More Xenon rumors

But your counting memory bandwidth wrong,

First there is eDRAM bandwidth,
then the L2 bandwidth
then the system memory bandwidth.

That means that 26 Gb/s has ALOT less to do than a topend PC video cards bandwidth.
 
Nite_Hawk said:
Sure, lower specs. When current top end videocards are pushing a memory throughput of ~36GB/s, and a next generation console that won't be out for atleast a year is talking about ~26GB/s, (less than 3/4th the throughput.) it makes sense to say it's got lower specs.

I don't agree. You should look at the system as a whole instead of picking one figure and comparing that to the best available on the PC. It will have a very high-performing CPU for both arithmetics and floats, and very high-performing GPU, both coupled to a rather damn nippy memory subsystem (no PC will have main memory running at even half of nextbox's bandwidth for YEARS to come).

So 25-ish GBs is less than some cards have today, BUT, there is eDRAM too which will be very low-latency compared to external memory on PC graphics cards. While it may lack the raw bandwidth, not having to wait for 100+ chip cycles for a read transaction to complete will be a major performance increase. Also, while it may lack the raw paper-specs fillrate of some of today's cards, you'll find they never come near in real-world situations, and simple textured pixels isn't what next-gen consoles will be targetting anyway.

So seen as a whole this WILL be an extremely powerful system, and I sure as hell hope somebody will release a Linus or somesuch for it! ;)
 
DeanoC said:
But your counting memory bandwidth wrong,

First there is eDRAM bandwidth,
then the L2 bandwidth
then the system memory bandwidth.

That means that 26 Gb/s has ALOT less to do than a topend PC video cards bandwidth.

Sure, but you seem to be talking a lot more about performance, where I'm talking about cost. I'm not saying that the next generation consoles are going to necessarily be slower than the next generation of PCs (though that's pretty much up in the air at this point), but only saying that in terms of video memory throughput, the xbox2 is utilizing slower memory than what is currently available. I'm not trying to infer anything about performance in this regard because as you said, various caches and optimizations can reduce the need for fast "main" memory. What I was getting at in my original post is that I don't think it's suprising that they are using lower "speced" memory than what is currently available because they need to keep costs down.

Nite_Hawk
 
Guden Oden said:
I don't agree. You should look at the system as a whole instead of picking one figure and comparing that to the best available on the PC. It will have a very high-performing CPU for both arithmetics and floats, and very high-performing GPU, both coupled to a rather damn nippy memory subsystem (no PC will have main memory running at even half of nextbox's bandwidth for YEARS to come).

So 25-ish GBs is less than some cards have today, BUT, there is eDRAM too which will be very low-latency compared to external memory on PC graphics cards. While it may lack the raw bandwidth, not having to wait for 100+ chip cycles for a read transaction to complete will be a major performance increase. Also, while it may lack the raw paper-specs fillrate of some of today's cards, you'll find they never come near in real-world situations, and simple textured pixels isn't what next-gen consoles will be targetting anyway.

So seen as a whole this WILL be an extremely powerful system, and I sure as hell hope somebody will release a Linus or somesuch for it! ;)

See response to DeanoC. :)

Nite_Hawk
 
Nite_Hawk what is the fastest speed MAIN RAM you can buy for your PC and what kind of bandwidth will it give your system?

:D
 
Tahir said:
Nite_Hawk what is the fastest speed MAIN RAM you can buy for your PC and what kind of bandwidth will it give your system?

:D

On my laptop it's 800MB/s :) Still, what is your point? The xbox2's system memory is shared with it's video memory. In terms of performance it'd be very difficult to accurately compare it with PC system memory due to there being so many other factors affecting performance. In terms of cost though, the memory chips in the xbox2 will cost the same as identical chips on a videocard (atleast if they are bought in bulk). This is why I'm argueing that it makes sense for Microsoft to stick with lower end video/system memory if they intend to keep costs down.

Nite_Hawk
 
Tahir said:
Nite_Hawk what is the fastest speed MAIN RAM you can buy for your PC and what kind of bandwidth will it give your system?

:D
ya but, as Nite_Hawk said if the leaked specs are to be believed, then the main RAM of Xbox Next is shared between the VPU and the CPU ... which increases the potential for a bottleneck even further..............
 
Tahir said:
My point is that system memory is not the same as GDDR3...

Is it? I haven't followed the xbox2 rumors that closely, but I wouldn't be all that suprised if they did end up using GDDR3 for their system memory. Perhaps it'll just be DDR or DDR2, but GDDR3 runs so much cooler than DDR that it seems like it might be somewhat advantageous for them to use it (though perhaps the latencies are higher or something)?

Nite_Hawk
 
Nite_Hawk said:
Sure, lower specs. When current top end videocards are pushing a memory throughput of ~36GB/s, and a next generation console that won't be out for atleast a year is talking about ~26GB/s, (less than 3/4th the throughput.) it makes sense to say it's got lower specs.

Assuming the web rumors are remotely close to real, this is a silly thing to say.

All the framebuffer and effects bandwidth will be siphoned off to the edram, the GPU command buffers can be sent directly from the L2 cache, and textures and vertex buffers (in compressed form) will come from main memory.

We're taking about an aggregate system bandwidth approaching 80-90 GB/s.

If you accept that most of the time the edram will reach 256 GB/s efficiency, we're talking a system that can approach 300 GB/s of system bandwidth.

That is nothing to sneeze at. Not to mention since the xenon design is integrated, the GPU can do cool things like pull stuff straight out of CPU L2 cache (try that on a PC), and shaders can read and write stuff straight to main memory (try that on a PC).
 
aaaaa00 said:
Nite_Hawk said:
Sure, lower specs. When current top end videocards are pushing a memory throughput of ~36GB/s, and a next generation console that won't be out for atleast a year is talking about ~26GB/s, (less than 3/4th the throughput.) it makes sense to say it's got lower specs.

Assuming the web rumors are remotely close to real, this is a silly thing to say.

All the framebuffer and effects bandwidth will be siphoned off to the edram, the GPU command buffers can be sent directly from the L2 cache, and textures and vertex buffers (in compressed form) will come from main memory.

We're taking about an aggregate system bandwidth approaching 80-90 GB/s.

If you accept that most of the time the edram will reach 256 GB/s efficiency, we're talking a system that can approach 300 GB/s of system bandwidth.

That is nothing to sneeze at. Not to mention since the xenon design is integrated, the GPU can do cool things like pull stuff straight out of CPU L2 cache (try that on a PC), and shaders can read and write stuff straight to main memory (try that on a PC).

Haven't you read the last like 3-4 posts I've made? I'm not comparing the aggregate memory performance of the PC and the xbox2. If I was, I'd have said so. I'm comparing the throughput of the memory used in the xbox2 to that of PC videocards for cost comparison purposes. In terms of the system/video memory used on the xbox (I only include system as it is shared for that purpose), it is lower spec than that in today's top end videocards. *Yes* the xbox2 could very well have an overall faster memory scheme in place once you factor in L2 cache and edram, and efficient x-bar design, and so on and so forth. All I am saying, is that it makes sense for microsoft to use lower-speced memory (as the rumors are inidicating) because otherwise the costs will be too high.

Sheesh. :)

Nite_Hawk
 
aaaaa00 said:
Nite_Hawk said:
Sure, lower specs. When current top end videocards are pushing a memory throughput of ~36GB/s, and a next generation console that won't be out for atleast a year is talking about ~26GB/s, (less than 3/4th the throughput.) it makes sense to say it's got lower specs.

Assuming the web rumors are remotely close to real, this is a silly thing to say.

All the framebuffer and effects bandwidth will be siphoned off to the edram, the GPU command buffers can be sent directly from the L2 cache, and textures and vertex buffers (in compressed form) will come from main memory.

We're taking about an aggregate system bandwidth approaching 80-90 GB/s.

If you accept that most of the time the edram will reach 256 GB/s efficiency, we're talking a system that can approach 300 GB/s of system bandwidth.

That is nothing to sneeze at. Not to mention since the xenon design is integrated, the GPU can do cool things like pull stuff straight out of CPU L2 cache (try that on a PC), and shaders can read and write stuff straight to main memory (try that on a PC).

I think some of that can be done on a pc...in theory at least, if you tried it you'd probably crash.(at least there's an option in my bios that's something like that, and does crash the computer if I enable it)

BTW, does the xbox2 gpu have edram? I would have thought if they're going through with the shared memory thing again, they would take the same approach and not have any smaller pools of ram. If it has edram(say like 8MB) than there shouldn't be anything to work about.
 
Fox5 said:
I think some of that can be done on a pc...in theory at least, if you tried it you'd probably crash.(at least there's an option in my bios that's something like that, and does crash the computer if I enable it)

Eh? Shaders on the PC can't touch main memory. And the video card on a PC is on the other side of a PCI-Express or AGP bus, there's NO physical way it can directly access something from the L2 cache of the CPU.

BTW, does the xbox2 gpu have edram? I would have thought if they're going through with the shared memory thing again, they would take the same approach and not have any smaller pools of ram. If it has edram(say like 8MB) than there shouldn't be anything to work about.

According to the spec floating around on the web it does. However, that spec may just be a bunch of BS.
 
aaaaa00 said:
Fox5 said:
I think some of that can be done on a pc...in theory at least, if you tried it you'd probably crash.(at least there's an option in my bios that's something like that, and does crash the computer if I enable it)

Eh? Shaders on the PC can't touch main memory. And the video card on a PC is on the other side of a PCI-Express or AGP bus, there's NO physical way it can directly access something from the L2 cache of the CPU.

BTW, does the xbox2 gpu have edram? I would have thought if they're going through with the shared memory thing again, they would take the same approach and not have any smaller pools of ram. If it has edram(say like 8MB) than there shouldn't be anything to work about.

According to the spec floating around on the web it does. However, that spec may just be a bunch of BS.

But a video card can't access system memory over an agp bus? And by extension couldn't the shaders?
 
But a video card can't access system memory over an agp bus? And by extension couldn't the shaders?

it could but thats the last resort deal . It slows things down drasticly.

Pci-e is supposed to fix this or at least make the hit less
 
jvd said:
But a video card can't access system memory over an agp bus? And by extension couldn't the shaders?

it could but thats the last resort deal . It slows things down drasticly.

Pci-e is supposed to fix this or at least make the hit less

Won't pci-e be limited by that system memory is much slower than pci-e is, plus a large portion of the system bandwidth is used by the cpu?
 
Won't pci-e be limited by that system memory is much slower than pci-e is, plus a large portion of the system bandwidth is used by the cpu?

it wont be that fast but it will be faster than the agp version
 
Well, my GFFX doesn't get close to using all of its 256MB RAM in most situations anyway, by next year we will have even more RAM on cards, and the games using all of that RAM will come out long after those cards are released.

I don't think PCIe will have such a huge impact, like AGP8c didn't on 4x and 2x. Situations where the video card uses the main RAM are not that many. Amount of Video Ram is not the bottleneck in video cards. Usually.
 
Nite_Hawk said:
Haven't you read the last like 3-4 posts I've made? I'm not comparing the aggregate memory performance of the PC and the xbox2. If I was, I'd have said so. I'm comparing the throughput of the memory used in the xbox2 to that of PC videocards for cost comparison purposes. In terms of the system/video memory used on the xbox (I only include system as it is shared for that purpose), it is lower spec than that in today's top end videocards. *Yes* the xbox2 could very well have an overall faster memory scheme in place once you factor in L2 cache and edram, and efficient x-bar design, and so on and so forth. All I am saying, is that it makes sense for microsoft to use lower-speced memory (as the rumors are inidicating) because otherwise the costs will be too high.

Sheesh. :)

Nite_Hawk

If you have EDRAM you don't need expensive GDDR3 modules, in 3D graphics rendering you need a lot of bw between graphics and memory chip so the first solution is the best thing you can do with performances in mind.
 
Back
Top