- Feb 5, 2006
- 2,778
- 1
- 81
I was just wondering instead of raising pipes and clocks like ati and nvidia are doing, maybe if they came out with a 512bit memory card it would blow everything away?
Just speculation..
Just speculation..
Is 512bit memory gonna come out soon?
- Thread starter SpeedZealot369
- Start date
lol 512bit is different than 512MB of on board memory. I think 512bit would be pretty nice addition though I don't know how bandwidth limited we really are right now... Either way that would be extremely expensive..
When people start wanting to play @ 2560x1600 w/ 6x AAA/16xHQAF we will be memory bandwidth limited.
Wasn't the Matrox Parhelia a 512-bit card?
edit:Matrox Parhelia-512
Matrox a while back(2002) introduced a 512-bit GPU with 256-bit ram.
edit:Matrox Parhelia-512
Matrox a while back(2002) introduced a 512-bit GPU with 256-bit ram.
http://www.samsung.com/Products/Semicon...DRAM/512Mbit/K4J52324QC/K4J52324QC.htm
These chips are 512Mbit each, so eight such chips suffice for a total of 512MB of graphics memory with 256-bit access.
R580 uses them
These chips are 512Mbit each, so eight such chips suffice for a total of 512MB of graphics memory with 256-bit access.
R580 uses them
Does anyone know how much of an increase in cost is involved per card if they went with a 512-bit memory buss or whatever it's called? I remember an interview with an ATI developer that said the main bottleneck for the highend cards is memory bandwidth right now. (Thus their reasoning for sticking with 16 pipes and adding so many vertex shaders instead of less shaders and more pipes). So I would think 512-bit memory would help a lot. But would people buy $750 cards? Maybe they wouldn't be that much. Anyone have an idea?
Originally posted by: toyota
its ok you are from Mississippi.Originally posted by: Ike0069
oops
I need to learn to read I guess.
Whohoo. The MS curse helps me out for once. :laugh:
toyota
Lifer
- Apr 15, 2001
- 12,957
- 1
- 0
lol. i am also from Mississippi thats why i said that. i moved to Birmingham, Alabama about three years ago but i am originally from Hattiesburg, MS. i had never heard of your town and had to look it up on the map.Originally posted by: Ike0069
Originally posted by: toyota
its ok you are from Mississippi.Originally posted by: Ike0069
oops
I need to learn to read I guess.
Whohoo. The MS curse helps me out for once. :laugh:
- Feb 5, 2006
- 2,778
- 1
- 81
akugami
Diamond Member
- Feb 14, 2005
- 6,210
- 2,551
- 136
I think most are getting bandwith and memory space mixed up. The ATI memory controller can use 512 bit memory (actually bi-directional 2x256 bit, effectively 512 bit each way) but most of the available RAM are 256 bit. This is a good move on ATI's part because no video card is truly built from the ground up, they do re-use parts from older GPU's and ATI's forward thinking means that the memory controller on the R600 core is likely nearly the same as on the R520/R560 GPU's. And the fact that ATI's memory controller supports GDDR4 bodes well for ATI's future GPU's as far as needed work goes. The less work you need to do on one aspect of the GPU means more time can be devoted to improving other areas.
nVidia would have more work (though I'm no engineer so I don't know how much) to get their GPU's up and running with 512 bit memory since AFAIK their memory controller is 256 bit only. I'm sure they have it well in hand, and if not, I'm sure the ones most happy would be ATI if nVidia should stumble again. After all, it's nVidia's own blunders with the Geforce 5x00 series, coupled with the excellent R300 core that allowed ATI to be in it's current position today.
So
Lifer
- Jul 2, 2001
- 25,923
- 17
- 81
having taken a number of classes in digital logic / microarchitecture...we wouldn't see much of any performance benefit going from our current memory to a greater bit/word scheme, and in some cases, it might hurt performance. Why do you think it took ~20 years to even START to move from 32bit CPU's to 64bit? Because there was no tangible performance gain.
Why do you think that the consoles dropped the 'bit' wars after the N64?
Why do you think that the consoles dropped the 'bit' wars after the N64?
zephyrprime
Diamond Member
- Feb 18, 2001
- 7,512
- 2
- 81
I'm not so sure about the idea that we have adequate bandwidth right now. Just look at the amount of proportional bandwidth between the 9700pro and the x1900.
9700pro: 8 pipes, 325MHz, 620mhz ram data rate, 256bit memory width
x1900: 16 pipes, 625MHz, 1450MHz ram data rate, 256bit memory width
Do the math and you'll see that the 9700pro had 64% more bandwidth per clock tick*per pipe. Now the x1900 probably has more efficient memory usage techniques up its sleeve but isn't this outweighed by the following factors:
1. Pipelines are more efficient.
2. proportional latency of ram is getting worse, not better
3. greater usage of pixel shaders increase memory pressure
The geforce 7900 is in even worse shape for memory bandwidth because it has 24 pipes. In fact, I would guess that the reason the 7900 doesn't blow away the 1900 even in games that don't have really heavy pixel shader usage is because the 7900 is more bandwidth constrained than the 1900.
The problem with increasing the memory bus width is that the PCB would probably be too expensive to make. The current PCB in graphic cards is already 12 layers thick. That's why I think companies need to get creative and create some sort of multichip module like IBM uses for some of their stuff. Yeah, the pcb for for the MCM would be expensive per square inch but since it's much smaller than the current board PCB, I think the whole thing would be cheaper overall.
(and Acanthus is right, the current 256bit memory busses are actually 4x64bit channels)
9700pro: 8 pipes, 325MHz, 620mhz ram data rate, 256bit memory width
x1900: 16 pipes, 625MHz, 1450MHz ram data rate, 256bit memory width
Do the math and you'll see that the 9700pro had 64% more bandwidth per clock tick*per pipe. Now the x1900 probably has more efficient memory usage techniques up its sleeve but isn't this outweighed by the following factors:
1. Pipelines are more efficient.
2. proportional latency of ram is getting worse, not better
3. greater usage of pixel shaders increase memory pressure
The geforce 7900 is in even worse shape for memory bandwidth because it has 24 pipes. In fact, I would guess that the reason the 7900 doesn't blow away the 1900 even in games that don't have really heavy pixel shader usage is because the 7900 is more bandwidth constrained than the 1900.
The problem with increasing the memory bus width is that the PCB would probably be too expensive to make. The current PCB in graphic cards is already 12 layers thick. That's why I think companies need to get creative and create some sort of multichip module like IBM uses for some of their stuff. Yeah, the pcb for for the MCM would be expensive per square inch but since it's much smaller than the current board PCB, I think the whole thing would be cheaper overall.
(and Acanthus is right, the current 256bit memory busses are actually 4x64bit channels)
memory is a big bottleneck on my 7800GT. with the core at 480 or higher, overclocking the memory from just 1.15 to 1.2 gives me a big increase in peformance
Memory interface width between CPU and external memory has little to do with how wide a CPU's ALU's are. For example, the memory interface between the sticks of DRAM on your motherboard and a cpu is 64 bits even if you have a cpu that can only handle 32 bit instructions. Likewise, the onchip connection between the cpu core and the L1 and L2 cache can be as wide as 128 bits on a K8 processor or 256 bits on Pentium4.
Limiting factor for 512bit wide external memory interface between a GPU and memory chips are packaging technology able to handle 256+ more pins and the board designers able to place and route 256 more wires from the extra memory modules to the GPU. Each individual DDR3 memory chip is only 32 bit wide for data. That is why 8 chips give a total of 256 bits of data width. 512 bit interface would require 16 memory chips. Look at a X1900XTX or a 7900GTX and you can see how much space just 8 memory modules take up and look at all the wires coming from them and connecting to the GPU. Now imagine if the number of lines approximately doubled.
Limiting factor for 512bit wide external memory interface between a GPU and memory chips are packaging technology able to handle 256+ more pins and the board designers able to place and route 256 more wires from the extra memory modules to the GPU. Each individual DDR3 memory chip is only 32 bit wide for data. That is why 8 chips give a total of 256 bits of data width. 512 bit interface would require 16 memory chips. Look at a X1900XTX or a 7900GTX and you can see how much space just 8 memory modules take up and look at all the wires coming from them and connecting to the GPU. Now imagine if the number of lines approximately doubled.
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes Discussion Threads
- Started by Tigerick
- Replies: 21K
-
-
-
Facebook
Twitter