KyleSTL, the thing is, anyone contemplating buying a card like this, who has an
older CPU like a 3000+, is bound to also consider getting a better CPU at the same
time, especially as I said before because of how cheap good CPUs like the 7850 are
now, along with 2nd-hand bargains like the 6000+ (45 UKP seems to be an average
price for this model in the UK). All this article shows is how the 4650 does over
an older X700 in a config where the CPU hasn't been upgraded to match. Plus, the
X1950 would have been a more useful base comparison anyway, given how well it sold
at the time (excellent price/performance reported in reviews). So we have some data
points with the results as they are, but the picture is incomplete without the same
tests done with a better CPU (not enough data atm to make an informed decision),
something which should have been obvious from the start, and if toms really want to
cover all the bases then why not show how the card performs when all possibility of
a CPU bottleneck is removed to whatever extent is achievable?
Consider the
older AGP board I have, it can use a Phenom2 X4 3.2GHz - how
would that change the results when fitted with a 4650? ie. despite the advantages
of PCIe, some people would rather just upgrade their existing system. What's the
best CPU an AGP board can take? It varies of course, but users of AGP boards really
need something much better than a 3000+ paired with the 4650 in site tests to gain
any useful insight from the results.
If I'm reading
Asrock's site correctly, the best possible Athlon64 X2 officially
supported on toms' board is the 2.6GHz 5200+ which could then be oc'd to 2.8 or
more. Or there's the FX 2.8Hz. But no wonder toms' 5600+ didn't work, it's not
supported by the BIOS for this board, something they could have easily checked
beforehand.
As for PCIe vs. AGP: I obtained better
3DMark results with my cheap Asrock board
and an X1950Pro AGP than hardware review sites were getting with the PCIe X1950!
I don't think the limits of the AGP bus were really reached until the 3850.
Perhaps my system summaries can be useful to people? They show how performance
changed with the same CPU (6000+) when moving from an AGP/X1950Pro to PCIe/8800GT:
OLD:
http://www.sgidepot.co.uk/misc/mysystemsummary.txt
NEW:
http://www.sgidepot.co.uk/misc/mysystemsummary2.txt
Btw, many of these issues were covered in toms' AGP Platform Analysis in Feb/2007.
Ian.