XGI Volari reviews next week?

Check www.ocworkbench.com

Reviews to follow I hope.

 

German print mag PC Games Hardware had a V8 (plain) and a V8 Ultra Duo for a couple of days, and have already sent them back to wherever.

A few bits have already been dropped: Look for postings by PCGH_Thilo

 

but these "tests" were with the same beta drivers!

 

I view the odd pure fillrate results as indication of some really odd issues, which seem to leave hope for some significant room for improvement if they are capable of overcoming the necessary driver challenges.

Some interesting characteristics:

• Why is a Z buffer depth of 32 bits listed as the default instead of 24-bit with an 8-bit stencil?
• Z pixel fillrate is significantly less than Pure Fillrate
• Dual texture and Triple texture results are higher than Single texture fillrate
• PS 1.4 results are lower than PS 2.0 results
• No performance penalty with 4 registers, even with (presumably) full precision!

Overall, assuming no glitches that invalidate the higher results among these, it gives the impression of dual core and general driver teething issues, with the general hardware potential fitting expectations. Also, decent initial results in some key areas (like PS 2.0).

For comparison, assuming resolution is accurately taken into account for calculations:

 

Yes, supplied by XGI. If you know where to get better drivers now, do tell

 

Another thing to note, partial precision does jack to fill rate, just like on the Radeon.

 

I wouldsay that there are some pretty serious driver bugs to sort out when a simple texturing app, when applying a single texture layer, causes a fill rate drop of 2822Mpix/s down to 495Mpixs/s (c.f. 5800U - 1958Mpix/s to 1650Mpix/s) and yet when performing the PS1.1 operation the throughput is significantly higher! If those results are genuine and not a misprint of any kind, and if XGI hope to have a Christmas release on their hands, then you'd hope that they would get the FF texturing performance sorted before anything else.

The z-fillrate results suggest that the Volari's z-buffer handling is not up to scratch compared to, say, the 9800. Their documentation states that there is some form of compression taking place but (a) it doesn't say what gets compressed as far as I can tell and (b) it could be that this is the only system they have to improve z-buffer read/writes (ie. no fast clear, hierarchical z, etc).

 

V5 & 8 are FP24 only.

 

so is the DeltaChrome as well, right?

 

Yup.