to extract performance out of an Intel system. As a result, we get a great result.

NPB, Fluid Dynamics: link

Aside from LINPACK, there are many other ways to benchmark supercomputers in terms of how effective they are for various types of mathematical processes. The NAS Parallel Benchmarks (NPB) are a set of small programs originally designed for NASA to test their supercomputers in terms of fluid dynamics simulations, useful for airflow reactions and design.

NPB gives a rather low score, based on millions of operations per second and per thread. As a result, a 40 thread system would easily outperform here, but the results are an indication of efficiency per thread.

Redis: link

Many of the online applications rely on key-value caches and data structure servers to operate. Redis is an open-source, scalable web technology with a b developer base, but also relies heavily on memory bandwidth as well as CPU performance.

Redis prefers single socket systems almost exclusively, and as a result despite having DDR4-2133 in the P900, the cross-talking for the database takes a hit in the result.

As I mentioned in the introduction to the review, Lenovo is at an advantage here due to the fact that the P900 is the first high end workstation we've had in for review in quite a while. As a result it is difficult to compare the experience with the other major workstation OEMs and easy to get carried away in the positive aspects. In order to address the balance, both Dell and SuperMicro are in the process or organizing systems for us.

Despite this, Lenovo's design is pretty well done and kudos are deserved to the engineering and design teams around the P900 series. One of the key elements in workstation chassis design is the ability for system administrators to update and replace components to minimize downtime. Lenovo's design ensures that almost everything is easily replaceable, especially when it comes to drives, storage, power supplies and fans. In an attempt to minimize cabling, storage is provided either via the FLEX system or integrated into a backplane that supports both data and power. The drives for the backplane work with two drives apiece, allowing for 8 drives on the backplane and a further four PCIe devices on the FLEX connectors.

Perhaps we could point a finger and say that the PCIe devices are slightly harder to adjust, requiring a screwdriver at a minimum, but the system splits the PCIe lanes from both the processors allowing as much bandwidth as possible should multiple coprocessors be required. In order to facilitate these devices, the design also includes direct airflow through three regions - at the top and bottom the air is blown through the front, over the storage drives and through the PCIe devices then out the back. For the two CPUs in the middle, air is taken in the back and split by an air baffle into two. Each air stream then passes over an individual CPU, and the air is then directed around the ODDs and then out of the front of the case. Due to Lenovo's design, this all happens due to the arrangement of the hardware and at a relatively quiet volume. When you start enabling the power viruses the fans do kick in at a rate, but nothing unexpected for a workstation.

Performance wise, we have to grin and bear that some of our performance metrics are not particularly suited to dual processor designs. Thus despite the high end system provided with two 160W ten-core processors, sometimes for regular workloads a single processor performed noticeably faster. It even affected our SPECviewPerf numbers, despite having the better Quadro than our P300 review. That being said, I always recommend that users looking into dual processor systems know their workloads and memory requirements. If you have the software to take advantage, it always makes sense, and in our performance numbers that could take advantage of this we saw large gains over a single socket implementation especially when large amounts of DRAM are required.

One critical juncture that will be noted is the price of the unit we were sent for review. At over $16700 for a web store purchased unit, it seems at least +50% more expensive than a self-build. We saw this in our P300 review as well, to which Lenovo explained that web sales account for single digits in percentage of sales and most customers are direct b2b or to resellers that outfit with their own hardware. The prices online, Lenovo has stated, means that they are not in direct competition with their resellers and b2b customers are directed to the regional representatives for pricing.

Thus despite the cost, the P900 is designed to be Lenovo's halo workstation product in a sea of potential contenders. From a purely design point of view, I loved poking and prodding with the chassis to see how everything connected and what went where. Whatever the end user might think, the sysadmin should find upgrading and replacing components easy.

Show more