E Pluribus Unum means "One from Many" or "Out of Many, One" in Latin. There's nothing religious about it. It is a part of the Great Seal of the USA and is unrelated to currency, though it is on the back of the $1 bill. Check out www.greatseal.com for more info.
So are you saying that a cluster isn't a supercomputer? Check out the Top 500 list. Notice how many Crays are in the top 15? ASCI White is a cluster and is definitely a supercomputer. Clusters are the future of supercomputing.
It looks like they're feeding the IF signal into an ADC, so you'd still need the part of the receiver that takes the RF down to IF. If I remember how FM works, you take the base signal and modulate it up to 10.7Mhz (IF). In the next stage you take the IF and modulate it up to whatever the final channel is, say 92.5Mhz for broadcast. What you're saying would only work if you had an ADC that could digitize the RF signal. This is probably possible for FM radio (~100Mhz), but would get expensive if you wanted to do it for cell phones (~1-2Ghz). DSOs that handle 5Ghz signals cost as much as $50k so those high speed ADC must be pretty pricey.
It would work if we had projectors with some kind of optical keystone correction. The problem is that we would need to project rectangular (as opposed to trapezoidal) images that overlap on the image plane but are coming from two different places. We'd have to do some kind of over-under mirror configuration. It's not impossible; it would just be pretty complicated mechanically. It would also be kind of tough to maintain the polarization. Who knows, we may be called on to do this some day.
It is an issue of bandwidth. There's no way we could feed that many pixels or vertices at an interactive rate with only 4 machines. If there were something like a 64bit 66Mhz PCI video card, it would help, but doing it like this is the way to get the best performance. The Myrinet network is approximately the same bandwith as regular PCI, so it's well balanced. Also, the resolution per screen *is* 1024x768. I'm not sure where people are getting the idea that it's less.
The screen is 9'x12'; the *image* on the screen is about 8'x8'. We're still using an old screen from our original 2x2 CRT based IWall. Each tile is 1024x768 and is about 25" wide. 1024 dots / 25" = 40.96 dpi. If you look at the image referenced in the news release or the ones I posted earlier, you can see what I mean.
How much would you pay to play quake on a display with head tracking, an 8'x6' screen, and stereo video? Seriously. See www.visbox.com and let me know.
One of the drivers for this wall is a scientist we have here that has 8kx8k images of radio astronomy data. He wants to be able to compare different regions by eye. Panning and zooming on a small monitor doesn't cut it. If you're going to image or calculate high resolution data, you don't want to throw it out when viewing.
Actually it is 4 projectors wide and 5 high. The screen is 9'x12', but the image is about 8'x8'. Each tile is about 2 ft. wide. You're right that the resolution in dpi is less than a monitor, but there are no monitors that can seamlessly display 4096x3840 or 8192x3840. The idea is that you can look at a large dataset and see the whole thing without panning and zooming. It will definitely be used for "real work".
The idea is to have a seamless high resolution display. The only way to do this is with projectors. You're right that a single graphics card can produce a QXGA signal, but there is only one projector that I know of that can display that many pixels (JVC DLA-QX1) and it will cost around $200k. There isn't any way to drive 4 projectors from a single output (that I know of). If you know of something, let me know.
We're using WireGL, which is a distributed OpenGL implementation out of Stanford. It has some code in it to synchronize the displays. Since we're using Myrinet (gigabit network with low latency), the pipes sync up pretty well. As for the stereo, we don't plan to do either. There are no projectors of this cost and form factor that are capable of active stereo and the complexity of mounting the projectors for passive stereo are too horrible to contemplate!:) We've got the other 20 projectors and plan to build another identical block next to the one we've already got.
Slashdot Mirror
E Pluribus Unum means "One from Many" or "Out of Many, One" in Latin. There's nothing religious about it. It is a part of the Great Seal of the USA and is unrelated to currency, though it is on the back of the $1 bill.
Check out www.greatseal.com for more info.
The 7000's are 64bit capable, but they don't require it.
So are you saying that a cluster isn't a supercomputer? Check out the Top 500 list. Notice how many Crays are in the top 15? ASCI White is a cluster and is definitely a supercomputer. Clusters are the future of supercomputing.
It looks like they're feeding the IF signal into an ADC, so you'd still need the part of the receiver that takes the RF down to IF. If I remember how FM works, you take the base signal and modulate it up to 10.7Mhz (IF). In the next stage you take the IF and modulate it up to whatever the final channel is, say 92.5Mhz for broadcast. What you're saying would only work if you had an ADC that could digitize the RF signal. This is probably possible for FM radio (~100Mhz), but would get expensive if you wanted to do it for cell phones (~1-2Ghz). DSOs that handle 5Ghz signals cost as much as $50k so those high speed ADC must be pretty pricey.
VREX builds projectors that do what you are describing.
We're using a cluster very similar to this to run our wall.
It would work if we had projectors with some kind of optical keystone correction. The problem is that we would need to project rectangular (as opposed to trapezoidal) images that overlap on the image plane but are coming from two different places. We'd have to do some kind of over-under mirror configuration. It's not impossible; it would just be pretty complicated mechanically. It would also be kind of tough to maintain the polarization. Who knows, we may be called on to do this some day.
It is an issue of bandwidth. There's no way we could feed that many pixels or vertices at an interactive rate with only 4 machines. If there were something like a 64bit 66Mhz PCI video card, it would help, but doing it like this is the way to get the best performance. The Myrinet network is approximately the same bandwith as regular PCI, so it's well balanced. Also, the resolution per screen *is* 1024x768. I'm not sure where people are getting the idea that it's less.
The screen is 9'x12'; the *image* on the screen is about 8'x8'. We're still using an old screen from our original 2x2 CRT based IWall. Each tile is 1024x768 and is about 25" wide. 1024 dots / 25" = 40.96 dpi. If you look at the image referenced in the news release or the ones I posted earlier, you can see what I mean.
How much would you pay to play quake on a display with head tracking, an 8'x6' screen, and stereo video? Seriously. See www.visbox.com and let me know.
The resolution is more like 40 dpi (1024/25). The cumulative light output of the full size wall will be around 40,000 lumens.
It is transparent to OpenGL apps. We're using WireGL out of Stanford which is a distributed OpenGL implementation.
One of the drivers for this wall is a scientist we have here that has 8kx8k images of radio astronomy data. He wants to be able to compare different regions by eye. Panning and zooming on a small monitor doesn't cut it. If you're going to image or calculate high resolution data, you don't want to throw it out when viewing.
Actually it is 4 projectors wide and 5 high. The screen is 9'x12', but the image is about 8'x8'. Each tile is about 2 ft. wide. You're right that the resolution in dpi is less than a monitor, but there are no monitors that can seamlessly display 4096x3840 or 8192x3840. The idea is that you can look at a large dataset and see the whole thing without panning and zooming. It will definitely be used for "real work".
The idea is to have a seamless high resolution display. The only way to do this is with projectors. You're right that a single graphics card can produce a QXGA signal, but there is only one projector that I know of that can display that many pixels (JVC DLA-QX1) and it will cost around $200k. There isn't any way to drive 4 projectors from a single output (that I know of). If you know of something, let me know.
http://brighton.ncsa.uiuc.edu/~prajlich/wall/newpi cs.html
We're using WireGL, which is a distributed OpenGL implementation out of Stanford. It has some code in it to synchronize the displays. Since we're using Myrinet (gigabit network with low latency), the pipes sync up pretty well. As for the stereo, we don't plan to do either. There are no projectors of this cost and form factor that are capable of active stereo and the complexity of mounting the projectors for passive stereo are too horrible to contemplate! :) We've got the other 20 projectors and plan to build another identical block next to the one we've already got.
That what happens when you get computer guys trying to cut, drill, and tap aluminum extrusion!