ok anandtech review, compared to their results for UT3 gpu performance they've done previously. in the previous article they ran 3 different maps and showed results for each map
http://anandtech.com/video/showdoc.aspx?i=3127&p=8
in the 8800gt review, the numbers it states for the GTS, 70fps, were best case scenario of the 3 maps, the fastest it can do. the GTX numbers it uses aren't from that same map, infact it seems to be about an average across the 3 maps, and the 2900xt numbers it uses are the LOWEST the 2900xt on its worst map. but on the map the 2900xt does its worst 79.5fps, the gts gets 60, over 30% slower. on that same map the gtx gets 83, some what short of whats shown in the 8800 gt vs gtx numbers.
on average in UT3, and bioshock(from other reviews around the web, mostly of 2900pro recent reviews with the 7.9/7.10 drivers) the 2900xt is around 30% faster than the gts, and within 2-3% of the gtx on all maps, in almost all situations. but almost every single review of the 8800GT seems to be using incredibly odd benchmarking.
when was the last time anandtech did a review without a direct bar graph of all top end cards current and last gen? when was the last time they didn't use aa/af in gpu reviews. when was the last time they refused to give gts/2900xt/gtx numbers on the same page? never?
what about the site that only used 2560x1600? what about the missing reviews off a ridiculous number of sites who would all normally have reviews up the second the NDA was up? what about the 25 slide presentation "some reviewers" were invited to that tried to lay out exactly how to benchmark in the way Nvidia wanted.
almost all the reviews are picking and choosing results to make the gts as close as possible, the 2900xt as slow as possible?
with anandtech you can plainly see they have used the 2900xt's lowest number on the hardest running ut3 map, against the gts's highest number on the easiest map, theres simply no two ways about it.
anandtech + hl2, the gtx/2900xt in their recent analysis shows the same numbers in for the 2900xt in the 8800gt review, but they've come up with another 8-9fps for the gtx, out of nowhere. but that would be the 3 of the biggest last 3 titles released that the 2900xt matches, or within 3-4fps of the gtx, while inbetween 20-30% ahead of the GTS. but no you're right, the 2900xt is crap, especially as you can i think still find them for £160 as pro's today in the uk
worth noting that lots of the review sites that do have reviews are comparing 2900pro's with pre overclocked 8800gt's. its just the most dodgey release of reviews i've seen in the 7-8 years i've been building computers.
i do give ati minus points for one thing, the difficulty of getting performance out of the gpu seems to entail waiting for a driver release for almost every game for it to be optimised, while nvidia do the same it would seem nvidia are going from close to final performance to final performance, where ati are going from(in some games at least) absolutely horrific performance to good final performance. now at least ati are fast with releasing new drivers, but its showing their gpu is great with the right optimising, and incredibly difficult for a default driver to seem to get good performance out of.
