Second Gen Cell info
- Thread starter Mikage
- Start date
OK TV guide seems a bit verbose example, it's useful when Marty junior previews his personal pr0n vid library in 2020Blazkowicz_ said:The similarity is you still see 25 (it was maybe 20, or 30, whatever) moving images at 25fps (no 30fps NTSC junk), and I was replying about Tacitblue's idea, which is good but has been implented more simply. could you hear anything if you heard all that crap at the same time? It's already painful when Marty junior watches six channels at the same time in 2015
Oops you're right, it was late last night when I posted and I read DTV there instead of SDTVERP said:
Anyway still, as noted below, shouldn't that be 720x480/576? (That's what DVD movies are encoded at anyway - heck even our PS2 FMVs are 720xsomething.Well it depends on their implementation - but that may not make much difference for performance.VNZ said:
On PS2, the IPU is rated at 768cycles per Macroblock for decoding, and realworld measured performance precisely matches that (~95MPix/sec) regardless of what bitrate I tried (although that may also be because it's fixed-hw solution).
I've tested I-Frames encoded as large as 200KB for 640x480, and as low as 20KB(which'd make 10x bitrate difference in I-Frame only stream) for the same dimensions and I'd get same speed in both cases.
Admitedly, and I've only run speed tests for my area of interest - I-Frame only streams, as I needed explicit frames for in-game texture compression, but those kind of streams are the worst case scenario for bitrate anyhow
ERP beat me to it. I was going to comment that the most interesting thing about the demo was that the system can sustain that kind of data input - provided as ERP pointed out - the data is not interleaved. I think we can safely assume it is not - else this demo is useless.
The latest top-end CPUs *may* have the raw performance to do something similar, but I *suspect* they may choke long before they reach that number...
After all it has been pointed out again and again that for CPUs today, it's not so much how fast it can run, but rather how quickly they can be fed.
The latest top-end CPUs *may* have the raw performance to do something similar, but I *suspect* they may choke long before they reach that number...
After all it has been pointed out again and again that for CPUs today, it's not so much how fast it can run, but rather how quickly they can be fed.
It's more sustaining 48 distinct streams that's difficult.
Seek times on DVD are hideous, better on HD obviously, but it's still not simple. Just seeking between the streams eats up the bandwidth. Using a large enough RAM buffer would let you do it, but I suspect that would be a lot of RAM for the 48 buffers.
That's why I qualified unless they are interleaved in some way. At that point you require no seeks and it's all down to bandwidth.
If I were a betting man I'd guess that the streams were in fact interleaved, but thats because the point of the demo seems to be the decoding speed, not the streaming capability.
Seek times on DVD are hideous, better on HD obviously, but it's still not simple. Just seeking between the streams eats up the bandwidth. Using a large enough RAM buffer would let you do it, but I suspect that would be a lot of RAM for the 48 buffers.
That's why I qualified unless they are interleaved in some way. At that point you require no seeks and it's all down to bandwidth.
If I were a betting man I'd guess that the streams were in fact interleaved, but thats because the point of the demo seems to be the decoding speed, not the streaming capability.
ERP said:
Not so sure I'd agree. At DVD bitrate, we're talking 1 MB/s per stream.
48 streams = 48 MB/s.
So a GB of RAM is enough to buffer 21 seconds. (I don't think you'll need anywhere near that much buffer though.)
If you read the streams in 1MB chunks from 2 hard disks, you should be able to meet the I/O throughput requirement pretty easily, even with the seeks -- 9ms seek time means you can do about 100 seeks a second... so with 48 streams/1MB chunks you'd spend half your time seeking, and the other half of the time you can be reading data.
Not that this level of HW is practical to ship in a cheap consumer device, but for the purposes of the demo, it seems pretty straightforward to accomplish -- we haven't even gotten into anything semi-exotic (like FibreChannel or RAID) in terms of I/O hardware.
V3 said:
Well that's nearly a 50% increase in bandwidth for a product supposedly shipping ~ 6 months later in 2006. Bandwidth growth has always been slower than other performance metrics. But that % increase doesn't sound too unreasobale IMHO.
V3 said:
The GPU has nothing to do with PC limitations like PCI-e. It only shares technology from NV's next gen GPU not bandwidth limitations of PC's.
Afterall what consoles lack in total RAM, they always make it up with bandwidth...so in 2006, I still expect that to be true...
Also don't assume work started on the GPU 'only' last December 2004...
PC-Engine said:
Cross platform comparisons like that are more or less meaningless wouldn't you say? The overheads of a pc operating system environment aka software interrupts from background processes or other programs means you'll never really get a fair or decent comparision. Hell if you ran SuperPi on the Cell it would undoubtedly monster anything out there but again apples and oranges. If you tried it though you'd have to run tests with both software or hardware decoding.
I think its a bit easy to see the kind of power that a streaming optimised processor like Cell has, when bottom line that's its strength, it was designed for that. a Pentium D with hyperthreading (isn't that the EE?) is not only prohibitably expensive (the processor all by itself costs more than a couple of PS3's expected cost) just has more burden from the bloatware it runs to keep us happy at work......or frustrated, depends on the time of day and what your deadlines are.
Tacitblue said:PC-Engine said:
Cross platform comparisons like that are more or less meaningless wouldn't you say? The overheads of a pc operating system environment aka software interrupts from background processes or other programs means you'll never really get a fair or decent comparision. Hell if you ran SuperPi on the Cell it would undoubtedly monster anything out there but again apples and oranges. If you tried it though you'd have to run tests with both software or hardware decoding.
I think its a bit easy to see the kind of power that a streaming optimised processor like Cell has, when bottom line that's its strength, it was designed for that. a Pentium D with hyperthreading (isn't that the EE?) is not only prohibitably expensive (the processor all by itself costs more than a couple of PS3's expected cost) just has more burden from the bloatware it runs to keep us happy at work......or frustrated, depends on the time of day and what your deadlines are.
The EE has 2 MB cache on the chip. Standard P4s (above a certain revision, dunno which one) have hyperthreading as well.
xbdestroya said:
Oh so when i go to the shop to buy one i have to tell them "Can i please have a Pentium D dual core Extreme Edition dual core with hyperthreading please? Dual core. Extreme. Please.".
Talk about a mouthfull.
marconelly!
Veteran
I am not aware something like he suggested was ever done in any game. Not that I think it would be particularily useful anyways.
Similar threads
- Locked