AMD: Volcanic Islands R1100/1200 (8***/9*** series) Speculation/ Rumour Thread

Not anything close to a weak link compared to the existing solution which is at best a 1 GB/s connection via a bridge link.

BeHardware among others have done extensive performance testing of the impact of PCI-e bandwidth. They last year tested a variety of PCI-e configurations with and without CF with a wide variety of workloads.

As far as max rez at 4k, theirs a nice limiter for that, its the cards themselves.

you'll only run into a problem if you are running multiple 4k rez monitors in eyefinity. And if you are doing that...

You might want to go look at a 1x PCI-e interface sometime.

 

In the bygone days that SLI and CF were introduced the support and reliability for peer-to-peer communications over PCI-e was haphazard at best. In the rare case that it actually worked reliably, the bandwidth was likely to be far from optimal esp in consumer chipsets.

 

You think of the 8+MB territory?

I really doubt AMD would invest that many transistors for it. Then I prefer 8MB L2.

 

Most (IE, all but two) of those pins are power and ground. Why would you have a lot of power and ground pins in a crossfire bridge...

 

Well if they upped the L2 cache from 128kB to 256kB per 64bit MC (which would seem like quite a reasonable thing to do, after all Cape Verde already has that much too though it doesn't really seem to help much there), so in total going from 768kB to 2MB, that would only add around 100 million transistors or so (more if they are using spiffy things like 8t srams). Even 4 MB in total would only add another 150 million transistors on top of that, so still not that much. And 4 MB not even talking about 8MB already sort of seems unlikely to me, unless ROPs are now also using L2 cache (though ROP caches were tiny before).

 

800mhz core: http://www.3dmark.com/3dm/1243672 ?

 

There's quite a bit more pins than just power and ground, for hotplug etc. too. The actual transmission at the very least needs one differential pair for sending, one for receiving, plus you've got a differential reference clock too. Also, the ground pins aren't just there for power, they act as shielding, so multiple pairs (if that's more than pcie x1) are always separated by ground signals.

 

BF4 beta benchmark and settings :
http://gamegpu.ru/action-/-fps-/-tps/battlefield-4-beta-test-gpu.html

 

http://translate.googleusercontent....i.html&usg=ALkJrhj2CHhVGJUO5FNmV-QAj1JIPFmeAw

 

http://www.hardware.fr/focus/89/amd-mantle-interview-raja-koduri.html
Probably better off without Google translate as it tries to translate the interview transcript from English to English which doesn't really give great results hehe

The analysis part, on top of the transcript is in French though.

 

or second BIOS with 800mhz ."X" may not be displayed in 3dm...
These two tests were found completely by accident!

 

Accident for "leaked" performance dont exist...

 

http://lab501.ro/editoriale/interviu-devon-nekechuk-senior-product-manager-amd

 

thanks you. Just for the good photo of the chip.

 

Google translated question about the cards(not 290 series) being reused 7000 series:

 

look like only the R260 use a Bonnaire core.

 

But the 260x is the only other chip to feature TrueAudio isn't it? So does that mean they've added it for the re-launch or that it's already there and just not activated? Hopefully they'll up the bandwidth a good bit. The 7790 is horribly bandwidth limited.

 

Anyone knows how long reference 290X is? I dont think it can fit in my case...