one said:128 bit bus enables them to reduce 512Mbit chips * 8 to 1Gbit chips * 4 when cheaper 1Gbit chips get available a few years later. Then they can retool the motherboard more simpler, and cheaper.
BTW, 256 GB/s memory bandwidth to EDRAM means it uses 4096bit bus running at 500Mhz if the GPU is a single chip.
Titanio said:Getting back to the 9bn dot products per cycle figure, does this reveal any "special sauce" on the vmx units, or is that what you'd expect to get out of 3 standard vmx units at 3.2Ghz (theoretically, of course)?
If a dot product is 4 multiplies and 3 additions, is that 7 floating point ops? So taking nAo's figure of 76.8Gflops, 76.8/7 = ~11bn dot products? Or if you round up to 8 flops (which you probably would do..can't do much with that extra spare flop per cycle per core..at least not another dot product), 76.8/8 = ~9bn? Is that were it comes from, or am I totally off track?
one said:128 bit bus enables them to reduce 512Mbit chips * 8 to 1Gbit chips * 4 when cheaper 1Gbit chips get available a few years later. Then they can retool the motherboard more simpler, and cheaper.Jaws said:barnak said:Sorry I havnet been keeping up with the specs as much as I like, so MS went ahead and went for the 512 ram?
If this spec is true then yes, 512MB is good.
Memory
-512 MB GDDR3 RAM
- 700 MNz DDR
Memory Bandwidth
- 22.4 GB/s memory interface bus bandwidth
- 256 GB/s memory bandwidth to EDRAM
- 21.6 GB/s frontside bus
700 MHz DDR (1400 MHz effective)
22.4 GB/s --> 128 bit bus
I was hoping it would be 256 bit ---> 44.8 GB/s
BTW, 256 GB/s memory bandwidth to EDRAM means it uses 4096bit bus running at 500Mhz if the GPU is a single chip.
BTW, 256 GB/s memory bandwidth to EDRAM means it uses 4096bit bus running at 500Mhz if the GPU is a single chip.
I'm pretty sure that 256 GB/s is 'effective' bandwidth and not 'true' bandwith as mentioned in the original 'leak'.
Eight pixels (where each pixel is color plus z = 8 bytes) can be sent to the EDRAM every GPU clock cycle, for an EDRAM write bandwidth of 32 GB/sec. Each of these pixels can be expanded through multisampling to 4 samples, for up to 32 multisampled pixel samples per clock cycle. With alpha blending, z-test, and z-write enabled, this is equivalent to having 256 GB/sec of effective bandwidth! The important thing is that frame buffer bandwidth will never slow down the Xenon GPU.
This VPU is being designed with the latest technologies in mind, such as GDDR2 SDRAM memory provided by Samsung running at 1600 MHz. A 128-bit configuration is capable of providing up to 25.6 GB/s peak bandwidth, while its 256-bit mode brings up to a shocking 51.2GB/s peak bandwidth!!! Samsung’s GDDR2 256-megabit memory will enable graphics memory cards of 512 MB, althought it is impossible to confirm if the Xbox 2 will feature such amount of system memory.
most of the mid to low end boards still have only a 128-bit bus .a 128-bit bus is kinda disappointing since R300 in Radeon 9700 has had 256-bit since 2002. (as did other new G|VPUs of 2002)
R300 didn't have a 48 (256) GB/s extra bus dedicated to render targets.Megadrive1988 said:I'm still holding out hope that with a 256-bit bus, the main memory/system bandwidth will be double or roughly 2.5x greater than the ~22 GB/sec which is because of a 128-bit bus.
Megadrive1988 said:but all *highend* ATI VPUs and Nvidia GPUs have 256-bit bus, or am I mistaken ?
nAo said:R300 didn't have a 48 (256) GB/s extra bus dedicated to render targets.Megadrive1988 said:I'm still holding out hope that with a 256-bit bus, the main memory/system bandwidth will be double or roughly 2.5x greater than the ~22 GB/sec which is because of a 128-bit bus.
therealskywolf said:Anyway guys, are those specs like......Good enough for next gen? Do they this years high end pcs or something like that?
vliw said:therealskywolf said:Anyway guys, are those specs like......Good enough for next gen? Do they this years high end pcs or something like that?
It's a Low Cost Next gen, imho....Ps3 will be another world.
In his mind where a 500$ system is low cost and a 500$ six months later will be a high cost system. ONly really because it says sony on it and not microsoft.How do you fathom "low cost"?
nAo said:My calculations assumed 8 ops per clock, I believe VMX units can do one fmadd per cycle.