Follow along with the video below to see how to install our site as a web app on your home screen.
Note: This feature may not be available in some browsers.
I say it's crap. It's clearly written by a very "uneducated" person - the R680 certainly does not "rely" on the PLX bridge for communication.
http://forums.vr-zone.com/showthread.php?t=285719
Linked through the memory? Does anyone have an idea of what this could mean?![]()
The main purpose of the bridge chip is to redistribute the PCIe lanes to the two chips. For communication, there's another interface - the "CF bridge" (which you use to connect two boards in CrossFire; it's routed through the PCB on X2 cards), although some communication goes through the PCIe bus. I wouldn't call that "rely on PLX for communication" though.Why? The two gpu are linked through the bridge chip, or not?![]()
No, certainly not. The games use 4 of the 5 "SPs" on average (there are different shader commands with different "width" and some games generally use "wider" shaders than others). There's no way you could leverage the potential of a 10-way unit with that, hence there'd be no real performance difference between, say, 320 5-way ALUs and 320 10-way ALUs.Had a strange thought last night... how about 80 10-way units instead of 160 5-way ones?
Curious why all the 32 TMU talk... 20 or 40 would seem to make more sense if it is 800 ALUs.
I went to bed last night thinking "if it's really 800:32, that's amazing."
I wake up and find it's definitely 800 and extremely likely to be 32 and I'm gobsmacked
I'd already concluded, based upon RV635->RV670 scaling where 60% extra die-space delivers 2-3x performance, that the scalability of this architecture is good - but this is outrageous.
We're looking at ~1.1 billion transistors for RV770?
With that much ALU it seems inevitable that it'll generally be TU:RBE limited. Crysis performance on a single RV770 is going to be an eye-opener because HD3870X2 scales really badly.
Jawed
Because 160 5D ALUs is either going to be 5 clusters of 32 or 10 clusters of 16. Each cluster will have 8TA and 4TF if it follows R6xx's design.no-X said:Why? 160 5D ALUs (800 SPs) and 32 TMUs ~ nice 5:1 ALU:TEX ratio...
2 TMU blocks like Fellix suggested?:I think the count, based on the size, is more on a 950 million transistors, give or less 50 million. And I think that each TMU block will be bigger, or, better, to have 2 TMU blocks tied to each sublevel in the SIMD.
#0 #1 #2 #3 #4 #5 #7 #8 #9 || #0 #1
[ALU] [ALU] [ALU] [ALU] [ALU] [ALU] [ALU] [ALU] [ALU] || [TEX] [TEX]
[ALU] [ALU] [ALU] [ALU] [ALU] [ALU] [ALU] [ALU] [ALU] || [TEX] [TEX]
[ALU] [ALU] [ALU] [ALU] [ALU] [ALU] [ALU] [ALU] [ALU] || [TEX] [TEX]
[ALU] [ALU] [ALU] [ALU] [ALU] [ALU] [ALU] [ALU] [ALU] || [TEX] [TEX]
http://forums.vr-zone.com/showthread.php?t=285719
Linked through the memory? Does anyone have an idea of what this could mean?![]()
The main purpose of the bridge chip is to redistribute the PCIe lanes to the two chips. For communication, there's another interface - the "CF bridge" (which you use to connect two boards in CrossFire; it's routed through the PCB on X2 cards), although some communication goes through the PCIe bus. I wouldn't call that "rely on PLX for communication" though.
Wasn't the shader clock rumor discarted?
Well, speaking theoretically, why not.A.L.M. said:If I understood well, so even an Hypertransport connection (I know that is unlikely, just doing an example) will be compatible with the presence of a PLX chip on the PCB?
2 TMU blocks like Fellix suggested?:
http://forum.beyond3d.com/showpost.php?p=1129739&postcount=604
Code:#0 #1 #2 #3 #4 #5 #7 #8 #9 || #0 #1 [ALU] [ALU] [ALU] [ALU] [ALU] [ALU] [ALU] [ALU] [ALU] || [TEX] [TEX] [ALU] [ALU] [ALU] [ALU] [ALU] [ALU] [ALU] [ALU] [ALU] || [TEX] [TEX] [ALU] [ALU] [ALU] [ALU] [ALU] [ALU] [ALU] [ALU] [ALU] || [TEX] [TEX] [ALU] [ALU] [ALU] [ALU] [ALU] [ALU] [ALU] [ALU] [ALU] || [TEX] [TEX]
Jawed