I installed the Legacy nodes in their new rack last night and today was spent preparing the shop for the new mains leg to be installed.
Rearranging was the hard part. It turns out running the new cord under the house is going to be easy. While I had been planning to tie a string to a rock and skip it under the house to the other side, I found I already have a string in place in the form of the very extension cord supplying the shop right now. I connect a lesser cord to that and pull it under the house to use for pulling both supply cords back for semi-permanent installation.
Saturday, October 31, 2009
Thursday, October 29, 2009
Much has been done.
The spider problem has been gone for months now. That's no big deal. What is, however, is the fact that Fragwell Labs, while once intended to receive a makeover, is going to be a barn for my server farm during the cold months.
With that in mind, I need more power. While the lab already has a 20A circuit, it has suffered some regrettable limitations due to the way I set the wiring up. When we moved in, the lab was wired with power via a 20A drop from the breaker panel to a piece of Romex and then to a junction box with a GFCI inside. From there, a grounded 12AWG extension cord was plugged into the box, run under the house and out to the lab. That was fine, for a bit, until I was heating the lab with a bunch of idling computers last winter to try to keep it at a temperature where I could work comfortably and ended up blowing the "15A" breaker on my ten year old power strip. This tripped the GFCI so I had to pull the skirting and push the reset button. Not easy with a foot of snow in my way. I have better power strips now, but there's another problem.
Directly inside the shop, the extension cord has a three-way splitter plugged into it. This splitter has a 13A breaker on it and it's very touchy, especially when the ambient temperature is high as is common during Utah summers. Running my lab computer, lighting, portable A/C unit and laser printer is a recipe for a breaker trip. Not fair.
But there's a way around it. This weekend, I run a second 115VAC line off the other side of the phase. While I can't go directly from a single outlet to a power strip and use all 20A since code is such that no single outlet can use more than 80% of the ampacity of a circuit, I can get a tw0-way splitter and use the full ampacity of both lines. While that circuit won't be enough during the summer months where cooling is a necessity, it will be more than enough for winter, even with the server farm and laser printer, assuming I use the printer all that much.
The compute cluster is meant to be powered by a 230VAC 15A circuit when fully loaded. With the 80% rule in effect, that's 12A peak they plan for. I still have 8A left on each rail, which would be enough for the AC unit and my laser printer to coexist.
I have nine compute nodes. Eight work. The cluster power rating looks at ten nodes.
With that in mind, I need more power. While the lab already has a 20A circuit, it has suffered some regrettable limitations due to the way I set the wiring up. When we moved in, the lab was wired with power via a 20A drop from the breaker panel to a piece of Romex and then to a junction box with a GFCI inside. From there, a grounded 12AWG extension cord was plugged into the box, run under the house and out to the lab. That was fine, for a bit, until I was heating the lab with a bunch of idling computers last winter to try to keep it at a temperature where I could work comfortably and ended up blowing the "15A" breaker on my ten year old power strip. This tripped the GFCI so I had to pull the skirting and push the reset button. Not easy with a foot of snow in my way. I have better power strips now, but there's another problem.
Directly inside the shop, the extension cord has a three-way splitter plugged into it. This splitter has a 13A breaker on it and it's very touchy, especially when the ambient temperature is high as is common during Utah summers. Running my lab computer, lighting, portable A/C unit and laser printer is a recipe for a breaker trip. Not fair.
But there's a way around it. This weekend, I run a second 115VAC line off the other side of the phase. While I can't go directly from a single outlet to a power strip and use all 20A since code is such that no single outlet can use more than 80% of the ampacity of a circuit, I can get a tw0-way splitter and use the full ampacity of both lines. While that circuit won't be enough during the summer months where cooling is a necessity, it will be more than enough for winter, even with the server farm and laser printer, assuming I use the printer all that much.
The compute cluster is meant to be powered by a 230VAC 15A circuit when fully loaded. With the 80% rule in effect, that's 12A peak they plan for. I still have 8A left on each rail, which would be enough for the AC unit and my laser printer to coexist.
I have nine compute nodes. Eight work. The cluster power rating looks at ten nodes.
Tuesday, October 20, 2009
Shake It Don't Break It
I took delivery of a "new" laptop a short time ago and just recently gave it a much-needed hard disk upgrade from 100GB to 500GB. The machine is still a current model in Dell's lineup and it rather powerful. Dual cores and plenty of RAM. Just what I needed.
As I wrote before, the Dual P4 Xeon 3.2GHz system scored a Yafaray test render time of 10.5 minutes with four threads. I knew I'd have to try the same on the new laptop's T7200 Core 2 Duo. The results surprised me. It was a tie.
If you look at clock speeds, it doesn't make sense. But clock speeds don't mean what they used to back in the Pentium 4 days. The Core and Core 2 CPUs architectures are based on the Pentium M, which was based on the Pentium III Tualatin. The Pentium 4 was abandoned after the Prescott core turned out to be a space heater and little else.
I mention the lineages because it predicted the future. I had a Pentium M laptop of 1.3GHz speed that would outpace my 2.4GHz Pentium 4 laptop in everything but floating point math. That's almost double the operations per clock tick right there.
The Core Duo and Core 2 Duo don't support Hyperthreading. That was brought back with the Core i7. The 15 minute time on the 3.2GHz CPUs when run without Hyperthreading explains the tie. The 2GHz Core 2 Duo in my laptop should be equivalent to a pair of 4GHz P4 Xeons when memory bus fighting is taken into consideration since the Dual Xeon rig is dual channel PC2100 DDR while the C2D is Dual Channel PC2-5300.
At 4GHz, the Xeon rig would complete the render in twelve minutes with Hyperthreading disabled. That's within a minute and a half. Something else is at work here and I don't know what it is, but it says a lot about the improvements made with the Core Architecture.
On top of all that, the Dual Xeon rig cranks out heat like mad, nearly maxing a 380W power supply, while the laptop is hard-limited by an AC adapter brick with a 120W output. And I have twice the RAM in the laptop.
I have yet to try any of my cluster computers. I wonder how they will do.
As I wrote before, the Dual P4 Xeon 3.2GHz system scored a Yafaray test render time of 10.5 minutes with four threads. I knew I'd have to try the same on the new laptop's T7200 Core 2 Duo. The results surprised me. It was a tie.
If you look at clock speeds, it doesn't make sense. But clock speeds don't mean what they used to back in the Pentium 4 days. The Core and Core 2 CPUs architectures are based on the Pentium M, which was based on the Pentium III Tualatin. The Pentium 4 was abandoned after the Prescott core turned out to be a space heater and little else.
I mention the lineages because it predicted the future. I had a Pentium M laptop of 1.3GHz speed that would outpace my 2.4GHz Pentium 4 laptop in everything but floating point math. That's almost double the operations per clock tick right there.
The Core Duo and Core 2 Duo don't support Hyperthreading. That was brought back with the Core i7. The 15 minute time on the 3.2GHz CPUs when run without Hyperthreading explains the tie. The 2GHz Core 2 Duo in my laptop should be equivalent to a pair of 4GHz P4 Xeons when memory bus fighting is taken into consideration since the Dual Xeon rig is dual channel PC2100 DDR while the C2D is Dual Channel PC2-5300.
At 4GHz, the Xeon rig would complete the render in twelve minutes with Hyperthreading disabled. That's within a minute and a half. Something else is at work here and I don't know what it is, but it says a lot about the improvements made with the Core Architecture.
On top of all that, the Dual Xeon rig cranks out heat like mad, nearly maxing a 380W power supply, while the laptop is hard-limited by an AC adapter brick with a 120W output. And I have twice the RAM in the laptop.
I have yet to try any of my cluster computers. I wonder how they will do.
Sunday, October 11, 2009
Life, the Universe and Everything
A while back, I hacked together a 32" LCD TV from a wide array of parts. It turned out to be too small. Funny, that, since it's got a widescreen picture the likes of which only the largest old-school tube TVs could approach. I guess it's the HD thing. More detail so you need to be able to see it.
Last Saturday, I got another LCD set. This one had real issues. While it would work correctly with no input signal, any kind of signal would cause it to scramble horribly. I had seen this before. It was power supply related. Replacing the capacitors in the main logic power supply allowed the display to work correctly. Basically, $5 got me a new display. And I grabbed it at the perfect time. The next morning, the ground was wet from fresh rain.
This new display didn't come with a stand of any kind and I knew I wouldn't be able to make one. It was time for a major purchase so I went to Walmart and found a very nice steel and glass LCD stand that would take advantage of the wall mount holes. It only cost me about $170 after taxes so I couldn't complain at all, having ended up with a 42" display for $175.
It doesn't have a tuner so I'll have to buy one. What is does have is a native 1080p image, HDMI input, two component inputs, VGA and the old-school S-Video and composite. It will take a 1920x1200 signal on the VGA input, but it looks noisy. I'll have to try a better cable. I do know that the VGA cable for my XBOX 360 will show a proper display, although it's cut off at the left.
Last Saturday, I got another LCD set. This one had real issues. While it would work correctly with no input signal, any kind of signal would cause it to scramble horribly. I had seen this before. It was power supply related. Replacing the capacitors in the main logic power supply allowed the display to work correctly. Basically, $5 got me a new display. And I grabbed it at the perfect time. The next morning, the ground was wet from fresh rain.
This new display didn't come with a stand of any kind and I knew I wouldn't be able to make one. It was time for a major purchase so I went to Walmart and found a very nice steel and glass LCD stand that would take advantage of the wall mount holes. It only cost me about $170 after taxes so I couldn't complain at all, having ended up with a 42" display for $175.
It doesn't have a tuner so I'll have to buy one. What is does have is a native 1080p image, HDMI input, two component inputs, VGA and the old-school S-Video and composite. It will take a 1920x1200 signal on the VGA input, but it looks noisy. I'll have to try a better cable. I do know that the VGA cable for my XBOX 360 will show a proper display, although it's cut off at the left.
Subscribe to:
Comments (Atom)