It isn't about the price of Opteron processors for 4 and 8-way systems, but rather the performance. We have a bunch of Dell hardware, and the 6650 (4-way P4 Xeon MP) wasn't much faster than the 2650 (2-way P4 Xeon). The problem is the shared frontside bus in the Intel architecture is a bottleneck, which Hypertransport solves.
Still, unless you need that many CPUs in one box or you need tons of RAM (our Opteron cluster has 16-32G per node), it is cheaper to buy multiple smaller boxes. However, if you need the bigger SMP boxes you can get good performance from them with Opteron and you can't with Intel without a very fancy external bus system.
If the administrative access that the voting official has could ever be used to corrupt the election process, then it is already broken regardless of whether that official gives access to a third party.
Besides, there should some verification test that can be run independently on the machine to verify it is working as intended, which would not require $40k and a plane trip to use. Clearly, as stated in the article, Diebold is wanting to make this person an example so no other election official will let anyone else take a look at the machine.
Actually I think we are not too far from building a functional AI. However, I don't think it will be by the current approach but rather by direct hardware simulation of the brain. We are gathering knowledge of how the internals of memory and brain functions at an increasing rate, and the computer processing power is advancing to the point it will be feasible to simulate each neuron at real-time speeds.
There are some Mexican railways that still use electric locomotives because a locomotive going down the mountain actually supplies a large portion of the electricity required by a different locomotive going up the mountain at the same time. I don't recall the dates, but I am pretty sure it predated anything in this patent. They are used as helpers added onto a set of diesel-electric locomotives when it approaches the high-grade track, and then are removed to wait for the next train.
Where exactly do you intend to get non-MPEG content? Satellite is delivered via MPEG2, digital cable is MPEG2, HTDV is MPEG2, and DVD is MPEG2.
Since I am quite certain you will agree HDTV is better quality than any NTSC analog signal you capture from analog cable or antennae, then perhaps the issue isn't using MPEG2 but rather what bitrate you use.
If you are recording more than one channel at a time, you will have difficulty building a disk array capable of keeping up with uncompressed D1 video, and you can forget uncompressed HD content with off-the-shelf hardware.
In my experience, MPEG2 at 6Mbps VBR has no noticeable loss of quality over any NTSC source. You can frequently get away with much less depending on the source material. So, I think the answer is to make sure that you choose the appropriate bitrate for your source material and your quality requirements and have a hardware encoder.
I found both the original drive (some Quantum Fireball model) and the larger drive (an IBM 120G drive, chosen for heat dissipation and noise level) to be loud enough to be noticeable. I eventually got used to it, but for a while it was constantly waking me up.
If you already have the computer equipment, the cost for adding an encoder card and a MediaMVP client is about $200 -- less than the cost of a moderately-sized Tivo/ReplayTV unit with the lifetime subscription.
HDTV changes the cost equation further -- right now, the only HD DVRs I know of are $900. I can add a pcHDTV HD-3000 card for $190, and a Roku box for each client for $299 (note I don't have one and don't know for sure it will be able to be a MythTV client, but it should just be a small matter of programming:). With HD, I can take the recorded HD programs and transcode them into widescreen SD for my widescreen NTSC TV as well. Add in the concern about the broadcast flag, removing the ability to skip ads, and forcing you to see adds, it seems like open source only looks better and better. Granted, if you want HD from satellite the only solution currently is something more expensive, like from http://169time.com/.
I am quite happy with my ReplayTV units, but I can see the day where they won't be that useful, and I want to start building my next-generation system.
I own two ReplayTV 2xxx units (one in the media room and one in the bedroom) upgraded with larger disks, yet I am building a MythTV system. One big reason is to have better utilization of disk space and tuners -- if I have recorded a program upstairs, I have to watch it there (the ReplayTV 5xxx series has ways around this but my understanding from friends with that setup is there are still limitations); if I have free space upstairs but not downstairs I can't record it downstairs. Another reason is disk space -- my main server has 2T of RAID5 disk space, which I can share between recorded video and other needs. Also, with it on the server it is easy to archive anything I have recorded to DVD if I like. Finally, I can use cheap MediaMVP units (or eventually Roku for HDTV) as the clients -- no bulky, loud computer by the TV (or a ReplayTV that spins up the disk in the night in the bedroom) and for $80 I can add another client on any TV in the house.
Granted, this is not ready for the average user just yet. MythTV works pretty well, but there are still rough edges. Using the MediaMVP as a client is still rather limited, but it works. Getting the kernel drivers for the PVR250 cards in a stock RedHat kernel (RHEL4 beta currently) took some effort as well -- I have not tried them in other kernels/distributions.
Using the PVR250 hardware encoders I get better picture quality at 4Mbps than the ReplayTVs do at 6Mbps. MythTV's scheduling is already much smarter than ReplayTV and still I think there is room for improvement. Program database access is faster as well.
So yes, right now it is more expensive and doesn't work as well as the dedicated devices. However, I don't think it will remain that way and will eventually be even better. I also get the source and can hack it to be any way I like:).
Slashdot Mirror
Still, unless you need that many CPUs in one box or you need tons of RAM (our Opteron cluster has 16-32G per node), it is cheaper to buy multiple smaller boxes. However, if you need the bigger SMP boxes you can get good performance from them with Opteron and you can't with Intel without a very fancy external bus system.
Besides, there should some verification test that can be run independently on the machine to verify it is working as intended, which would not require $40k and a plane trip to use. Clearly, as stated in the article, Diebold is wanting to make this person an example so no other election official will let anyone else take a look at the machine.
Actually I think we are not too far from building a functional AI. However, I don't think it will be by the current approach but rather by direct hardware simulation of the brain. We are gathering knowledge of how the internals of memory and brain functions at an increasing rate, and the computer processing power is advancing to the point it will be feasible to simulate each neuron at real-time speeds.
There are some Mexican railways that still use electric locomotives because a locomotive going down the mountain actually supplies a large portion of the electricity required by a different locomotive going up the mountain at the same time. I don't recall the dates, but I am pretty sure it predated anything in this patent. They are used as helpers added onto a set of diesel-electric locomotives when it approaches the high-grade track, and then are removed to wait for the next train.
Since I am quite certain you will agree HDTV is better quality than any NTSC analog signal you capture from analog cable or antennae, then perhaps the issue isn't using MPEG2 but rather what bitrate you use.
If you are recording more than one channel at a time, you will have difficulty building a disk array capable of keeping up with uncompressed D1 video, and you can forget uncompressed HD content with off-the-shelf hardware.
In my experience, MPEG2 at 6Mbps VBR has no noticeable loss of quality over any NTSC source. You can frequently get away with much less depending on the source material. So, I think the answer is to make sure that you choose the appropriate bitrate for your source material and your quality requirements and have a hardware encoder.
If you already have the computer equipment, the cost for adding an encoder card and a MediaMVP client is about $200 -- less than the cost of a moderately-sized Tivo/ReplayTV unit with the lifetime subscription.
HDTV changes the cost equation further -- right now, the only HD DVRs I know of are $900. I can add a pcHDTV HD-3000 card for $190, and a Roku box for each client for $299 (note I don't have one and don't know for sure it will be able to be a MythTV client, but it should just be a small matter of programming :). With HD, I can take the recorded HD programs and transcode them into widescreen SD for my widescreen NTSC TV as well. Add in the concern about the broadcast flag, removing the ability to skip ads, and forcing you to see adds, it seems like open source only looks better and better. Granted, if you want HD from satellite the only solution currently is something more expensive, like from http://169time.com/.
I am quite happy with my ReplayTV units, but I can see the day where they won't be that useful, and I want to start building my next-generation system.
Granted, this is not ready for the average user just yet. MythTV works pretty well, but there are still rough edges. Using the MediaMVP as a client is still rather limited, but it works. Getting the kernel drivers for the PVR250 cards in a stock RedHat kernel (RHEL4 beta currently) took some effort as well -- I have not tried them in other kernels/distributions.
Using the PVR250 hardware encoders I get better picture quality at 4Mbps than the ReplayTVs do at 6Mbps. MythTV's scheduling is already much smarter than ReplayTV and still I think there is room for improvement. Program database access is faster as well.
So yes, right now it is more expensive and doesn't work as well as the dedicated devices. However, I don't think it will remain that way and will eventually be even better. I also get the source and can hack it to be any way I like :).