*Official* ATI Radeon HD4850X2 Thread

[rico2001]

Good score vicious_fishes, added tot he list, welcome to the club.

Update:
Owners list up to date 8-3-2009

[rico2001]

@vicious_fishes
Crossfire uses on a method of gpu load balancing known as alternate frame rendering (AFR), in which each gpu is responsible for rendering a whole frame and frames are distributed to gpus sequentially and the shared memory is the pool of which they work in. Frame 0 will go to gpu 0, frame 1 to gpu 1, frame 2 to gpu 2, and so on. Because of the three-frame limit in rendering, the fourth gpu in a crossfire setup will have no effect in some applications. ATI is said to be working on combining split-frame rendering (where one gpu renders the upper and another gpu renders the the lower part of the frame) with AFR in order to improve scaling in such applications.

[vicious_fishes]

Originally Posted by rico2001

@vicious_fishes
Crossfire uses on a method of gpu load balancing known as alternate frame rendering (AFR), in which each gpu is responsible for rendering a whole frame and frames are distributed to gpus sequentially and the shared memory is the pool of which they work in. Frame 0 will go to gpu 0, frame 1 to gpu 1, frame 2 to gpu 2, and so on. Because of the three-frame limit in rendering, the fourth gpu in a crossfire setup will have no effect in some applications. ATI is said to be working on combining split-frame rendering (where one gpu renders the upper and another gpu renders the the lower part of the frame) with AFR in order to improve scaling in such applications.

shared memory pool eh ? here i was thinking it's gpu specific.

thanks for clearing that up.


so should the users in this thread have a benchoff ? a 25k run would be quite nice to see

[SgtHop]

Once I get my rig rebuilt, I'll give you 25k alright. 30k possibly, but that might be considered cheating, cause it wont be just a 4850X2, lol.

[rico2001]

Originally Posted by vicious_fishes

shared memory pool eh ? here i was thinking it's gpu specific.

thanks for clearing that up.


so should the users in this thread have a benchoff ? a 25k run would be quite nice to see

You know, I think you are right vf about the memory. I just looked at a crossfire diagram again and I think my description of how the memory in a crossfire setup was wrong. The alternate frame rendering (AFR), is correct but from the diagram, each gpu works on their frame using their separate memory. I believe where the sharing comes in only refers to the speed and bandwidth. Since memory speed/bandwidth is determined by the gpu with the lowest memory bandwidth, to maintain a constant way off the card through the pci-e. Visual example: Two kids (gpus) in a classroom working on a math problem on two separate chalk boards (memory) and the class only has one door (pci-e) out. And although one kid may be faster at working out the math problem, both kids can talk to each other in the classroom. If that make any sense.

[vicious_fishes]

Originally Posted by rico2001

You know, I think you are right vf about the memory. I just looked at a crossfire diagram again and I think my description of how the memory in a crossfire setup was wrong. The alternate frame rendering (AFR), is correct but from the diagram, each gpu works on their frame using their separate memory. I believe where the sharing comes in only refers to the speed and bandwidth. Since memory speed/bandwidth is determined by the gpu with the lowest memory bandwidth, to maintain a constant way off the card through the pci-e. Visual example: Two kids (gpus) in a classroom working on a math problem on two separate chalk boards (memory) and the class only has one door (pci-e) out. And although one kid may be faster at working out the math problem, both kids can talk to each other in the classroom. If that make any sense.

kinda. but the kid with gddr5 (he can read his formulas in his textbook faster) can get more done than the kid with gddr3 no ? for arguments sake, they went (respective to gpu) frame 0-1-0-1-0-1-0-1-0-1-0-0-1 etc. core speeds don't drop either.

if it downclocked i'm sure ati would have said something. and you did get an extra fps with the same core speed on the 4870 as the 4850.

[sulphur]

Here is my Card: not overclocked yet.



My Benchmarks:


btw are these good benchmarks all at stock nothing OCed.


After OC:

[rico2001]

@sulphur

Outstanding scores sulphur. Thanks and welcome to the owners thread. rep+

Update:
Owners list up to date 8-4-2009

[rico2001]

Comparing gpu scores: 16X pci-e vs. 8X pci-e


4850 X2 + 4850 (tri-fire at 16X/16X)
3DMark 2006
Cpu 5090
Sm2.0 7346
Sm3.0 10490
Phenom II 940X4 @ 3.53G
Foxconn AD7A-S 790GX
http://i400.photobucket.com/al...TI4850X3700-1000-1.jpg

4850 X2 + 4850 (tri-fire at 16X/16X)
3DMark 2006
Cpu 5441
Sm2.0 8319
Sm3.0 11727
Phenom II 720X3 @ 3.9G
MSI K9A2 Platinum
http://img170.imageshack.us/img170/8531/capturehiv.png

Me: 4850 X2 + 4850 (tri-fire at 8X/8X)
3DMark 2006
Cpu 3633
Sm2.0 8454
Sm3.0 11495
E8300 @ 4.0G
ASUS P5Q PRO



4850 X2 + 4850 (tri-fire at 16X/16X)
3DMark Vantage
cpu 11848
gpu 15967
Phenom II 920X4 @ 3.5G
DFI LP DK 790FXB-M2RS
http://service.futuremark.com/result...eResultType=19

4850 X2 + 4850 (tri-fire at 16X/16X)
3DMark Vantage
cpu 12574
gpu 17966
Phenom II 720X3 @ 3.9G
MSI K9A2 Platinum
http://img13.imageshack.us/i/16k.jpg/
Note: a little low on the gpu score here from guy above me, not sure if he is stock. I’ll overclock to 700/1100 to see if I can catch him. If not, then there is a bottleneck.

Me: 4850 X2 + 4850 (tri-fire at 8X/8X)
3DMark Vantage
cpu 8024
gpu 16300
E8300 @ 4.0G
ASUS P5Q PRO


Note: So far, so good, I don't see a pci-e bottleneck. Need to find more people benching games. More later…..

[SgtHop]

Wow. 3DMark is way too arbitrary.