Thursday, March 5th 2009
ATI M97 Already Using 5 GT/s GDDR5 Memory, Highest Memory Bandwidth for Any mGPU
In an interesting observation, it has come to light that AMD's soon to be released ATI Mobility Radeon HD 4860 MXM modules will carry the industry's fastest 5 GT/s GDDR5 memory chips made by Qimonda. From the part name on the chips, IDGV1G-05A1F1C-50X. One can infer "50X" to be the bandwidth per pin at reference speeds: 5 GT/s (5 GB/s in one direction). Across the 128-bit wide memory bus the M97 HD 4860 has, the chips can churn-out a staggering 80 GB/s memory bandwidth, never before seen on any mGPU board. In comparison, NVIDIA's G92b-based GeForce GTX 280M delivers 60.08 GB/s and AMD's own M98 RV770-based Mobility Radeon HD 4850 doles-out 56.8 GB/s.
The 5 GT/s GDDR5 chips are yet to be used by AMD in its desktop products. Development cycles have gone as far as including "40X" (4 GT/s) memory chips labeled IDGV1G-05A1F1C-40X on company-internal RV790 samples. Another DRAM major, Hynix had announced its plans to introduce 7 GT/s GDDR5 chips back in November 2008. The company is known to commence volume production of the 7 GT/s chip by the end of Q2 2009.
Source:
GPU Café
The 5 GT/s GDDR5 chips are yet to be used by AMD in its desktop products. Development cycles have gone as far as including "40X" (4 GT/s) memory chips labeled IDGV1G-05A1F1C-40X on company-internal RV790 samples. Another DRAM major, Hynix had announced its plans to introduce 7 GT/s GDDR5 chips back in November 2008. The company is known to commence volume production of the 7 GT/s chip by the end of Q2 2009.
13 Comments on ATI M97 Already Using 5 GT/s GDDR5 Memory, Highest Memory Bandwidth for Any mGPU
That would be sweet.
And I should know - Ive been burned by all of that very underhanded marketing on a few processors before (And YES their theoretical bandwidth WAS better, but they still performed worse).
Effective mathematical bandwidth does NOT ALWAYS translate into performance (And can even AGGRAVATE pre-existing bottlenecks, creating buffer overruns that can actually cause entire transfer cycles to drop).
Don't get me wrong, in GFX processing , bandwidth is vital to taking full advantage of a GPU. This does indeed LOOK like good news on the surface, but I'm going to wait for the benchmarks before I begin swishing my credit-card around...
If ATI/AMD can keep this up it will mean great news for us consumers.
Check that review out ppl.