Given the article is supposed to be about whether or not older
Athlons are limiting for newer games, I find it strange the
review used a gfx card that is not exactly in the upper end of
modern performance, or even midrange for that matter. They should
have used a decent 8800GT, not an 8800GS.
My original system was AGP-based, fitted with an X1950Pro 512MB
and 6000+. I obtained much higher frame rates than the article
shows for the 8800GS at 1600x1200 no AA and 16X AF. Even with
high detail and 6x AA, the frame rate was still very decent,
between 30 and 90fps. I believe this is because of the good
memory I had and a decent CPU capable of feeding the gfx card
at suitable speed. I've not checked for a while, but my system
used to be #6 on FutureMark for 3DMark06 for a setup with
an Athlon64 X2 and X1950 AGP (the only systems faster than mine
were those with the newer 6400+ and/or factory overclocked
X1950s that didn't exist when I bought mine).
So, why didn't the article also test with a 6000+ 3GHz and
the faster Black Edition aswell? Especially given how much
cheaper these CPUs have become in the last year, nearly
60% less now. Speaking to other PC users, why do people in
the real world so often observe better results than are
shown in online articles?
Back in May/08 I upgraded with new mbd & gfx (for my birthday),
but I kept the 6000+ CPU, RAM disks, PSU, etc. I switched to
an 8800GT 512MB (Gigabyte, Zalman fan, 700MHz core) and ASUS
M2N32 WS Pro mbd (I wanted proper PCIX and PCIe for SCSI RAID).
See:
http://www.sgidepot.co.uk/misc/ASUS_M2N32-WS-Pro.jpg
http://www.sgidepot.co.uk/misc/Gigabyte8800GT-Zalman.jpg
Baring in mind I still have the same 6000+ CPU and RAM, with
the new card at high detail 1600x1200 and 16X AF with no AA,
results range from 48 to 169fps, with 4X AA they are 47 to 153
fps (not much of a change for minimum fps) and with 8X AA they
are 39 to 107fps! These make the article's 5600+/8800GS results
look positively snail-paced.
Even more amazing, at 2048 x 1536 res, max detail, no AA and
16X AF, I get 51 to 152fps! With 4X AA it's 33 to 109fps, with
8X AA it's 25 to 77fps (still playable!!). In some cases results
are similar no matter what the AA, ie. a CPU bottleneck, though
in practice it doesn't matter because the frame rate is so high
anyway. Just shows that for newer games a newer quad-core CPU
should help, but I was very surprised at how fast the system was.
For reference, 3DMark06 increased from 5583 for the X1950Pro to
11762 for the 8800GT (SM2/3 results are 5457 and 5704, better
than many other 8800GTs I've examined on FutureMark, but the
overall score is lower because other systems have quad-core CPUs
of course).
I have observed the same speed increase for Stalker. In both
cases, I now run the games at 2048 x 1536, max detail, no AA,
16X AF, which is much better than I'd been expecting. Note that
Oblivion doesn't tend to respond well to oc'd systems, but for
Stalker I run the CPU oc'd to 3.225GHz and the gfx oc'd to
790/1790/980, just for the hell of it, though in practice the
frame rates are so high I don't notice much of a difference.
Also, for Oblivion I run without AA as I find just having a high
2K resolution looks fine.
In other words, to be a proper test, the article should have
used a better gfx card (my 8800GT was not expensive, only 128 UKP)
and should have run the same tests with the upper-end Athlon64
X2 CPUs aswell. Indeed, why not test with a GTX280 and 4870x2
to see where the _real_ bottlenecks lie? I keep reading how many
games are not written to take proper advantage of multi-core
CPUs - this would be a good way to test the issue. Go on Paul,
let's see how these results change with really fast cards! 8)
If any of you have an AGP mbd with a good AM2 CPU, I hope my
results will be useful. It _is_ possible to switch to a PCIe
board, keep your CPU and RAM, get a better gfx card and obtain
really good frame rates for modern games. I could even go SLI
if I wanted to (mbd has two PCIe wich both run at x16 for SLI)
though atm I'm using one of the slots for a PCIe U320 SCSI card
(system disk is a 147GB 15K U320).
Here is the data for my original system (when I upgraded systems,
keeping the X1950 AGP I already had from an older Dell 650) and
for my current system with the same CPU/RAM/disk but newer
mbd/gfx (URLs for the screenshots are included):
http://www.sgidepot.co.uk/misc/mysystemsummary.txt
http://www.sgidepot.co.uk/misc/mysystemsummary2.txt
The first page shows how a better _system_ allowed the same
gfx board to run much, much faster because of the better RAM
(also included are results for my brother's Athlon64 3400+
at 2.64GHz also using an X1950Pro AGP). The second page shows
how a nice AM2 CPU is still a decent solution when paired with
a good gfx card.
Having said all that, besides gaming I use my system for video
encoding, so my next system will likely be a QX9650 or somesuch,
or a Nehalem if the prices aren't crazy and the oc'ing works ok.
Ian.