TXB on Xbox2 = 51.2GB/s bandwidth??
- Thread starter chaphack
- Start date
Where did this come from? Is there actually anything new and real in this? It strikes me that the poster has cherry picked a number of official statements and then hobbled them together with some stuff thats happened recently (1600MHz GDDR2).
The PS/VS3.0 things is surely a but of a large gaff - R500 will be very different from R420 since its another new architecture, so PS/VS4.0 is almost certain IMO. I'm wouldn't be too sure about GDDR2 support either, since when ATI were initially talking about R400 (whatR500 used to be) they were also talking up the GDDR3 which they had been developing with a number of manufacturers - IIRC GDDR3 had slightly better properties for working as a wider bus than GDDR2 does - so it wouldn't surprise me if GDDR3 was used.
The PS/VS3.0 things is surely a but of a large gaff - R500 will be very different from R420 since its another new architecture, so PS/VS4.0 is almost certain IMO. I'm wouldn't be too sure about GDDR2 support either, since when ATI were initially talking about R400 (whatR500 used to be) they were also talking up the GDDR3 which they had been developing with a number of manufacturers - IIRC GDDR3 had slightly better properties for working as a wider bus than GDDR2 does - so it wouldn't surprise me if GDDR3 was used.
Its from the web no.1 unofficial Xbox site +==+ www.teamxbox.com
BUT like i said, validity NOT confirm.
BUT like i said, validity NOT confirm.
zurich said:
Woops!
Given the physical constaints of a console for heating, etc., wouldn't it make more sense to go with lower clocked ram with a wider data path (as opposed to the quoted 1600mhz GDDR2 ram)? I know it means more traces and more board layers, but it would cut down on heat, yes?
edit: also, isn't say 500mhz 256bit ram more efficient than 1ghz 128bit ram?
GDDR3 consumes ~ 50% less power than graphics DDR2. Who knows, it might even use water cooling coupled with a silent fan not that it will need it...
A low-speed 120*25mm fan right in the top cover (or side, for upright PS3 design - if that's the way they go) ducted across a big-ass skived copper heatsink (or maybe extruded alu would be enough really if area big enuff) covering CPU, GPU, RAM would do the trick I guess. Heck, the Gamecube is cooled by the piddliest fan of the current three consoles and it is rated at like 43W.
*G*
*G*
DaveBaumann said:
Yes, GDDR3 seems way more probable. Micron, the current memory supplier for Microsoft, worked with ATI on finalizing the GDDR-3 spec. Micron and ATI have an established working relationship which should have some political influence in addition to the better performance GDDR-3 has.
My opinion on the memory architecture of the Xbox 2 is that it will be segmented.
A pool of high band video ram for the VPU. GDDR-III
A pool of very low latency ram for the CPU. RLDRAM-II
Low latency can make online games feel more responsive. Getting a smoother experience online can happen if everyones sending out packets at a high sustained rate. Microsoft really can't do much about bandwidth upstream/downstream, but they can attack the latency problem.
They wont need alot of ram for the cpu. THere is no os to hog memory. I can see them go with 64-128 megs of whatever is the best cpu ram of the time. Have the sound chip and the cpu use that and then have the gpu running on 512 megs of whatever is the fastest they want to put in i t. Ati being very good at making low power chips compared to nvidia should be able to keep whatever chip in there running cool. We will most likely see a 1ghz slower than top of the line chip in the xbox2. That way it can run under volted and much cooler. Prob a 3ghz-3.5ghz if they go intel.
Joe DeFuria
Legend
PaulS said:
Doesn't seem low to me at all.
What is your expectation, other than generally "higher than 51.2 GB/sec by a fair margin?"
Megadrive1988
Veteran
Yay!
I hope so.
Joe DeFuria said:
Given that the NV40 is rumoured to be in the 40gb/s range, and that comes say half a year after the NV35 with 27gb/s, only having 51.2gb/s over 2 years from now makes no sense.
I realise that they'd have to finalise everything towards the of 2004, but that seems quite a drop in typical bandwidth increases. If the above mentionned increase holds true over the next few generations, you'd expect something like 60gb/s by the end of 2004.
DeadmeatGA
Banned
...
I agree with one thing, XGPU2 must have eDRAM frame buffer or it will be a toast. Nothing massive like Sony is sticking in their GS3, but just enough to hold one frame buffer, ala Flipper style. 5 MB should be enough to hold 1 Z-buffer and 4 subsample buffers.
I agree with one thing, XGPU2 must have eDRAM frame buffer or it will be a toast. Nothing massive like Sony is sticking in their GS3, but just enough to hold one frame buffer, ala Flipper style. 5 MB should be enough to hold 1 Z-buffer and 4 subsample buffers.
Panajev2001a
Veteran
Deadmeat, that would be probably barely enough at HDTV resolutions.
1280x720 * 32 bpp = 3.51 MB for the back-buffer ( I assume this will be the on-chip one ).
1280x720 * 24 bpp = 2.63 MB
So, for 24 bits back-buffer and 24 bits Z-buffer you would use 5.26 MB.
Maybe we could do non wide-screen 720p ( 1024x720 ).
720p is quite high-res for most people so we might forget about additional AA or they might drop the Z-buffer to 16 bits and turn MSAA on and/or add more e-DRAM.
32 bits for back-buffer and Z-buffer would take 7.02 MB and if you added a 24 bits front-buffer that would mean 9.65 MB.
1280x720 * 32 bpp = 3.51 MB for the back-buffer ( I assume this will be the on-chip one ).
1280x720 * 24 bpp = 2.63 MB
So, for 24 bits back-buffer and 24 bits Z-buffer you would use 5.26 MB.
Maybe we could do non wide-screen 720p ( 1024x720 ).
720p is quite high-res for most people so we might forget about additional AA or they might drop the Z-buffer to 16 bits and turn MSAA on and/or add more e-DRAM.
32 bits for back-buffer and Z-buffer would take 7.02 MB and if you added a 24 bits front-buffer that would mean 9.65 MB.
Panajev2001a
Veteran
Dural said:
This just summed up while I like XDR solution better.
400 MHz base clock ( external bus ) * 4 ( PLL ) * 2 ( DDR ) = 3.2 GHz signaling ( on chip ).
64-128 bits memory controller = 25.6-51.2 GB/s with 128 or 256 data pins.
Move to a 800 MHz base clock:
800 MHz base clock * 4 ( PLL ) * 2 ( DDR ) = 6.4 GHz signaling.
64-128 bits memory controller = 51.2-102.4 GB/s with 128-256 data pins.
At 1.6 GHz of base clock you could have 102.4-204.8 GB/s with 128-256 data pins.
