Scott Savage’s Blog

It appears that I have suffered the curse of Nvidia again, this time on my brand new work unibody Macbook Pro 13in. Under Windows Vista and 7 the Nvidia Geforce 9400M graphics card can only send a 640×480 resolution signal to projectors through the official Apple Mini DisplayPort to VGA adaptor.

It appears that I am not the only one, there is a large thread that is continuing on the Apple Discussions forum. There are a bunch of other people reporting the same issue, but Nvidia doesn’t have a solution yet. In fact, they have flat out said there are a huge list of unsupported (and will NEVER be supported) features. The following are from page 16 of the 190.62 Nvidia drivers release notes:

The following are features and functionality that were available in driver releases
supporting Windows XP, but are not–and will not be–available in driver releases for
Windows 7:
• High resolution scaling desktop (HRSD)
• MultiView Display Mode (for NVIDIA Quadro NVS graphics cards)
• NVKeystone
• Unified back buffer (UBB) controls
• OpenGL Video Overlays – This is an operating system limitation.
• Overclocking – GPU overclocking is no longer supported in the default GPU driver control panel. This feature is available in the NVIDIA System Tools software, which you can download from NVIDIA.com.
• GPU Temperature Monitoring – Temperature monitoring is no longer supported in the default GPU driver control panel. This feature is available in the NVIDIA System Tools software, which you can download from NVIDIA.com.
• AGP Settings Adjustment
• Video Zoom
• Pan & Scan ‐ the process of panning across the desktop in order to display a desktop on a monitor with lower resolution
• Per‐display Desktop Color Setting Adjustments – For Clone mode, the desktop color setting adjustments through the NVIDIA Control Panel can only be made across all displays in a system, and not on a per display basis.
• Per‐display Video Color Setting Adjustments – For Dualview mode, the video color setting adjustments through the NVIDIA Control Panel can only be made across all displays in a system, and not on a per display basis.
• Edge Blending
• Run display optimization wizard
• Run multiple display wizard
• Run television setup wizard
• nView Horizontal and Vertical Span Modes – Due to architectural changes in the new Windows Vista Window Display Driver Model (WDDM), span mode cannot be supported in NVIDIA graphics drivers. NVIDIA recommends using the built‐in Windows Vista multi‐display modes.
• Display/Connection Wizard (such as was provided with Windows Media Center Edition)
• DVD/MPEG Extensions (such as was provided with Windows Media Center Edition)
• Audio Extensions (such as was provided with Windows Media Center Edition)
• NVIDIA nView Desktop Manager – The nView Desktop Manager will not be included in drivers for GeForce products.

If you are seeing this post then you have reached the new server for my blog. I moved to Crucial Paradigm as I found their Windows VPS servers to provide good HDD limits and low Australian pings. Please let me know if you experience any improvement in speed or accessibility. This is all part of my plan to improve the exposure and performance of my blog this year, as well as launch other paid services.

I have mentioned Pipe Networks before, more specifically their “Project Runway” surviving a near death credit crunch experience. There have been two interesting developments since that time.

Guam Beach Cable Hole

The first is that Pipe have put up a blog tracking the progress of the cable installation. You can view it at pipeinternational.com. As with their DC3 data centre blog they have uploaded a ton of photos and commentary on a very regular basis. I am sure customers, creditors and other participants in the Australian networking industry appreciate the transparency of information delivered through these blogs. It is fascinating to see that, in the end, the data is flowing through a rather unglamorous combination of copper, concrete, seawater and sand. It is also great to see a company recognise the power of blogs as PR delivery mechanisms. I am sure they raise the public profile of the company significantly.

Also on the PR front the Pipe Networks CEO, Bevan Slattery, participated in an interview with Business Spectator’s Isabelle Oderberg. This interesting interview starts off with a glimpse at how close the project came to collapsing, and concludes with some comments regarding the NBN process and goals. I do agree that the Government is tackling the problem from the wrong end, they should be starting at the core and moving outwards. I guess it is easier to present end solutions to the common voter, you can’t deliver bit size statistics like 50% more bandwidth will mean 50% wholesale price cuts which will mean 50% consumer broadband saving.

Maybe the same is also true for the NBN tenderers? If the tender was only for backhaul, could they be guaranteed enough customers to viably support duplicating the infrastructure? Do they need retail customers and the bundling of products to put together a firm business case and profit margin? Again the big question is, if this rollout is viable then why has no-one done it already? At the end of the day the problem always lies with Telstra, they simply own everything that matters. Scarily, this monopoly is now extending into the wireless spectrum. Backhaul competition is a great first stage, and perhaps with the current financial state the tender process should be limited to that for now. But there should always be a second stage planned, one that makes the whole end to end process competitive. Until that happens we will be facing different versions of the current monopoly.

Personally I believe that Telstra should be seperated and that the core infrastructure (backhaul, exchange buildings and ducts) should never have left the public’s hands. There is enough infrastructure around, it is just being crippled to maximise profit and prevent competition. Now we are either going to have to buy them back or build around it, either way this is not an efficient or cheap process.

Cloud Computing should be understood as an evolutionary rather than revolutionary approach to delivering software. I think the rPath video below is the best explanation I have seen of Cloud Computing. It represents this evolutionary process without getting bogged down in the technical jargon, it even uses a car metaphor (which everyone loves).

Blackberry make good phones, regardless of whether you use them for email or not. Like everyone I am trying to cut costs at the moment, and one of those costs was my Blackberry plan. Optus charge a minimum of $24.95 extra a month for a Blackberry “bolt-on” ($29.95 without a 24 month contract), and on top of my existing cap contract this was getting expensive.

So I dropped the bolt-on. Everything seemed fine, with the only noticeable change being that the email icon disappears as expected. Then I try browsing the web. No dice. Google Maps doesn’t work, even the Optus Zoo home page doesn’t work. I call Optus and they advise me that I need the Blackberry plan (with BIS – Blackberry Internet Service) or a 3G phone in order to get internet. Surely not!

Anyway after a bit of googling I stumbled on a couple of good sites. The first is Ross Barkman’s GPRS settings listing page. This gives me the APN name that I plug into the Options -> Advanced Options -> TCP -> APN: field (although I left the username and password blank). The second and really great site by Anworm provides some Service Book configuration files and neat Java tool. The basic steps are remove the browser, modify your Service Book, and then reinstall the browser. Although he doesn’t say this on his website, other apps such as Google Maps Mobile also need to be removed and reinstalled in order to pick up the new non-BIS settings. Interestingly not all apps require this Service Book hack, some other apps such as Opera Mini will find their own non-BIS way or working even without the settings. Who knows why? At least everything seems to be working nicely over GPRS now, just watch your data usage.

We recently purchased a Netgear ReadyNAS unit (formerly made by Infrant Technologies). It is a nice compact little unit, 1RU with 4 hard drives across the front. It runs an onboard Debian install with some custom software to support X-RAID, the front panel buttons and a nice web interface.

We have run into some issues lately in relation to the performance of the device over the network. Their general advice is to do a direct connect and check your network drivers, but this hasn’t helped our fault. When logged into the SSH server on the system I can see that the CPU is running at 90%+ pretty consistently during usage. These are smb processes running under the various usernames that have access to the file shares. Even when the desktops are idle they are chewing CPU cycles on the NAS.

After about 24 hours of usage the NAS starts to become unresponsive. In particular the web interface actually crashes the browser (both IE and Firefox). I am trying leaving oplocks turned off at the moment as some people have suggested, but I am not seeing any reduction in CPU usage. Apparently these problems have been fixed in the latest beta, with the next prod version due in ‘a couple of weeks’. It can’t come soon enough as far as I am concerned.

�

The relentless pace of technology is exciting, but sometimes this fast paced environment
leaves its past behind a little prematurely. I was on the receiving end of one such
problem this week.

I purchased a new ATI graphics card, the x1950Pro AGP version to work with my existing
nforce3 motherboard. It is running an Athlon X2 processor and has 2GB of RAM, so it
isn’t an outdated machine by any stretch. I plugged in the new card expecting everything
to go smoothly, however it kept failing with a code 43 “Device
failed to start” error. Vista recognised the card correctly, however it just would
not leave the graphical ’safe mode’. I initially suspected the card was faulty, however
after a bit of googling I managed to find others with the exact same problem.

In fact not only did I find others, I found an acknowledgement
from Nvidia themselves of the problem. It was snuck in at the end of one of their
website FAQ’s and reads:

There is a known issue with ATI AGP cards with NVIDIA nForce3 and Vista. This
is currently being looked into and will likely be resolved with an MCP driver update.

In fact this problem has been known about and acknowledged since the start of this
year, and reported on the 7th of
February. Still nothing has been done, there is no MCP update and I am still left
sitting out in the cold. Nvidia has been infamously slack with their Vista drivers
in general, now they are blocking ATI cards too (Nvidia cards work fine)? I think
that is a little too convenient and completely unethical to be honest.

The following photo set shows the progress of the rack rebuild that I performed yesterday.
It starts with the old rack setup and then progresses through each level of the build:

It was good fun putting all the new equipment into the rack, but it is very time consuming
especially when everything needs to be adjusted to fit. We had to pull all the vertical
rails as well, so everything had to come out and be rebuilt. It is fun doing it while
the everyone is at work as well, you have to plan the build so you only have one shutdown
during the lunch break! It all ended up working out nicely in the end, everything
is on rails and is easily accessible.

My suggestions for anyone looking to build/rebuild a rack are:

• Allocate twice as much time as you think you will need

• Pull as much out as you can

• Don’t start running cables until you have as much equipment in as possible

• Get a bag of zip ties

• Plan everything beforehand

• Get as many decent screwdrivers and spanners as you can

It is worth it in the end. We now have better air flow, if there is water in the room
the servers will be fine, better accessibility (important especially in emergency
hardware failures) and better physical security.

Tonight I performed a tower-to-rack conversion of the Naiman Clarke primary server, a new Dell Poweredge 2900. You may ask why we didn’t just get the rack mountable 2950 version to start with? Well that IS what we ordered, however Dell got a little confused along the way. Anyway they made it up to us with some good discounts and a free rack conversion and rails kit.

Tonight I simply changed the face of the server and moved the control panel so it will be correctly oriented when the server is rack mounted. This was a bit of a process however and required the removal of all drives and bays as well as around 60 Torx screws. You can see these lying around on the floor while I am in the process of reassembling the machine. I also took a few photos of the internals just out of interest. There are some pretty big and noisy fans in there! I thought Woodcrest chips were meant to be power efficient and quiet?

Hover over the pictures for a description:

The Zalman VF700 is a flower style copper heat sink and fan that can be installed on a variety of graphics cards, including the Nvidia Geforce 6600GT (both AGP and PCI Express). My version is a Gainward 6600GT AGP 128MB Golden Sample. One major problem with the AGP installation however is that the PCI Express bridge converter causes the GPU to be set higher on the card than normal. This results in a 30mm overhang as shown on this Zalman installation instructions site.

As I have a desktop style computer case (Antec Overture II) there is no room for this overhang, and I really prefer to have the lid on my case. As a result I decided to embark on a bit of a modding project. Basically the way the heat sink is built one side has much larger fins than the other, and these fins are the ones that overhang. If I could rotate the heat sink 180 degrees then this would no longer be a problem.

 First of all the old heat sink must be removed. This is as simple as pushing together and then pushing through the 3 white pins that hold it in place. It comes off very easily and is not held on with heat transfer adhesive like some other GFX cards. 

With the heat sink removed you can clearly see the two major chips. The larger one in the centre is the main GPU and the one to the right of that is the troublesome AGP -> PCI Express bridge (the 6600 is a native PCI Express chip). You can also see the 4 10mm x 10mm memory chips surrounding the GPU.

First of all I installed the heat sink rotated 180 degrees. Because the holes are symmetrical this is not a problem and it fitted nicely.

The next problem was that if you insert the card the larger fins actually touch the adjoining PCI slot connector. To fix this I trimmed some fins and bent some of them around out of the way. Copper is very flexible and easy to cut, any loss of surface area this small would be irrelevant to the heat sink’s heat dissipating capacity.

The memory chips also needed to have their heat sinks installed. The guide recommends that you place one on top of each of the 4 memory chips and one underneath. Looking under the board I don’t see how this is possible (you just stick them on top of the component ‘legs’??).

Not only wouldn’t this hold very well, the tiny surface area and fact that the heat would have to travel through the board means that the cooling effect would be minimal. As a result of that I decided to stick 2 on each memory chip. Each sink only has a 50% coverage of the surface area, however they are located where they will still get a decent air flow. Note also that I had to bend and trim the heat sinks a bit so that they fitted within the bounds of the card (yes my case is that close to my cards). Of course rotating the heat sink around is going to cause more problems than just hitting the PCI slot! I also purchased a heat sink from Zalman for the AGP bridge chip (ZM-VHS1 HSI Heat sink).

The large fins now cover the area directly above the bridge chip, so this makes installing the heat sink a little more difficult. As the picture shows the fins actually hit the heat sink and prevent it from being moved in and underneath the fan. Some more modding was required!

Using a grinder and some careful testing I ground down the heat sink so that it could fit in underneath the fan. This reduces the surface area of the heat sink, however the heat sink is located almost directly underneath the fan (as opposed to well away from the fan where Zalman intended), meaning the increased airflow should compensate for the loss of area.

The bridge chip is also a little difficult to install underneath the heat sink, however it is certainly manageable. When the bridge chip is then installed with some thermal paste you can see just how nicely it fits underneath the fan. Needless to say I was pretty happy with this!

The card could now be installed in the case. You can see the fins down near the slot are a little bent, but that the fan now fits nicely within the bounds of the PCB.

I fired up my computer and it looks pretty nice I have to say! The fan is also dead silent when connected to the 5V input of the included power cable. I am very happy with the result and I think it complements my Golden Orb II quite nicely. Sorry about the blurriness of the photo, hard to hold time delays still!

While I was in the modding mood I decided to brand my white box PC with a Dell badge that I ‘found’ elsewhere. It fits absolutely perfectly over the Overture on button and has a nice blue glow coming through around it. I am not a Dell fanboi, it just matches my monitor so nicely now!

I hope that this simple guide gives people with a similar problem to myself some help. There is no other 3rd party cooling system available for the 6600GT AGP card, so this was my only choice. I think this guide will be particularly useful to people setting up media centres which often have pretty tight space restrictions like my case.

I would recommend this card as it is pretty cheap these days and still offers very solid performance. Please be really careful not to damage your GFX card in the process of attempting this mod and also make sure you clean any metal filings off your heat sinks etc. before putting them anywhere near computer equipment! Good luck with it!