Dell Precision 690 upgraded to Precision T7400 Motherboard


Well, I did it!
Took less than 2 hours in the end.
Excuse shoddy photos, almost didn’t take any but remembered how hard it was to find good data so took pity on everyone else.

What you will need:

Long reach crosshead screwdriver.
Hacksaw with bimetal blade.
Bastard file.
Magnet on a stick type thing. If you don’t have one just shred an old HDD and mount the curved magnet to a stick with cable ties or glue.
Good lighting.
Clear flat work surface.
Even though I have a decent workshop I found the bed to be best, good natural light and decent working height. And no solder splats or metal shavings.

Tools and Cards

^^^ Crappy DIY tools but decent bi-metal hacksaw blade ^^^

Xeon X54xx or X52xx CPU(s) + heatsinks from the 690 or a Mac Pro
RAM from your 690
RW199 Precision T7400 motherboard
MM776 T7400 front IO panel. FF219 is just the metal case
WY005 T7400 front IO panel ribbon cable
HX637 T7400 front audio cable (optional if you have a proper soundcard)


Riser 1 +2 0JF806 (0G9460 + Riser Riser 2 0H9376)
Riser 3 + 4 0JF807 (0M9008 + Riser Riser 4 0JF807)

690 MoBo

^^^ Precision 690 Motherboard ^^^

T7400 MoBo

^^^ Precision T7400 Motherboard ^^^

Notes on parts:

There are a lot of common parts. Air ducting, RAM risers, heatsinks, cables, RAM, PSU, case parts – most of it really. The front panel + cable and CPUs + motherboard are the main differences.
You will need the front panel and its ribbon cable.
T7400 and 690 versions are different although they look the same. There are auction listings for front panels that say they’re for both but its a nomenclature fail on their part. FF219 is common between them but its just the metal shell. The boards are different. The ribbon cables are very similar but have different key pins.
If you felt brave you could drill out the blank one but you’d have to connect to it too. Better to buy one, got my cable for £3 inc P&P, front panel for £7.50 inc P&P.
The ribbon cable is not 0.1″ pitch like an IDE cable, its 2mm or smaller pitch. Just in case you’d thought to use an old IDE cable.

690 Front Panel^^^ Precision 690 Front Panel ^^^

Behind Front Panel^^^ Precision T7400 Front Panel ^^^
PSU cables fit both motherboards fine.
Pro tip: take TONS of photos at every stage for reference.
Stripped out all cards and drives, bent all cables etc out of the way.

Case Cables
Disconnected all PSU cables, fan connectors, intrusion switch, front panel Firewire and the front panel ribbon cable.
Removed front dual fan assembly (2 screws).
Removed CPU/RAM air guide, heatsinks, RAM risers if fitted, RAM fan and support frame.
Removed screws holding motherboard frame in position (at rear by IO ports).
Slid motherboard tray out.
Unscrewed front IO panel (3 screws).
Unplugged Firewire and ribbon cable at front panel end.
Leave Firewire cable in place where it passes under motherboard.
Remove the ribbon cable, noting its route.
Beware when unplugging the ribbon cable for the IO panel, it has little teeth that hold it in and in mine they had become brittle with age or heat. They broke even with gentle leverage with a flat bladed screwdriver.

Clean EVERYTHING! Surprised at how little dust there was, but its best out of the way. For static reasons don’t do it till all active devices are removed; drives, cards, motherboard etc. PSU is OK.

MoBo IO Port Lineup

Stacked motherboards together to see the difference at rear.
Most IO ports line up, but a cluster in the middle don’t.

MoBo Slot Lineup

Card slots line up fine.
Took me a few minutes with a decent hacksaw blade and some files to make it right. Use a magnet in a bag to get the metal chips then hoover the case very thoroughly. The thought of what a steel filing might do stuck in a slot or between some pins on the motherboard made me do a good job.

690 Case Rear Hackery

T7400 PS2 Port Alignment

Feed the T7400 front panel IO ribbon cable through the slots in the case under the motherboard area.
Fit the T7400 front IO panel. You will find that only the 2 side screws will fit, the centre one is offset by about 1/4″. I couldn’t be arsed to remove everything in order to drill and tap a new hole, the IO panel is held fine with 2 screws and the power switch works so I left it.

When fitting the T7400 motherboard tray be sure to make certain you have engaged all the tabs in the case, you need about 5 hands to apply pressure in all the right places.
Use a couple of screws in the front fan assembly’s holes to hold the motherboard & tray in position while you upend the case to screw it in place from the rear (2 screws).

Motherboard wiring:
Plug in the front Firewire cable first, then intrusion switch, then the front panel IO ribbon cable.
Then HDD fan(s), power, then dual front fan block (remember to remove the temp screws first).
Replace SATA cables and cards, then drives, then done!

Getting there

Complete sans RAM Risers

My experiences afterwards:
T7400 BIOS was A01 when I got it, now A11.
Annoyingly you lose a PCI-e slot in the upgrade, its a more modern machine so why swap a useful 8x PCI-e for another PCI-X? You also lose the IDE port that the 690 DVD burner used.
The second GPU card slot won’t accept a RAID card or any non-GPU card, at least not so far. There may be an nVidia SLI chip mediating between the 16x slots that won’t work unless it finds some magic string in the GPU boot ROM or something horrid like that.
In the 690 I had a Revodrive 240Gb in the first PCI-e, then nVidia GTX660 or Quadro FX 4600, then PCI-e RAID HP P400 512Mb + battery, then Soundblaster X-Fi in a PCI slot. Since replaced the X-Fi with an Audigy 2 ZS as it sounds better than the X-Fi.

1600Mhz FSB CPU with 667Mhz RAM instead of 800Mhz

Because the RAM in my 690 was 667 not 800 I was unsure if a 1600Mhz FSB Xeon X5492 would work, there is much contradictory advice out there.
I can assure you that it DOES work, just at 640Mhz rather than 667. Or so the BIOS says.
I got a cheap Xeon X5272 (dual core 3.4Ghz 6Mb cache 1600Mhz FSB) just in case it wouldn’t work. X5492s are still a bit expensive for an experiment.
I imaged the Revodrive before starting just in case, but it booted fine.

“God said 640×480. Realism is over-rated.”

TempleOS style CRT shot from later reinstall

^^^ Screen ‘shot’ from a later reinstall ^^^

Fine for TempleOS, not so much for me.
First boot was 640×480, found about 35 devices, installed drivers automagically, rebooted twice, all fine.
I’m allergic to all these registry/driver optimiser junkware programs but there is one that actually works: DriverMAX.
Used it on the 690 as well, it finds much newer drivers than windows or Google does. Its a legit program, surprisingly. I only used it because there’s a blog post somewhere by a guy upgrading his 690 to Windows 7 x64 and he reported good results.
Once you’ve used it to get all the drivers you need it can export them as a zip for later use.
You can then uninstall it.

After the T7400 upgrade I noticed that a few times a minute my audio would stutter. I put it down to crappy X-Fi drivers and change of mobo. I used DPC Latency Checker from here.
Its free and shows latency spikes. Very useful.
Sound card was a red herring, I had the Intel Matrix Storage Manager installed on the 690 and as soon as I killed its process and deinstalled it the problem went away. I replaced the X-Fi card with an Audigy 2 ZS anyway because it sounded better.
Its only a problem if you don’t do a clean OS install.

Great Success although not photographicly

^^^ Screen ‘shot’ from Windows 7 x64 Experience Index ^^^

Machine is not quite as fast as it was (Solidworks Performance Test – desktop is very snappy) but I put that down to going from 8 cores at 3ghz to 2 cores at 3.4Ghz as well as only having 1 CPU. Probably memory channel efficiency and other chipset and bus related stuff, you need to have both CPU sockets populated to get quad channel RAM. CAD is  mainly a single thread performance thing apart from raytraced rendering and some physics simulations.
When I have a pair of X5492s in there it’ll fly. (Later note: it did)
800Mhz RAM is still a bit pricey compared to 667, but I’ll gradually get that too.  (Later note: ended up with the RAM risers and 16x 2GB 800Mhz as well).

In summary its well worth doing, lots of crap info around about what’s needed and how to do it but I hope this will be the definitive practical guide. I know its a bit verbose, but it should be helpful.

Its 2013, but these machines are still fast. And absolutely stable, which means a lot for CAD, code compilation, rendering – anything iterative that runs for a good while. With a bit of careful eBay lurking you can get a 690 to a T7400 for under £50.
There are still quite a few people out there with love for these machines. I’m a Starving Hacker(tm), but even if I COULD afford some i7 quad SLI Titan monster I think I’d rather spend the money on a top end scope or tools or something.
I’m even still using a CRT monitor, but it is a 24″ widescreen @ 1920×1200.
I may get a 1440p 27″ IPS when it dies.

Most of this is from a forum post I intended to randomly drop somewhere but never finished. Better late than never I suppose? Let me know if there is anything more I need to add.

So you’ve got a Broken Thing…

So you’ve got a Broken Thing…


First Step:

What’s your time worth? Is it worth 4 hours of your time to repair a £10 commodity item? Clearly there are “sentimentality” and “convenience” modifiers that can be applied.

Everything is a time/money trade off. Modern equipment, with some honourable exceptions, is not really meant to be repaired. You need to be prepared to improvise.  See bottom of page for notes about tools and test gear (spoiler: you don’t need as much as you think you do).

Bricolage, as well as being an great album by Amon Tobin, is the French word for a worthy improvisation or creative workaround. Jugaad is an Indian word with similar meaning. Its sad to note that the English ‘bodge’ has pejorative connotations. But we’re post-industrial now, with our Ballardian landscapes of flyovers and endless callcentres and cheap conference centres. But that’s a rant for another time.

Stop and Think Step:

Probably half of all ‘faults are caused by operator error. It is essential that before you roll up your sleeves and operate you should take a photo with your phone of any switch or dial settings, power off,  then set them all to their defaults. If you haven’t found a manual use your intuition. Then restart.

Most things with a microcontroller inside (which is most things these days) have lots of exception handling code and the classic method of yanking the power and leaving it off till the caps have discharged before powering up again will often force it to reinitialise itself and clear whatever pathalogical condition it has found itself in.

Once you’re inside (see Third Step) you may find a jumper to clear internal settings. Be careful with this, without a (service) manual you might brick it.

There are many easy wins to be had here, thanks to the unappreciated firmware coders.

See Sixth Step below for more on dealing with embedded systems.

Second Step:

Check the fuse or external PSU. Check any mechanical connection ie leads, connectors, etc.

Third Step:

Gain access to the internals.


It’s natural to want to get stuck in but it is so easy to snap off little plastic tabs or gouge edges levering with a screwdriver. It’ll never look right or fit together properly again; OK for your own stuff but incompatible with scoring brownie points by reviving yer gal’s favourite curlers.

Check for hidden screws under stickers. Some cunning manufacturers use 3 normal screws and one tri-lobe or similar. If you don’t have a security bit set its sometimes possible to jam in a smaller flat bladed screwdriver and turn the security screw. Failing that just drill it’s head out with a small carbide burr or stub drill. Rigidity is important to avoid drifting off course. Secure the work and use a drill press if you have one.

For cases that use those horrid little click tabs I’ve found that guitar picks are an excellent way to avoid leaving gouge marks like you would if using a flat blade screwdriver. Some picks have a very thin edge. I’m sure there’s a whole vocabulary to describe them if you’re into that sort of thing. They’re cheap, ‘used’ ones can possibly can be got for free from music shops and studios.

Some power bricks are joined by ultrasonic welding or glue. In these cases you have no choice but to use a hacksaw or mini-tool cutting disc and glue the case back together afterwards. It can be done neatly but it takes patience and a steady hand.

Fourth Step:

Clean it. Use a 1″ paintbrush to remove dust and clean fan blades. Sometimes hoovers can develop a large static charge from the air movement past the nozzle. This is bad. PC World had to rethink their ‘PC Health Check’ after zapping quite a few motherboards this way.

Check EVERYTHING mechanical. That means anything that isn’t glued or soldered together permanently. Unplug and replug all internal wiring looms, check for obvious corrosion or weak pins/sockets. If it has linkages or other moving stuff check for broken parts or foreign objects jamming movement. Try and sequence them by hand, does it operate smoothly? Beware – some mechanisms use a worm gear on the motor as the first reduction so if you don’t apply the test movement to the motor shaft you may well break the worm.

Plastic Worm on Motor Shaft

Be careful when lubricating. Some lubes will attack plastics or mix with dust and form a gummy paste that makes things worse. A lot of plastics are self lubricating anyway and most bronze sleeve bearings are made of sintered metal and are already impregnated with enough oil.

‘Mechanical’ is also electronic, reseat any chips that are socketed. These days its rarer to find but in slightly older gear its often the case that system firmware is stored in a DIP or PLCC socketed EEPROM. Pins corrode but often just reseating the chip is enough to scrape through the muck, just like with connectors.

Fifth Step:

Electronics. First thing to check is electrolytic capacitors. If you’re lucky they’ll look like this:


and it will be obvious. Not always though. When they’re in a reasonably high power PSU and are no-name caps I would replace them all anyway.

Look for broken power devices and any sign of excess heat.

Cracked MOSFET

Burned Resistor

In some switch mode PSUs there can be slight discolouration on the board below some power resistors, this is a red herring and fairly normal. The above is not that.

Sixth Step:

There are often unpopulated TTL level serial headers or JTAG headers that can be very useful. If you have JTAG stuff this post is probably too basic for you.

Serial headers are usually a group of 4 unpopulated holes of 0.1″ pitch. You can solder small patch wires to them but its easier and more reliable to solder in a section of pin strip or 0.1″ pitch male connector. The Vcc connection is not required, just Rx, Tx, Gnd.

Pin Strip 3 way0.1″ Pin Strip (usually comes as strip of 40 very cheaply)

Unpopulated Serial Header Example

Unpopulated header

TPLink TL-WR2543ND-serial-pinout

TP-Link TL-WR2543ND Router with header added. Note no V+

TTL serial headers can be very useful, all you need are one of the £2 USB to TTL serial dongles from ebay or Amazon; just hook the 2 data lines and the ground, ignore the power line. Then get PuTTY, and assuming the device has a linux core like so many things do these days, when you power up your device the PuTTY window will show you the boot data and may well let you change the firmware. Most embedded systems have, to their shame,  default logins. If not have a search around, its likely someone else has had to face this before.

If you get nothing over serial first check your PuTTY settings, then your connections, then try swapping Rx and Tx. Or maybe it isnt a serial header, or maybe the MCU is fried. If you have a scope try probing the Rx and Tx lines as you power the device up, you should see some random looking square waves. You might have to switch the scope to DC coupled mode as the serial bitrate might not be high enough to masquerade as an AC signal and make it through the DC blocking caps. Even if it does the signal edges will be very distorted.

I had some drama with a TP-Link TL-WR2543ND Router whilst upgrading its factory firmware to DDWRT that involves most of the above, there’ll be a post on it soon with details on how to set up PuTTY and TFTP etc. Far too much data for a general purpose page like this. Once its up I’ll link to it from here.

I will also make a page on converting ATX PSUs into general purpose PSUs, including how to wire them in series without an explosion! ATX PSUs are essentially free these days from scrap PCs; even buying new ones costs about £12 for 500W.


Power => connectors => moving things => capacitors => power devices => main silicon => firmware

Component Testing and Tools:

The minimum you will need is:

Assortment of screwdrivers, mains tester screwdriver, pliers, cheap multimeter, soldering iron, solder sucker, croc clip jumper wires, ATX PSU, 1″ paint brush (unused).

Nice to have:

Security bit set, ‘lab’ adjustable PSU, soldering station ie temp controlled, fluxed solder wick (I like Chemtronics Chem-Wik), a real oscilloscope (NOT a crappy audio bandwidth USB or soundcard based one), wrist grounding strap, LCR meter or AVR based component tester.

If you don’t recognise any of these things just select the text, right click, do Google Image search.

The AVR based component tester needs a mention by itself, see below. I happened to have a pretty posh digital LCR meter already, LCR400 LCR Meter.


LCR Probes

The AVR version does everything it does plus an enormous amount more. Maybe not as accurately but its cheap, £8.20 delivered. Needs 20 mins soldering up but no SMD. Based on the brilliant Open Source work of a German group headed by Karl-Heinz Kübbeler at (hence the cheap Chinese kits). The firmware is updated fairly often. Depending on if you have the reference hardware version or not you may need to fiddle with it a bit. There is usually a how-to somewhere for most of the Chinese clones when a firmware update comes out.

AVR Tester

Apart from having a semi decent multimeter its the most useful bit of test gear you can own. Within 2 days of making one I had tested, categorised, and stored ALL of my unknown or reclaimed component tubs. That was a lot of caps, diodes, FETs and inductors. It will tell you the function, pinout, and characteristics of almost any component that isn’t an IC. Like a magic trick. Many more functions than the Peak analysers, also <10% of the price. Accuracy caveats apply but it’s easily good enough. You zero out the probes or whatever when you set it up. ESR and other cap measurements are accurate and quick. Just hook the unknown up any pin to any pin, it works it all out its own self and tells you. Don’t connect a charged cap!

As copy’n’pasted from it’s bumpf the AVR one can do:

Automatic detection PNP and NPN type bipolar transistors , N, P -channel MOSFET, JFET FET , diodes , two diodes, thyristors , resistors , capacitors , inductors. Automatic detection pin definitions.
Measurement of the bipolar transistor current amplification factor (B) and conduction voltage emitter junction (Uf). Darlington transistor may be identified by the amplification factor of the high threshold voltage and high current.
Can be detected inside the bipolar transistor and MOSFET protection diodes and displayed on the screen.
The threshold voltage and the MOSFET gate capacitance measurement.
Support two measuring resistors , the potential can also be measured . If the potentiometer is adjusted to its end , the tester can not distinguish the two ends of the pin and the middle.
Resistance measurement resolution is 0.1Ohm , the highest measurement value of 50MOhm.
Capacitance measurement range from 25pf to 100mF (100000UF). Resolution up to 1PF.
Can detect more than 2UF capacitor equivalent series resistance (ESR), with a resolution of 0.01Ohm. This feature is very important for the detection capacitor performance.
You can display symbols of the two diodes in the right direction , also shows the forward voltage drop.
LED detection of diode forward voltage drop higher than normal . Dual LED detected as double diodes. Simultaneous detection of light-emitting diodes will flash.
Each test time is about two seconds , only large capacitance and inductance measurements will take a long time.

New conversation function, you can select the following additional functions via the menu:
1. The square wave signal generator, optional built-in within the range of 1HZ-2MHZ stall square wave signal;
2.PWM pulse signal generator for 1-99% of the pulse width modulated signal;
3. The frequency meter function, can test 1HZ-25KHZ or higher frequency signal (in this case a slight decline in accuracy, this feature should make their own extensions)


Things, Ideas, People

I’ll add to this periodically, it’ll be a list of stuff its worth paying attention to.

In the mean time I’ll just leave this here:

XKCD, sublime engineering and human stick people

JWZ, wrote lots of Mosiac – eventually became Firefox. Has an amusing and interesting blog. Runs a goth/rave nightclub now.

Bunnie Huang, another interesting blog about having hardware made in China. Noted open source hardware designer.

Max’s ‘Little’ Robot Shop, amazing stuff. The man is making his own injection moulding machine amongst MANY other hardware and software projects. Makes most things from scratch and documents it. Amazing skills.

Dan Gelbart has an 18 part series on prototyping skills, processes, materials, and tools. Its HERE. I consider it compulsory watching for anyone building one off or low volume prototypes or instrumentation, its a goldmine.

I may embed it all at a later point.  Here they are:


^^^ THIS ^^^ is extremely dark. You have been warned. Also very well made.



On building PCs or buying workstations

I’ve owned and used a pretty wide array of computers over the years.

Started with a BBC Model B at school, access to an Amstrad PCW8256 (Z80, 256k, CPM and Logo, nearly proper computing!), then an old NCR COBOL based batch processing machine with a 9″ text only monitor and seperate dual 8″ floppy drives. Had about 16k of RAM. Used various Sun and SGI machines at university.

My first PC was an IBM 8086 motherboard with 64k RAM. Individual socketed chips, no I/O at all bar keyboard – all on 8 bit expansion cards. If it couldn’t find a boot sector it would load MS BASIC from ROM! Played with that in the early 1990s.

Then usual array of 386, 486, Pentiums etc. Used to assemble PCs from parts and sell them for a decent markup. A complete system with decent specs was at least £1k, so making one up for £600 led to a decent income. Sold about 10 – 15 a month for quite a while. Around the Pentium 4 era it stopped being worthwhile except for friends. Still used to make up my own machines till about the Core 2 era. I used AMD or Intel CPUs depending on price and performancealthough AMD dropped the ball CPU wise after the 64×2 line. Then I discovered the wonders of off lease workstations.

It was a tossup between HP and Dell, I happened to find a good deal on a Dell Precision 690, 2x quad core Xeons, quad channel 32GB ECC RAM, RAID, tons of slots, nice quiet (very heavy) case.

1336071486_368955464_5-dell-precision-t7400-workstations-punjab DSC_0851 T7400



It was a revelation: a PC that NEVER EVER crashes! Everything just works. Dell Precisions or HP Z series are good bets.

Haven’t built myself a PC since. Workstations seem to be overlooked by the masses so the prices stay low. Between the Xeons, ECC RAM, good thermal and hardware design, and vendor software verification they seem to be indestructible. The cases are huge, tools are not required, the fans are large so run slow and quiet – and the fan control firmware is sanely configured. ECC RAM is much cheaper than desktop, as are the Xeons. They all have serial and LPT ports too, handy if you mess with electronics. As they’re Xeon based they have all the hardware virtualisation CPU instruction so VMs run at close to bare metal speeds. With 8 or more cores you can just use affinity to shunt your VMs off onto the last few cores and not even notice they’re running.

I moved onto a Precision T7400, then my current Precision T7500 (dual quad core i7 Xeons with HT (hex core available but £££), 96GB DDR3 ECC, 240GB PCI-e SSD, etc). 8 cores, 16 threads. Not quite current gen but its amazing how quick it is. The new AES instructions mean that Truecrypt whole disc encryption is essentially free. Assuming you trust Intel that is…

Downsides are size and power consumption. The T7500 is more reasonable power wise. Buying a current model would bankrupt most, decent spec workstations are £10k – £15k new.

Personally I’d rather have a second hand Mercedes than a new Ford. Or, as someone cleverer than me once said: ‘The bitterness of poor quality lingers long after the sweetness of low price has gone.’

There are laptop versions, I have a Dell Precision M6400. It only has a Core2Duo era CPU but its fine for everything I use it for. Backlit keyboard with numeric keypad, jog/shuttle mousepad, 4x RAM slots, 1GB Quadro FX 3700M, 2x drive bays with RAID, TONS of I/O, 17″ 1920×1200 EdgeToEdge glass RGBLED screen with wide gamut. Its beautiful but quite heavy. The 230W PSU weighs as much as some netbooks! The case is pretty much completely made of metal, looks great. Cost £4700 new, I paid £220 a couple of years ago.



I can now just USE my machines. Don’t have to think about them at all, they just work. There’s so much CPU and RAM spare these days you don’t have to get too uptight about some driver or Windows process running in the background.

Windows 7 x64 is stable and fairly lightweight, I will eventually ‘upgrade’ to Windows 10 but I’ll let the early adopters iron out the bugs first. I’d like to use Linux on principle but its just not there yet; I can’t be bothered with the constant fiddling. Fine when you have 3000 identical servers to admin, not such fun with a single desktop. And most engineering software etc only supports Windows and doesn’t like WINE. Eagle PCB is an honorable exception.

I do use LinuxCNC, but thats a topic for a different post.