[Disclaimer! The following analysis was done for fun.....and for science.]
6520G
vliw5
no dedicated video ram
400 MHz Core Clock
320 SPs
16 TMUs
8 ROPs
Pixel fillrate: 3.2 GP/s
Texture fillrate: 6.4 GT/s
Memory bandwidth: 17 GB/s shared (assuming DDR3 1333MHz ram in a dual channel configuration)
Peak shader throughput: 256 GFLOP/s
Expected shader throughput (assuming 56% utilization since it is vliw5): 143 GFLOP/s
Expected pixel fillrate (assuming the cpu eats no more than 5GB/s which no consumer application should do): around 1.0 GP/s, this totally depends on what the cpu is doing though
6650m
vliw4
1GB of dedicated video ram
600MHz core clock
900MHz memory clock (1800 effective)
480 SPs
24 TMUs
8 ROPs
DDR3
128 bit wide memory bus
4.8 GP/s pixel fillrate
14.4 GT/s texture fillrate
28.8 GB/s memory bandwidth
Peak shader throughput: 576 GFLOP/s
Expected shader throughput (assuming average utilization of 70% since it's vliw4): 403 GFLOP/s
Expected pixel fillrate (based on memory bandwidth constraints): probably around 3.0 GP/s
7640G
vliw4
No dedicated video ram
497-655MHz core clock (configurable by the OEM)
256 SPs
Unknown number of TMUs, it will be somewhere between 16-32 that's all we know, 16 is expected
Unknown number of ROPs, we can assume 8 since every single low end AMD GPU has 8
Texture fillrate: Unknown (not that it matters since it makes no difference at all to performance these days)
Pixel fillrate (assuming 8 ROPs): 4-5.2 GP/s depending on core clock rate
Memory bandwidth: 25.6 GB/s shared, your 6650m shouldn't have much of a memory bandwidth advantage over this thing
Peak shader throughput: 256 - 333 GFLOP/s (depending on core clock rate)
Expected shader throughput (assuming 70% utilization): 179 - 233 GFLOP/s
Expected pixel throughput: probably around 2.0 GP/s
A 6650m is going to have at most a 40% advantage in usable memory bandwidth and 50% in usable pixel fillrate.
So there are two ways we can compare these, shader throughput, and benchmarks.
Comparing shader throughput is surprisingly easy since both the 6650m and 7640G use the exact same SP architecture. In other words with any game their shader utilization will be exactly the same assuming no outside bottlenecks. Depending on the clock rate the 6650m is going to be between 1.75 to 2.33 times as fast as the 7640G in shader throughput. Most manufacturers are going to be clocking it near the maximum but there is going to be a small drop in performance from the shared memory hierarchy so we can predict the throughput to be almost exactly twice as fast in most scenarios. Under the absolute best case scenario you should gain about a 30% performance boost from asymmetric crossfire making your configuration about 2.6 times as fast, and you would need near perfect scaling for that.
Basically with all variables accounted for you're looking at a 2-3x difference in performance depending on what the core clock rate is set to, how well the game uses asymmetric crossfire, and how much framebuffer and texture buffer read/write operations the game is doing. Since game performance doesn't scale perfectly with these things it will likely be closer to 2x in the vast majority of circumstances. This all assumes the application is shader bound, which most games are.
In memory bound situations the 7640G will actually do better compared to the 6650m. At the absolute minimum the 6650m is going to be at least 40% faster under any scenario. I would expect memory bound applications to perform in the 1.7x-2.5x range depending on exactly how memory bound they are.
Now what about actual benchmarks? Well there are none. At least not for the A10-4500M with the 7640G. Only the top of the line llano and trinity chips have been benchmarked by the major review sites, the 4600M and 3500M. In those tests we see a 20% increase in IGP performance on average. So that's literally all we have to go on. Which means......
https://www.youtube.com/watch?v=gENVB6tjq_M
Quote:not to mention I have dual graphics.
You keep saying that as if it's a big deal. Go look at some benchmarks for asymmetric crossfire, it barely makes a difference.
Quote:AMD's second gen APU is trinity, how much of an improvement do you think the Trinity A8-4500 has over the Llano A8-3500 in terms of regular pc gaming.
20%
Quote:Also if I overclock the Trinity chip do you think improvement in emulation will be noticeable?
Yes.
