What i would love to see is a review where the same cards with different RAM amounts were tested against each other...

[dug777]

I'm honestly sceptical of the benefits of say a 256mb 6600GT over a 128mb 6600GT (especially taking into account the price difference, based on my experiences with Ti4200s & 9550s with 64/128mb and 128/256mb respectively, where i could perceive no playability differences AT ALL (and the 9550s both overclocked to beyond 9600XT speeds)), so i would love to see a comprehensive review that shows you the declining benefits of increased RAM size @each mem bus width (64/128/256bit)...

*subtle hint*It would be awesome if AT would do this*subtle hint*

It would be particularly amusing at the lower end of the market, i can't see a 512mb 6200 beating a 128mb one by much

 

[dug777]

bUmP

 

[Gstanfor]

What are your system specs and what would you like to see?

My 256mb 6600GT is in a system as follows:

ABIT KW7 motherboard
Athlon XP2400+
1 gig PC2100 memory (runs single channel on this POS m/b due to it being 1x 512, 2 x256 - never had a problem running dual channel on the nForce2)
ExpertVision 6600GT 256mb GDDR3, clocked at 550 core, 1 gig memory
Win XP original release, no service packs applied
DirectX 9.0c feb 2006 update
Forceware 83.90 v2, extreme-G version

 

[deadseasquirrel]

I'm not sure how relevant this review is, but it compares an x800 with 256mb and an x800XL with 512mb on COD2 demo. It shows the 512mb making a HUGE difference.

This might be limited to COD2 demo because, as you can see from their BF2 tests, the x800XL 512mb gets beat by the x850xtpe easily, whereas on COD2, it easily beat the xtpe. In fact, Firing Squad ran the same tests on COD2 with the x800XL 256/512 and confirmed Amdzone's results.

In another review, Firing Squad showed no improvement in FEAR or Quake 4 across all resolutions with a clocked-down 512GTX and 256GTX. On the next page, they showed COD2 and HL2 having a little increase from the 512GTX, but not near as much as on the x800XL 512mb percentage-wise.

I think we're still a little ways off from really taking advantage 512mb video memory. Though, with all of ATI's high-end cards, and nV's soon-to-be-released high-end card, having 512mb, it's going to become more commonplace. And by R600/G80, it'll likely be mainstream, and maybe actually used.

 

[dug777]

Originally posted by: Gstanfor
What are your system specs and what would you like to see?

My 256mb 6600GT is in a system as follows:

ABIT KW7 motherboard
Athlon XP2400+
1 gig PC2100 memory (runs single channel on this POS m/b due to it being 1x 512, 2 x256 - never had a problem running dual channel on the nForce2)
ExpertVision 6600GT 256mb GDDR3, clocked at 550 core, 1 gig memory
Win XP original release, no service packs applied
DirectX 9.0c feb 2006 update
Forceware 83.90 v2, extreme-G version

Wicked mate

I've got a AXp rig too, so we should be pretty comparable :thumbsup:

tbred-b @ 2.1Ghz (167x12.5)i can clock back to 2400+)
1gb RAM (2x512mb @167mhz)
ASUS A7V8X (old skool ****** yeah )
Xp Pro SP2 (latest patches)
Latest Dx
Hmm, might have to get me some of those drivers, i've been away from my rig for a couple of months

& a 128mb Sparkle 6600GT @590/1145 with my hawt new Vf700 Cu(which i can clock back for comparability sake).

Do you have FEAR? Hl2? 3dmark05/6? I'll stick to stuff that is easy to benchmark...

 

[blckgrffn]

Yes, mr seasquirrel. For example, I can definitely tell the difference between 64 and 128 meg cards right now, and 128 to 256 is defintely there when you are trying to run CS:S at 1400*1050 4xaa 8xaf. All the bandwidth in the world can't save you from running out of frame buffer. The issue is you have to have enough GPU to actually be able to run at those settings/texture levels, and I don't think the 4200's and 9550's have ever been that kind of card.

The difference between 512 and 256 is sure to come soon as games have that amount of frame buffer available on this gen of consoles. I doubt we will go much past that for a while. I would guess 1024 in maybe two years time on the highest end cards. F'in crazy...

*The declining returns bit would be most interesting to me.

*Make sure you guys run nice AA levels, etc. Also, watch for the split second zeroes that loading from ram can bring. That is what drove me nuts with my 6800NU, even though most of the time it was a cool 50FPS...

 

[Gstanfor]

I have all the software you mentioned Dug, have never tried to bench HL2 before, but you point me to whatever I need to do it and I will.

 

[dug777]

Originally posted by: Gstanfor
I have all the software you mentioned Dug, have never tried to bench HL2 before, but you point me to whatever I need to do it and I will.

i usually just run the CS:S Stress test...not sure how to best go about it for hl2 proper...

 

[deadseasquirrel]

Originally posted by: dug777
i usually just run the CS:S Stress test...not sure how to best go about it for hl2 proper...

Link to Anandtech's HL2 bench demos.

To run, I believe you just simply go to the console and type in:
playdemo [demoname]
Make sure the AT dem files are in your \Program Files\Valve\Steam\SteamApps\[username]\half-life 2\hl2\ directory (or whereever you installed).

Alternatively, you could use HardOCP's benchmark or Hardware OC's.

 

[Gstanfor]

Don't play CS myself, and the PC the 6600 is installed in has no 'net or lan access

 

[dug777]

No worries

We'll just use the AT hl2 bench demos then ya?

Need to work out some settings:

FEAR: Everything maxed @1024, no softshadows, No AA/8xAF then maybe a run with 2xAA/8xAF?

Hl2: 1024/4xAA/8xAF everything set as high as it will go? Maybe a 1280 run too?

Both 3dmarks at standard 3dmark settings?

Have to agree on driver settings too...

I have to go to uni soon, so i'll get on with this later this arvo/tonight, but if you want to confirm some settings then you can get startedsooner

 

[Gstanfor]

Those settings sound okay to me, I'll always specify the settings for each result anyhow. I won't start testing until later myself - the working day isn't over yet.

 

[dug777]

Originally posted by: Gstanfor
Those settings sound okay to me, I'll always specify the settings for each result anyhow. I won't start testing until later myself - the working day isn't over yet.

No worries

i got the drivers so i'll install them now...

i'll run everything at my current speeds, then back of to your CPU & GPU speeds & run them again. I'll post all my settings & do the easier stuff first (like 3dmark 05/06 & FEAR).

 

[dug777]

FEAR: 1024x728, Forceware 83.90 v2, extreme-G version as they install (just installed them & ran the benchmark), CPU @2.1Ghz (167x12.5), GPU @590/1145, FEAR @ all settings maxed in the computer panels (so sound, detail etc), everything maxed in the video card panels (NO SOFT SHADOWS), No AA/8xAF.

Min: 17
Avg: 38
Max: 89

&

20
45
35

With same settings but now 2xAA:

MIn: 14
Avg: 30
Max: 75

&

40
42
18

EDIT: & now at 550/1000

Min: 15
Avg: 37
Max: 87

&

25
42
33

With same settings but now 2xAA:

MIn: 13
Avg: 28
Max: 68

&

46
40
14

 

[BlacKJesuS]

Dug i love you

 

[dug777]

Originally posted by: BlacKJesuS
Dug i love you

:shocked:

 

[Gstanfor]

3DMARK03
7247 = Default benchmark
5978 = 1024x768 2x AA 8x AF
4593 = 1024x768 4x AA 8x AF
4750 = 1280x960 2x AA 8x AF
3489 = 1280x960 4x AA 8x AF

For reference purposes, my 6800GT (in a different PC remember) scores 10601 in the default test, 6078 @1280x960 AA 4x, AF 8x
Driver Settings 1 Driver Settings 2 Driver Settings 3

3dmark05
3593 = default benchmark
2702 = 1024x768 4x AA 8x AF
2954 = 1024x768 2x AA 8x AF
2196 = 1280x960 4x AA 8x AF
2443 = 1280x960 2x AA 8x AF

For reference purposes, my 6800GT (in a different PC remember) scores 4819 in the default test, 2794 @1280x960 AA 4x, AF 8x
Driver Settings 1 Driver Settings 2 Driver Settings 3

Final Fantasy XI benchmark
yes, I know, not on the list, but I like it.
Benchmark setup
The benchmark has no way of specifyin AA & AF, so I force them in nHancer.
2959 = 4x AA 8x AF
2970 = 2x AA 8x AF

For reference purposes, my 6800GT (in a different PC remember) scores 4785 for 4x AA 8x AF, and 4795 for 2x AA 8x AF.
Driver Settings 1 Driver Settings 2 Driver Settings 3

F.E.A.R.

1024x768 4x AA 8x AF
MIN 10
AVG 18
MAX 40
83% below
17% between
0% above

1024x768 2x AA 8x AF
MIN 13
AVG 27
MAX 60
50% below
36% between
14% above

1280x960 4x AA 8x AF
MIN 7
AVG 12
MAX 24
100% below
0% between
0% above

1280x960 2x AA 8x AF
MIN 10
AVG19
MAX 39
77% below
23% between
0% above

800x600 4x AA 8x AF
MIN 12
AVG 26
MAX 62
46% below
40% between
14% above

800x600 2x AA 8x AF
MIN 14
AVG 38
MAX 96
22% below
41% between
37% above

For reference purposes, my 6800GT (in a different PC remember) scores as follows

1024x768 4x AA 8x AF
MIN 17
AVG 35
MAX 88
18% below
58% between
24% above

1280x960 4x AA 8x AF
MIN 14
AVG 25
MAX 58
53% below
38% between
9% above

Driver Settings 1 Driver Settings 2 Driver settings 3

FEAR settings 1 FEAR settings 2

Half-Life2 CS:Source Video stress test
66.23 = 1024x768 0x AA 0x AF
66.66 = 1280x960 0x AA 0x AF
65.70 = 1024x768 4x AA 8x AF
62.19 = 1280x960 4x AA 8x AF
66.92 = 1024x768 2x AA 8x AF
64.60 = 1280x960 2x AA 8x AF

Driver Settings 1 Driver Settings 2 Driver Settings 3

In conclusion, I need to source a faster Socket A CPU and PC3200 for my spare rig - those two items are holding me back way more than the 6600GT is.

 

[dug777]

Wow, give me some to digest that I've got some more stuff to run , and then i'll clock back to 2400+

 

[dguy6789]

7800GTX 256MB vs 512MB of the same clocks:

http://www.anandtech.com/showdoc.aspx?i=2607&p=10

 

[Gstanfor]

I discovered I could use CS:S offline, so I've added some (rather strange) CS:S stress test results to my benchmarks post.

 

[rstrohkirch]

Gstanfor your rig link needs to be this

http://www.anandtech.com/myanandtech.aspx?member=4126


Also, I will say that 512mb cards CAN be of use right now, problem is map makers don't want to hurt performance for all those people out there with 128mb or lower cards...or even 256 in some cases.

You know much stuttering levels like this would get people with those types of cards in modern game engines?

 

[Gstanfor]

Done.

edit: I agree more memory on video cards is useful, so long as the memory is the same speed as a lesser amount. If it's slower, it's not a great deal of use.

 

[dug777]

Those CS:S results are whacked

well, i'm working on mine now...

 

[dug777]

Meh, same problem for me in CS:S Stress test, everything is between 70-72 FPS

 

[Gstanfor]

weird - but it is the source engine... and I don't get what was so stressful about their test either - perhaps it is a stress test for a GF2/Radeon 7200, not for much else though...