X360 has better MEMORY Bandwith ??

[sub1zero]

The amount of memory is like the size of a tank, say the tank holds X gallons of stuff. The bandwidth is how fast you can fill or drain the tank, say Y gallons per second. You can argue about whether the tank is big enough for a given purpose (X), and you can argue about whether you can fill it fast enough for a given purpose (Y), but what would it mean to talk about the quantity (X*Y)? That's the total memory the system can process in a second.

PS3 GB/s Memory Total
22.4 GB/s x 256MB = 5734.4 GB/s
25.6 GB/s x 256MB = 6553.6 GB/s
ps3 total: 12288.0 GB/s

GB/s Memory Total
22.4 GB/s x 512MB = 11468.8 GB/s
256 GB/s x 10 = 2560 GB/s
x360 total: 14028.8 GB/s

There we have it. The Xbox 360 can process more memory in a second than the PS3. Seriously, how can anyone argue against this?

P.S :: I am not being very original but I don't know tupence about these stuff so I'm quoting someone from the IGN BOARDS. Is this true. DISCUSS

[Rapture]

the extremely high number is the bandwidth to the edram on the gpu as far as i understand it. it really doesnt count for system bandwidth in the manner they say it.

now i re-read what you're saying... that makes no sense.


PS3 GB/s Memory Total
22.4 GB/s x 256MB = 5734.4 GB/s
25.6 GB/s x 256MB = 6553.6 GB/s
ps3 total: 12288.0 GB/s

GB/s Memory Total
22.4 GB/s x 512MB = 11468.8 GB/s
256 GB/s x 10 = 2560 GB/s
x360 total: 14028.8 GB/s

ps3's real total: 48gb/s
xbox's real total: 22.4gb/s + 256gbs (that doesnt really count in the manner they make it out to)

its not 22.4gb per 1mb of ram... it doesnt matter if they had 2024mb of gddr3, the transfer speed would still be 22.4gb/s

[sub1zero]

Why not ?? Care to elaborate ??

[Rapture]

say your local gas station has a flow rate of 1 litre/second on their pumps.

you drive in your car with its 50 litre tank. it takes 50 seconds to fill.

next comes a bigger car with a 200 litre tank. it takes 200 seconds to fill.

just because there is more room to fill doesnt mean the petrol-pump is magically going to be able to fill faster. the size of the tank does not increase the speed, merely the capacity.

*i have no technical knowledge to back this up, but everywhere else ive looked the transfer speed is the same as i have stated, common sense dictates that having more ram doesnt suddenly mean you can transfer it faster and also 14028.8gb/s sounds absolutely absurd to me*

[spyshagg]

exactly...

xbox GPU communicates with the 10Mbytes EDRAM (on the gpu) at 250Gb/s, so that's internal bandwidth between the Gpu core and the GPu Edram only.
Its not a bad thing! its usefull in some things.

but Outside the GPu, everything communicates at 22.4Gb/s... so those 250Gb/s are irrelevant to the rest of the system. you cant Add 22.4 + 250 because its apples and bananas!
The only advantage of that Internal GPU bandwidth is Aliasing filtering without almost any perfomance penalty.

Microsoft trick you into believing in false facts.

1 + 1 is always 2,
but for microsoft 1 + 1= 3 because one of the 1's is a little larger... its nonsense

[Mordecai]

Actually, the speed between the eDRAM and the R500 is only 32GB/s real bandwidth, the speed between the logic inside the eDRAM chip and the 10 MB of memory in the chip is 256 GB/s. ATI has been quoted as saying this.

[the legendary ice man]

I thought 1+1 = 3 for MS because so far they have yet to design an OS that can count properly.


Isn't the transfer speed always going to be slower transfering between two devices than that of internal chips?

[cpiasminc]

ps3's real total: 48gb/s
Actually, if I followed the presentation right (I only listened to audio), it suggests that PS3's real total would be about 83 GB/sec. Reason being that the FlexIO bus can be used to read and write from/to the oppposite device memory. i.e., the CPU can access main memory with it's main mem bus but can also use FlexIO to access video memory.

I wouldn't take this entirely to heart just yet unless someone were to confirm that part of it, but it's probably safe to assume you can access both simultaneously. I think the GPU->CPU is 15 GB/sec and the CPU->GPU is 20 GB/sec. Granted, this bus has more purposes than just memory access. Main reason the CPU->GPU is faster is because you need to transfer render packets across this bus as well. It's more of a general-purpose point-to-point bus, so it's probably best counted separately.

In any case, I said elsewhere that the eDRAM DOES help the efficiency of the GPU by itself. But its bandwidth is not point-to-point at all. It does not communicate externally at that full 256 GB/sec, so it is therefore not comparable at all to the point-to-point memory busses. Considering again the fuel tank analogy that Rapture brought up...
Say that your 200 litre tank was actually constructed as two smaller 100 litre tanks that are connected by a very high flow tube -- one that can handle flow rates of up to 20 litres/sec. If the pump at the fueling station only fills up the tank at 1 litre/sec, does the tank fill up any faster for your 20 litre/sec tube?

[Mordecai]

Hey cpiasminc... if what you're saying is correct, does that mean that the GPU could read/write directly to main memory? aka they could use the flexio bus to access the XDR instead of having to go through the CPU?

[rpgamer_2k5]

Hey cpiasminc... if what you're saying is correct, does that mean that the GPU could read/write directly to main memory? aka they could use the flexio bus to access the XDR instead of having to go through the CPU?

Yes. This was confirmed by Kutaraji himself. The Cell will be able to have access to the VRAM, while the RSX could do the same with the CPU's XDR memory. In addition both Cell and RSX has independent bi-directional bandwidths, so no slow downs.

[Mordecai]

Yes, but my question was whether or not the GPU would go through the CPU to get to the XDR or whether it has a straight path directly to the XDR via the FlexIO bus...

[cpiasminc]

The FlexIO bus is not direct to memory -- rather, the GPU would use the FlexIO bus to queue requests onto the XDR controller on the CPU. There's added latency all right, but the only XDR controller in the entire system is on the CPU, so you have to use its DMAC. While the actual CPU *cores* themselves aren't involved (the DMAC gets a device ID with each request, so it gets routed directly), you can't go direct to the opposite RAM block. Similarly, the CPU can't directly access GDDR-3. It has to queue requests onto the GPU's mem controller, AFAIK.

[Mordecai]

Ok, let me reword my question... The GPU has to make a request to the CPU to access the XDR... We know that there is a direct bus link between the CPU and GPU, but would the CPU have to use that bus to transfer the data that was requested by the GPU? Or can it make the request to the CPU and the CPU sends it by way of the FlexIO (which does not take up any of the bandwidth of the direct CPU-GPU bus or the GPU-GDDR bus)? If it can transfer it seperately through the FlexIO then this really increases the bandwidth and in my opinion shows just how flexible the PS3 architecture is.

[cpiasminc]

The direct bus link between CPU and GPU IS the FlexIO bus. So when you're transferring data from GDDR<->XDR, you end up using that bandwidth. That's why in my earlier post, I still said it's worth considering FlexIO bandwidth separately because it's not exclusively for memory transfers.

In practice, this is why you'd never use the FlexIO bus for memory transfers while also needing to render. You'd use it, perhaps for framebuffer readbacks so you can do some post-processing or something. Either way, if you're using bandwidth to transfer from XDR->GPU->GDDR, you are, at that moment, utilizing bandwidth that might otherwise be used to transfer renderpackets and instructions.

It's not that different from the way AGP cards could access main memory without the intervention of the CPU (well, because the Northbridge houses that control), but you still only have one outgoing/incoming link from CPU<->AGP bus, which is through that Northbridge. Except with FlexIO, it's much faster, and much lower latency... then again, carrier pigeons would be lower latency than goddamn AGP. [mini-rant]

[Mordecai]

Ok, I know what you're saying now, I was hoping that it was something that I hadn't heard as of yet, but I've been reading over at B3D that the PS3 GPU should be able to store and transfer things in XDR and GDDR simultaneously, so they would be able to split up some of the workload and alleviate some of the BW issues.

[High Lander]

The amount of memory is like the size of a tank, say the tank holds X gallons of stuff. The bandwidth is how fast you can fill or drain the tank, say Y gallons per second. You can argue about whether the tank is big enough for a given purpose (X), and you can argue about whether you can fill it fast enough for a given purpose (Y), but what would it mean to talk about the quantity (X*Y)? That's the total memory the system can process in a second.

PS3 GB/s Memory Total
22.4 GB/s x 256MB = 5734.4 GB/s
25.6 GB/s x 256MB = 6553.6 GB/s
ps3 total: 12288.0 GB/s

GB/s Memory Total
22.4 GB/s x 512MB = 11468.8 GB/s
256 GB/s x 10 = 2560 GB/s
x360 total: 14028.8 GB/s

There we have it. The Xbox 360 can process more memory in a second than the PS3. Seriously, how can anyone argue against this?

P.S :: I am not being very original but I don't know tupence about these stuff so I'm quoting someone from the IGN BOARDS. Is this true. DISCUSS

MAAAN, you have really NO IDEA of what you are talking about.

It´s so absurd, that there´s no point discussing it. So PS3 has 22 GB/s PER Megabyte of memory :lol:

And those 10MB people are talking about for 360 are the same PS2 have 4MB, and people say it´s waaay too few.

If 4MB is too few for PS2, for sure 10MB is not enough for 360.

PS2 could also use the shared 32MB for video, the same way the 360 can use the 512MB.


The REAL trick on PS3 is:
Cell can ALSO use the GPU memory, and RSX can ALSO use the Cell memory.

Cell can feed the video buffer directly , without using the RSX. Exactly what they did with the London (Getaway) demo. Everything you could see there was done in real time by the Cell ONLY. No GPU was used. And the Cell could use the 256 + 256 MB

Cell and RSX together is something new and unique, waaaaaaaay beyond X360.

The last trick: It seems the huge 256GB/s bdwdth between the GPU and 10MB cache is not really true. 256GB/s is actually between the edo logic and the ram itself... no more than 32GB/s link GPU and Edo. Anyone could confirm this ????
But again, 4MB of cache was too few for PS2, so 10MB is too few for X360.

There´s no more escape for Microsoft. X360 is the new Dreamcast, unfortunately.

[High Lander]

Hey cpiasminc... if what you're saying is correct, does that mean that the GPU could read/write directly to main memory? aka they could use the flexio bus to access the XDR instead of having to go through the CPU?

Yes. This was confirmed by Kutaraji himself. The Cell will be able to have access to the VRAM, while the RSX could do the same with the CPU's XDR memory. In addition both Cell and RSX has independent bi-directional bandwidths, so no slow downs.

Yes, everything you said here is correct, and is confirmed.

It´s AWESOME.

[imported_The_One]

The last trick: It seems the huge 256GB/s bdwdth between the GPU and 10MB cache is not really true. 256GB/s is actually between the edo logic and the ram itself... no more than 32GB/s link GPU and Edo. Anyone could confirm this ???? Yeah, the 256GB/s of bandwidth is only between the logic and eDRAM. The bandwidth between the GPU and the eDRAM is 32GB/s write and 16GB/s read.