[ChrisRay]

Take a look at the actual chip and its labels. Doesnt that seem a bit strange? I havent seen any other close ups of the chips.But its labeled an X800 Pro. Perhaps a last minute decision to mark the X800XL? Is it the process? It does seem rather strange.


http://www.techreport.com/reviews/20...l/index.x?pg=1

http://www.beyond3d.com/reviews/ati/r430/

[Dave Baumann]

As I mentioned in our review all signs pointed to R430 being released as a 12 pipe ASIC. Evidently they thought that would be a waste and let a version go out with 16 pipes.

Its not the only one labelled interestingly X700 PRO is labelled X700 XT (that is the PRO ASIC - you can just see the blue of the Sapphire board round the edge).

[Jawed]

Tech Report asserts that R480 is about 297 square mm, whereas R420 is 260 square mm. No size is given for R423.

So is R480 bigger than R423? If so, why?

If not, are R423/480 bigger than R420 solely because of the PCI Express interface?

Is this down to dual-TMDS on the R480?

Jawed

[Dave Baumann]

Bear in mind that these measurements are probably being done with a fairly innaccurate ruler (it is in my case). I'd be surprised if there were any differences between R423 and R480, bu not surprised to see differences between R420 and the others because of the PCIe interface. IIRC the second TMDS is one with an external transmitter on R480.

[Jawed]

Originally Posted by DaveBaumann

Bear in mind that these measurements are probably being done with a fairly innaccurate ruler (it is in my case). I'd be surprised if there were any differences between R423 and R480, bu not surprised to see differences between R420 and the others because of the PCIe interface. IIRC the second TMDS is one with an external transmitter on R480.

I notice you use a flatbed scanner to produce pix of the chips. Since your scanner has a fixed resolution, you could easily measure the length of the sides of the chips in pixels and convert that to mm. Photoshop has a tool for this (for those situations when the alignment of the chips in the scan is not perfectly square).

What dpi are you scanning with, and is it consistent over the last 9 months?

I did think the difference could be solely due to the PCI Express interface, but I haven't found a size for R423 yet. I don't think you scanned R423.

I can't find an explicit description of the configuration of the dual-TMDS in X850XTPE.

http://www.ati.com/products/radeonx850/specs.html

"Single and dual link external TMDS transmitter support for high resolution and/or multi-monitor DVI configurations" which could imply that the second TMDS is not integrated - dodgy grammar though.

Jawed

[Dave Baumann]

Originally Posted by Jawed

I notice you use a flatbed scanner to produce pix of the chips. Since your scanner has a fixed resolution, you could easily measure the length of the sides of the chips in pixels and convert that to mm. Photoshop has a tool for this (for those situations when the alignment of the chips in the scan is not perfectly square).

What dpi are you scanning with, and is it consistent over the last 9 months?

Its fairly difficult to get the perfectly square, and it can also be very difficult to follow exactly where the chip edge is rather than the material round the edge. The scans are usually done at 600dpi, but the final images are resized for the web, so they aren't of any use in relation to one-another.

I did think the difference could be solely due to the PCI Express interface, but I haven't found a size for R423 yet. I don't think you scanned R423.

Yes, my comparison with R423 against R430 (they were on the same scan).

I can't find an explicit description of the configuration of the dual-TMDS in X850XTPE.

http://www.ati.com/products/radeonx850/specs.html

"Single and dual link external TMDS transmitter support for high resolution and/or multi-monitor DVI configurations" which could imply that the second TMDS is not integrated - dodgy grammar though.

[Jawed]

Originally Posted by DaveBaumann

Yes, my comparison with R423 against R430 (they were on the same scan).

Sorry, forgot that R423 was the chip scanned in first posting on this thread - couldn't find it in the 3D Tables.

Using Photoshop's Measure tool (assuming 300dpi), I find that both R430 and R423 are square.

- R423 is 17.4x17.4=303 square mm
- R430 is 15.5x15.5=240 square mm

Jawed

[Jawed]

Oh and assuming that both chips are mounted on the same-sized piece of board, then both R480 and R423 are 303 square mm in size.

Jawed

[Geo]

Originally Posted by DaveBaumann

Bear in mind that these measurements are probably being done with a fairly innaccurate ruler (it is in my case). I'd be surprised if there were any differences between R423 and R480, bu not surprised to see differences between R420 and the others because of the PCIe interface. IIRC the second TMDS is one with an external transmitter on R480.

Dave--

This post will take this thread sideways, so I'll understand if you want to move it (or even delete it). Many months ago, when GF6 and X800 hit the wild, you interviewed Orton, and one of the subjects of that interview and board speculation here, was how NV4x could only be 10% bigger in size yet have so many more transistors. Orton at the time said he didn't understand it and it would take more investigation to figure out what was really going on there.

Well, it's more than six months later. . . any updates? As I recall, the leading candidates were "ATI and NV don't count transistors the same way" and "there must be an extra layer in NV40". The former never held water for me due to the extra trannies that FP32 require, and the clock-for-clock performance difference suggests that NV's pipes are more capable (and thus probably have more trannies each). But it would be nice to get that extra layer confirmed, and what --if anything, it means for the future.