AMD: R9xx Speculation

That being said, I've been thinking...

If the rumor that Barts/Cayman will be multi RPE chips is true and that there will be a Tesselation unit in each one of the RPEs, what kind of performance increase are we looking at? If each RPE has a Tesselation unit that is as much capable as Cypresse's, would it be safe to assume that Barts, being a supposed 2 RPE part, could be twice as fast in tesselation? Could they scale linearly?

I am talking about theoretical tesselation performance here, not framerate. I do understand that there are other factors that could not let it scale 100% on every engine.
 
Which slides are you referring to?



He leaked more of them (and removed them quite early), but I didn't save it. I remember he leaked also the slide with feature-set (expreview leaked content of this one, too) and some less interesting ones...
 
As far as I know, those rumors were started by Charlie, and he later went back on them. So I for one am dismissing them.
 
As far as I know, those rumors were started by Charlie, and he later went back on them. So I for one am dismissing them.

But Gipsels analysis of the driver code showed a different ALU-configuration (4 components instead of EGs 5) of a chip family called "NI" and 2 months ago catalyst 10.8 was released with the known codenames of upcoming cards also combined with the "NI".

These were all informations from AMD. Or were these disinformation?
 
But Gipsels analysis of the driver code showed a different ALU-configuration (4 components instead of EGs 5) of a chip family called "NI" and 2 months ago catalyst 10.8 was released with the known codenames of upcoming cards also combined with the "NI".

These were all informations from AMD. Or were these disinformation?

Sorry, my post was ambiguous. I mean the rumors about SI (now NI) having Evergreen shaders + NI (now SI, I know, it's confusing) "uncore" were started by Charlie, who later stated that NI (that is, the real NI :D) would have 4D shaders.

In other (and much simpler) words, I think the HD 6000 series will indeed have 4D shaders.

You have to admit, abstract codenames such as Evergreen and Northern/Southern Islands serve their purpose very well: they confuse the hell out of anyone trying to figure things out.
 
Probably because gddr5 uses differential data signals, which needs more wires.
Except GDDR5 has SE data lines, not differiential.

The excess traces/pads are mainly for clock (outgoing clocks) and power (same power + less voltage = higher current = more solder balls needed)
 
But hopes about massively reworked TMUs and a NV-HQ-equivalent AF might be diminished.
I don't think anything new has emerged on that subject. I've talked about texture architecture patent documents but I don't know of any other information.

But Gipsels analysis of the driver code showed a different ALU-configuration (4 components instead of EGs 5) of a chip family called "NI" and 2 months ago catalyst 10.8 was released with the known codenames of upcoming cards also combined with the "NI".
http://forum.beyond3d.com/showpost.php?p=1453118&postcount=1025

I tend to find Gipsel's findings pretty compelling.

May be command bus is differential then, I seem to remember that gddr5 used some differential signalling.
http://www.interfacebus.com/GDDR5-Memory-IC-Pinout.html

Just clocks.
 
Sorry, my post was ambiguous. I mean the rumors about SI (now NI) having Evergreen shaders + NI (now SI, I know, it's confusing) "uncore" were started by Charlie, who later stated that NI (that is, the real NI :D) would have 4D shaders.

In other (and much simpler) words, I think the HD 6000 series will indeed have 4D shaders.

You have to admit, abstract codenames such as Evergreen and Northern/Southern Islands serve their purpose very well: they confuse the hell out of anyone trying to figure things out.

My view on them is that the now upcoming lineup is in fact "SI" due the island names, the rumored other "NI/SI" island names lineup had all but one island norther than these 4.

So "NI" in the names would point out it's "NI family", but NI family would be divided between the SI (HD6) and actual NI (HD7?) due 32nm being cancelled.
 
This silly season is sooo boring, it's not worthy of the silly season nickname :cry: Were the f are the leaks??

I'm trying to remember the last time he was right ;)...

During Rv770 pre-launch speculation, he said he was dissapointed by Gt200, because he saw 4850CF prowess :D. He went on and off saying he had a 4850, that he was gonna post pics blah blah, but then said amd asked him not to xD
 
This silly season is sooo boring, it's not worthy of the silly season nickname :cry: Were the f are the leaks??

Haven't seen this posted yet.

AMD's upcoming HD 6800 series HD 6850 and HD 6870, has recently been exposed on the network of its 3Dmark scores, AMD Radeon HD 6870 has a 900Mhz core clock, and with 960 stream processors and 256-bit memory frequency wide, with a 1050Mhz (effective frequency of 4200Mhz) of GDDR5 memory. 3DMark Vantage performance preset results P16270 points and X7538, 3DMark06 score of 19,480.
AMD Radeon HD 6850 core clock to 775Mhz, and equipped with 800 stream processors and 256-bit memory with a 1000Mhz (effective frequency of 4000Mhz) of GDDR5 memory. 3DMark Vantage performance preset results P14872 points and X6549, 3DMark06 score of 18,750.

3DMark Vantage (Performance-preset) / 3DMark06:

NVIDIA
GeForce GTX 480: P18376/19671 GeForce GTX 480: P18376/19671
GeForce GTX 460 (256-bit & 1 Gt): P13623/18601 GeForce GTX 460 (256-bit & 1 Gt): P13623/18601
GeForce GTX 460 (192-bit & 768 Mt): P13386/18259 GeForce GTX 460 (192-bit & 768 Mt): P13386/18259
GeForce GTS 450: P9792/15793 GeForce GTS 450: P9792/15793

AMD
ATI Radeon HD 5870: P17924/19433 ATI Radeon HD 5870: P17924/19433
AMD Radeon HD 6870: P16270/19480 AMD Radeon HD 6870: P16270/19480
ATI Radeon HD 5850: P15593/18762 ATI Radeon HD 5850: P15593/18762
AMD Radeon HD 6850: P14872/18750 AMD Radeon HD 6850: P14872/18750
ATI Radeon HD 5830: P14014/17298 ATI Radeon HD 5830: P14014/17298
ATI Radeon HD 5770: P11017/16358 ATI Radeon HD 5770: P11017/16358
ATI Radeon HD 5750: P9124/14966 ATI Radeon HD 5750: P9124/14966

Source: http://translate.google.com/translate?hl=en&sl=zh-CN&tl=en&u=http://www.xfastest.com/cms/tid-52081/
 
Back
Top