Cruncher2014

Cruncher2014 is my second build in my modified Silverstone FT02 case.

 

 

The objective of this build is to achieve maximum compute performance for minimum power consumption. It is not intended as a high-end gaming machine and as such has a relatively modest Nvidia GTX750Ti graphics card. Being on the latest Maxwell architecture at 28nm, these chips are very power efficient and Crucially support 4K monitors on DisplayPort at 60Hz with G-Sync (I’m waiting for the Acer XB280HK to arrive). The Core i7-4790K processor is Intel’s latest and greatest (at the time of writing) and the Asus Sabertooth Z97 Mark 1 motherboard was chosen for its looks and high degree of instrumentation. Storage is all re-used: an OCZ Octane 256GB SSD for the system drive and two Seagate hard disks (3GB SATAIII and 1.5GB SATAII) for data files. Aside from a re-used BlackGold BGT3600 TV receiver card, the rest of the build is all Corsair: 16GB of Vengence Pro memory, a RM550 Power Supply and a Hydro H80i cooler.

 

The H80i fans are configured in exhaust rather than the recommended intake direction in order to support the vertical airflow concept of the Silverstone case. Note that the case fits the Corsair radiator assembly like a glove with the cable recesses in the top panel providing welcome extra space. It is almost like the two manufacturers cooperated on the design!

 

A tip I’d like to pass on to other PC builders is this product: www.sugru.com. This is a Plasticine-like putty which cures to hard rubber (but the cured form can easily be pared with a modelling knife). It comes in a variety of colours including black and adheres well to plastic and metal. It is ideal for making ad-hoc connector back-shells, mounting fixtures and covering parts that spoil the colour scheme such as multi-coloured wire tails. I made several back-shells for board-to-cable connectors (USB, HD audio) and the mounting for the two internal USB connectors for radio dongles and the direct-to-Bios reset switch inside the top cover. I also improved the look of three Silverstone Air Penetrator intake fans by removing a tacky label covering the circuit board recess at the center and filling that and the spiral cable duct with black Sugru hiding unsightly multi-coloured wires.

Here is the minimum load performance on Corsair’s Link software:

 

 

Power consumption is under 100W, less than half my previous build. For a constantly on machine, that represents a very worthwhile saving in electricity. Note that to get the PSU fan RPM and Amps instrumentation you have to buy the “Corsair Link Analog to Digital Bridge cable” in addition to the RM550. This can be hard to find – I got it for the princely sum of £5.94 on Ebay, plus an annoying £4.98 postage. It is the small rectangular dongle below the left hand edge of the motherboard. My mains energy meter shows that the RM550 is drawing less than 100W, so the efficiency of this PSU is very high – borne out by the fact that the fan is stopped and the PSU is barely warm to the touch! The top left part of the picture is generated by Piriform Speccy freeware and shows the bus multipliers on a per-core basis. Actually this is a snapshot and the multipliers are all over the place from a minimum of 8x to a maximum 44x.

 

Next is a maximum CPU and GPU load torture test:

 

 

I have the Uningine Valley Benchmark running at 1920x1080 with 8x anti-aliasing to load the GPU and Prime95 Small FFT torture test running for over 9 hours when this snapshot was taken. This represents a VERY severe test of the CPU. Throughout this test the multipliers were rock solid at 44x, so Intel seem happy with temperatures in the high 80s, and who am I to argue? I must say that the temperature gradient between the chip and the waterblock is disappointing and suggests the thermal resistance here is the limiting factor rather than between water and air. Notice that the power has risen only to 260W, so I have over-speced the power supply. The fan on the amazing RM550 is still only ticking over and in fact for much of the test it was not running at all (and the RM550 is still only slightly warm to the touch!).

 

The next test is a memory and GPU stress test:

 

 

As before except that I have used a custom torture test setting in Prime95 with an FFT size of 4096 and a memory size of 14000MB. This stresses the memory and the chip memory interface rather than raw compute power. You have to tune the memory so it does not page, which will lower the CPU load waiting for the SSD. I found that leaving 2GB for the Valley benchmark and other overheads insured the CPU load remained at 100% (with Speccy indicating 98% physical memory usage). Note that the total power consumption is 60W lower and the temperature is much more reasonable. Clearly there is 100% CPU load and then there is 100% CPU load!

 

In both the load tests the GPU temperature does not even graze 60°c running a demanding benchmark (it averages about 22FPS and looks spectacular), so full marks to EVGA and Nvidia!

This picture shows the top of the build with the lid removed:

 

 

The FT02 is a vertical airflow case which rotates the motherboard through 90° relative to the traditional tower case and provides space inside the top cover for connectors and cabling. This is a great idea both in theory and in practice. In theory it gives a clean airflow which is assisted by convection (hot air rises). In practice, it produces a much neater build and locates the cabling within easy reach (when the case is placed on the floor).

 

The main modification I made is seen on the left side of the case in the picture. The FT02 is supplied with conventional mounting for five front panel bays in the vertical face (the left side in the picture). Unfortunately filling these goes against the vertical airflow concept with modules such as a DVD drive seriously blocking the airflow through the vertically mounted hard drives at the bottom of the case. I re-engineered the case so it supports three bays in a carrier with various mounting options including the top-mounted configuration shown. There is a 5mm gap between bays which allows good airflow and also makes the carrier square allowing it to be rotated by 90°. I went to town on this idea as a proof of concept and the carrier can also be mounted in the conventional location and also facing either of the two side panels (which would have to be modified with a suitable square cutout). This gives four location options each with two orientation options for eight options in total including the one shown. Five of these eight options support vertical airflow. I have used two of the three bays in this build. The black module is a hot-swap SATA HD bay used for my rotating backup regime. The central module I built myself contains the front panel switches and indicators, the HD Audio jacks, two USB 3 ports and a laptop DVD drive in an unconventional “pop-up” orientation (the trade-off for vertical mounting bays is that you have to be selective about optical drives – a tray will not work here!). I have also fitted two USB 2 ports under the top cover (to the right of the bottom edge of the bay carrier) for radio dongles. The dongle for my Logitech keyboard/trackpad combo is already fitted and I intend adding a Bluetooth 4 dongle later. I was concerned if these would work under the metal top cover, but tests so far are encouraging. This assembly (made with Sugru - see first post) also accommodates a direct-to Bios power switch. This is a useful innovation on the Sabertooth motherboard which saves having to hit some specific key at exactly the right time during the boot sequence!

 

When I first un-boxed my FT02 a few years ago, I immediately noticed a problem: it is virtually impossible for one person to lift it safely! Even empty, it is very heavy and there is only one good handhold at the top (the cable aperture on the right of the picture). The only safe lift requires four hands in the air intake slots at the base of the case. To rectify this I fitted the “ammo case” handle you can see to the right of the bay carrier. This makes for a reasonably safe two-handed lift, although you still have to be careful not to put your back out! The handle folds neatly under the top cover and is fully concealed when the case is closed as shown in the next picture:

 

 

The final modification is to the PSU mounting on the right side of the case in the pictures. The airflow for the PSU intakes from a filtered aperture on the right case side and turns a right angle within the PSU housing before exiting with the rest of the airflow through the top of the case. However the intake has not properly lined up with any of the PSUs I have tried in this case. For the previous build I had a longer Gigabyte Odin unit, but the fan intake was still at the top and I put up with about a third of the fan area being occluded. For this build, I have made an extra metal bracket to move the PSU about 20mm lower in the case so the fan lines up properly with the intake filter.

 

For reference here is what the unmodified case looks like, showing the original bay arrangement:

 

The BGT3600 TV card is a few years old and re-used from the previous build. It even has analog TV tuners (as well as DVB-T and DVB-S) although these are now useless as the analog signal has been turned off in the UK. I’d not noticed the S-shaped distortion in the PCB in the previous build which was much more cluttered. Really guys, I hate to criticize a small UK company, but when you cannot even get the physical dimensions of a PCIe card right in your premium product it does not build confidence you have the necessary attention to detail to write robust device drivers! The problem is that the plane of the PCB (and therefore the PCIe edge connector) is a good 2mm displaced from its proper position relative to the back plate. This is in turn determined by an inadequate stand-off on the antenna connectors. No amount of gold plating makes up for an amateurish bodge like this!

 

Secondly, the need for an additional power cable. I realise this is necessary to support a dish pointing motor, but I found satellite reception would not work at all without it connected, even though I do not have a motorised dish.

As others have noted, the clamping hardware for the H80i waterblock/pump is not good enough. First build attempt I followed the instructions to the letter and the performance was worse than the Intel stock air cooler. Examination showed that the top left 1/3rd of the chip cap was not wetted with thermal compound and had not therefore been in contact with the waterblock. This was due to the clamp not being tight enough to take out the strain moment caused by the stress of the stiff water pipes at the opposite corner. I had to improvise hard rubber washers between the back plate and the motherboard, and also insert packers in the corners of the front support frame to enable the support screws to clamp down tightly. I can only surmise that the clamp was designed for much thicker PCB stock.

 

The “Analog to Digital Bridge cable” for the RM550 should be supplied as standard or preferably incorporated on the circuit board inside the PSU. The latter would save a lot of the combined cost since a separate enclosure and PCB would not be required and the cabling could be much simplified. It should not add more than a couple of pounds to the list price of the PSU. As it was, none of the discount suppliers of the RM550 bothered to stock this accessory and when I eventually found a supplier, it was irritating to have the trivial cost almost doubled by the postage! Another mistake is that both the USB cable for the H80i and that for the PSU dongle use a single USB channel in a dual-USB motherboard socket housing wasting a USB channel each. Single-row housings should be used allowing both to plug into the same socket. I carefully modified mine by extracting the contacts from one housing and re-inserting them into the spare row in the other, a tricky operation requiring good tools and knowledge.

 

The Corsair Link software is the best monitoring software I’ve seen, but it could (and should) be better at discovering non-Corsair sensors. For example Speccy (which is not manufacturer specific) finds temperature sensors on the two Seagate hard drives. It would be nice to be able to display this information in Link at the right location on the system photograph, but while Link detects the devices, it does not find any sensors on them. More disappointingly yet, Link only finds three out of twelve temperature sensors on the Sabertooth motherboard, and does not see any of the nine monitored fan speed channels displayed in Asus’s own AI Suite 3 software. Since Speccy does not see any Asus sensors, this lack of integration may be more Asus’s fault than Corsair’s. It takes two to interface!

 

It would be useful to have peak detection on sensor channels and configurable moving averages to make fast-changing values more readable and meaningful. It would also be good to have the option to display PSU load in watts rather than amps.

 

Finally, attention to detail guys! Although the graphical display knows RM PSU 12V Current is measured in amps, the taskbar menu shows it as a temperature “RM PSU 12V Current: 7.4°c”! There is even an empty “System Power Info” menu item available.

Part of Corsair’s clamping problem is really down to poor thinking by you (Intel) in the chip package design and the standardization of the chip and cooler mounting features on the motherboard.

 

Have you considered supplying your K series overclockable chips with machined copper caps? Such caps could have bolt holes in the corners providing a proper engineering solution to clamping the chip to a much simpler board pin array on one side and to waterblocks or air coolers on the other. With a significantly larger contact area, a better clamp and a more thermally conductive material, it should be possible to achieve a significantly lower, and more repeatable, thermal resistance between the chip and the cooler as well as imposing much less stress on the PCB.

 

For the non-K chips a simple aluminium or even plastic adaptor block could enable the current chip package to be installed in the same motherboard configuration without the expense of the machined copper block.

The Sabertooth is a great product and the Thermal Armour and TUF Fortifier (metal backplate) are valuable innovations. But your marketing is poor: since when did “Mark 2” mean a simpler, lower cost version of “Mark 1”?

 

If the lack of visibility of the excellent array of sensors to third-party software (see above) is your fault, please correct this as everyone benefits when components from different manufacturers play nicely together (and you could copy Corsairs idea of displaying sensor values over a system image in AI Suite).

 

One small further improvement: the socket shields for PCIe x16 sockets should be supplied as an x1 shield plus a separate shield for the rest of the socket. This way, if the socket is used for an x1 card, the unused part of the socket could still be covered.

I tried to persuade you guys to give or sell me a second top cover so I could cannibalize it to match the grill material when making my bay cover. Predictably, you ignored me. Makers of expensive enthusiast-level PC cases ought to sell the individual parts separately as damage replacements and sources for modification materials.

 

The design of the air intake for the PSU could easily be improved. Simply make the intake extend about 20mm further up the case. It would then fit a wider range of PSUs as well as continuing to support whatever long unit it was designed for (note that the case is not supplied with a PSU). The aperture behind the CPU on the motherboard carrier (for fitting coolers) is also slightly misaligned (or too small), but that seems to be an issue with many cases designed for the previous generation of motherboards.

 

I had to rebuild three times before I found a layout for the cables behind the motherboard which allowed the case to be closed up without visible strain on the back panel. An extra 5mm clearance behind the motherboard carrier would make all the difference and nobody buying a case this size is going to notice that much extra width! It could also use more cable tie anchors.

 

You should consider producing a PWM (4-Pin) version of your 180mm Air Penetrator fans to allow better speed control. The manual switch does not really cut the mustard any more since the ideal balance between noise and airflow is radically different for minimum and maximum load situations on a modern build. Also the open cable duct and the paper label covering the motor PCB are tacky on a premium product such as this.

 

Please also note the "Case Modifications" posting above - I would be happy for you to adopt any of these innovations in a future iteration of product. Recognition and freebes welcome, but not insisted upon!

This is a new maximum load test using the FurMark benchmark instead of the Unigine Valley Benchmark to stress the GPU. As before, this is running concurrently with with the Prime95 small FFT torture test to stress the CPU.

 

 

This appears to get closer to the maximum possible load as the load on the PSU has risen from 260W to 282W and temperatures (particularly for the GPU) are a little higher.

By purchasing an extra set of H80i/H100i cables, I was able to connect two of the three intake fans to the H80i pump bringing them under control and instrumentation of Corsair Link:

 

 

 

Unfortunately the Silverstone Air Penetrator fans are 3-Pin rather than 4-Pin PWM types (I have been unable to find 180mm PWM fans anywhere). This combination gives a rather poor range of control and the minimum speed was too noisy. The air penetrators have a three-position speed control switch and I found that the middle position allowed for a quiet enough minimum speed and an adequate maximum speed for this build (the maximum speed of the fan using the manual switch is 1060RPM). The motherboard fan headers by contrast accept both three and four pin fans and give a much better speed range for the air penetrators. Neither system however allows the fans to be stopped completely which is a pity as this should be perfectly feasible technically.

 

With the aggressive power saving modes on modern CPU and GPU designs cooling manufacturers really need to up their game. Manual speed control on fans is not good enough any more. All fans should be 4-Pin PWM and all fan controllers should be able to stop fans completely as well as controlling their speed over a wide range.

http://www.overclockers.co.uk/showproduct.php?prodid=WC-664-EK&groupid=701&catid=2331&subcat=2335

Thanks for the tip thenosbod. I'll probably try a couple of these when they become available.