New Vid Card Thoughts...

[Hpt. Lisse]

Hey ya'll, just found this lovely chart at THG, with a hardware swap imminent...

Yeeowza

Click on the chart to increase the size... Note that the 6800GS is roughly equivalent to a 6800GT.

While my eVGA FX5900/BartonXP 2500+ (OC'ed to 2.1 Ghz & 360 FSB)/ 768 MB memory has worked well for two years, really large campaigns like CMBB's "Road to Ihantala" can bring a stutter to the landscape (boning up for CMC, oh yeah.) Small to Med. battles, even with 8xS AA turned on at 1280x1024, run pretty smoothly.

It'll be interesting to see where NVidia's mid-range stuff in the 7000 series ends up, performance-wise.

Is anyone here running CMBB/CMAK with a 7800GT or GTX?

[Jack Carr]

Herr Hpt.

For what its worth, I just upgraded my video card from a Chaintech GeForce Ti4600 4xAGP 128MB to a BFG GeForce 6800GT 8xAGP 256MB.

I see absolutely no difference whatsoever with CMBB and CMAK. I also added 512MB of RAM giving me 1.5GB. I upgraded for the same reasons that you are considering an upgrade.

A while back, I upgraded my motherboard and processor. My box is a bit behind like yours. I'm maxed out with an AMD Athlon XP 3200+, although I haven't done any overclocking on the processor yet.

I don't currently have any FPS games on my box, but I was thinking that the video card upgrade I chose would have helped those kinds of game.

I'm also hoping that my current configuration can handle the next generation of CM games and DropTeam.

Perhaps Battlefront will release the minimum specs soon.

Good luck with your upgrade.

[Redwolf]

Originally posted by Hpt. Lisse:

Is anyone here running CMBB/CMAK with a 7800GT or GTX?

No, I just wanted to mention that the 7x00 cards are PCIe only. There are no AGP variants. So for you it probably means full upgrade.

[Hpt. Lisse]

Yes, Looks like an NForce4 mobo, 2 GB memory, and a 3700+ San Diego processor. There were rumors floating around that the 6800GS was going to be released in AGP as well, but I haven't seen one yet. I may try to make the stretch for a 7800GT vid card.

But Jack Carr brings up a VERY interesting question - to make his kind of vid card upgrade and see zero improvement, that doesn't seem right. Does Sandra read your card as using the 8xAGP? What is your AGP aperture set to? Drivers? Did you have a chance to run 3DMark'03 before and after the new card? It would be quite notable if the 3DMark scores changed significantly, but you see no difference in CMBB/CMAK...

'Course, this isn't the first thread about "Where does CM bottleneck computers", either.

<community patiently waits for Schrullenthaft to throw in...>

[ December 26, 2005, 04:35 PM: Message edited by: Hpt. Lisse ]

[Jack Carr]

Originally posted by Hpt. Lisse:

But Jack Carr brings up a VERY interesting question - to make his kind of vid card upgrade and see zero improvement, that doesn't seem right. Does Sandra read your card as using the 8xAGP? What is your AGP aperture set to? Drivers? Did you have a chance to run 3DMark'03 before and after the new card? It would be quite notable if the 3DMark scores changed significantly, but you see no difference in CMBB/CMAK...

I never ran any benchmark software before or after installing the new card. It just seemed that there was no significant improvement visually or in speed with CMBB/CMAK. My previous configuration with the Ti4600 was really more than enough to handle CMBB and CMAK, so the 6800GT didn't make much of a difference. Anyway, that's my technical explanation .

I'm currently using the 77.76 drivers. I set the aperture in the BIOS to 8xAGP.

I'm hoping that the upgrades I've made are enough to handle the next generation of CM. Sounds like you're getting a 64 bit processor. A good move over the long haul.

Is it possible to get a shareware version of 3dMark or Sandra?

And remember, speed kills!

[Hpt. Lisse]

Today is your lucky day.

Sandra from Guru 3D - free

3DMark2003 - at guru

All shareware.

AGP aperture should read something like 32MB, 64MB, 128MB, etc.

Another test you need to do is run a "Huge" battle with plenty of units, and then report back if 6800GT improves your FPS or not. I loved my Ti4600 card, too, but you may not see much of a difference until you really tax the scenery, graphically speaking. It may look similar until you load up larger battles. Obviously, the vid card will not speed up the rendering of a turn, which is CPU-based.

Remember, too, that a Venice 3200+ CPU and NForce4 mobo runs well under $300 - if your current memory is PC3200, it'll swap right over... though you'll need a PCI-E vid card. One could still get an NForce3 mobo in order to retain AGP but upgrade to socket 939...

[Redwolf]

One reason why CM doesn't show that drastic improvements as other games with faster cards is that CM doesn't use any advanced feature that is supported in new hardware. In particular no fancy shaders are used, or any shaders for that matter.

When all what is done is push polygons around and paint textures on then a GeForce 4x00 is actually pretty competitive even with today's cards.

Jack, what exactly do you do, where do you want a speedup in CM? Is it when moving over the map, when switching views?

[Hpt. Lisse]

Redwolf - not buying it. Even without the advanced features, Carr's 6800GT is processing 16 pipes in one pass, where the Ti4600 is doing 4 (or my FX5900 is running 8). Certainly, alot of the subroutines per pipeline that has been engineered since the 4X00 line has to do with advanced features, but he's still pulling with 16 cylinders vs. 4 before. (BTW, the very fact that those fancy DX 9.0c calls aren't being used may only speed up vid card throughput.) Again, if he doesn't play larger than med. battles, he may not subjectively notice a difference... esp. if he doesn't use any AA/AF, which I do with the FX5900 in med. or smaller battles (at 1280x1024).

[ December 26, 2005, 10:41 PM: Message edited by: Hpt. Lisse ]

[Jack Carr]

Thanks for the links. I'll download them today. I checked the BIOS setting. I have the AGP aperture set to 256. The AGP transfer rate is set to 8x. My current motherboard is an ABIT NForce2 chipset. 400mhz bus, I think. Socket A for Athlon XP. Thanks for the suggestions.

Redwolf - I think at this point, I would like to see a pickup in speed when CM figures out the move. Scrolling over the battlefield is smooth as glass right now, even with all the doodads turned on.

[Redwolf]

Originally posted by Jack Carr:

Redwolf - I think at this point, I would like to see a pickup in speed when CM figures out the move. Scrolling over the battlefield is smooth as glass right now, even with all the doodads turned on.

Well, then you obviously have a fast enough video card and your processor isn't fast enough.

Which also answers the previous question.

Lisse, of course the GF4 is not as fast as a more modern card, even when just pushing around textured polygons. What I am saying is that a card that might be 10 times faster in modern games might only be 3 times faster in CM. Should have been pretty obvious.

BTW, the pixels pipelines in newest generation cards mostly do the shader calculations. You don't use shaders you don't get much out of more pixel pipelines. On a modern card the pixel pipelines will mostly be idle when displaying CM.

[Hpt. Lisse]

Sorry there, Jack, I had sotra prefaced my upgrade discussion by focusing on the choppy video, not the turn rendering times. As Redwolf says, that's completely CPU/FSB based.

Mostly idle, Redwolf, 'cause them fancy shader engines aren't being called on. But they're still rendering pixels. A 24-pipe 7800GTX is spitting out twice the amount of CMAK as a 12-pipe vanilla 6800. Believe you me, I was relieved (as we have similar rigs) when Jack commented on the smooth scrolling of his 6800GT. And it may even be fair to say (trying to claw towards "the obvious" here) that, past a certain point, there wouldn't be any discernable difference. Jack's 6800GT may well scroll as smoothly as a 7800GTX 512MB in CM, who knows.

Being that I'm in the market for a new build, I was just trying to get some feedback from CM players who have newer video cards. Jack's info leads me to believe that a 6800GS may be enough for now, instead of spending the extra $100 for a 7800GT. 'Course, the 16-pipe 7800GS are about to be released, but I'm not sure yet where it ends up on the price/performance curve.

[ December 27, 2005, 10:35 AM: Message edited by: Hpt. Lisse ]

[Jack Carr]

Yes, I am CPU bound. Not that I'm complaining though. As I said, my current CPU handles CMBB/CMAK well. The amount of money I would have to spend to pick up a few extra seconds figuring out what happens in a turn would be an absolute waste. CMX2 might be a different story though.

Redwolf, I believe in one of the CM forums you put a scenario out there for people to download and post their five best times. In retrospect, that was a good test. My times were not the best, but they were close to machines that were running big 64bit CPU's and high end Pentium 4's. AMD Athlon XP socket A is pretty much a dead platform, so I know I'll have to break down at some point. I'm just hoping I have a year or two.

Good luck and thanks for the info. I downloaded those benchmark programs today and will be checking them out this evening if I get a chance.

[Jack Carr]

I just got finished running the 3DMark03 software. My machine's score was 4034. Is that good or bad?

Specific scores were as follows:

Main Test Results

3DMark Score 4034 3DMarks

CPU Score 691 CPUMarks

Detailed Test Results

Game Tests

GT1 - Wings of Fury 123.4 fps

GT2 - Battle of Proxycon 25.8 fps

GT3 - Troll's Lair 22.3 fps

GT4 - Mother Nature 29.2 fps

CPU Tests

CPU Test 1 77.7 fps

CPU Test 2 12.2 fps

Feature Tests

Fill Rate (Single-Texturing) 1328.5 MTexels/s

Fill Rate (Multi-Texturing) 2678.9 MTexels/s

Vertex Shader 24.3 fps

Pixel Shader 2.0 66.2 fps

Ragtroll 17.9 fps

Sound Tests

No sounds 41.5 fps

24 sounds 36.4 fps

60 sounds Not Supported

* NO NEED TO REPLY *

I did some research and found 3DMark scores above 10000. My score is low.

[ December 28, 2005, 01:15 PM: Message edited by: Jack Carr ]

[Hpt. Lisse]

Jack -

I ran the test out of interest - please note:

Internet connection disabled, then real-time virus scan turned off - no AA/AF for tests in 1024x768 (default game test)

Also Barton 2500+XP OC'ed to 2.1 GHz and 362 FSB on an Epox 8RDA+ NForce2 mobo

eVGA FX5900 OC'ed to 438 MHz, and 926 Memory clock.

3DMark score - 5592

CPU Mark - 612

Tests:

GT1 - 167.1 fps

GT2 - 40.9

GT3 - 33.1

GT4 - 33.7

CPU tests

test 1 - 65.8 fps

test 2 - 11.3

Feature Tests

fill rate, sing. text - 1480.5 meg/text

multi. text - 3135.3

vertex shader - 17.9 fps

pixel shader 2.0 - 47.4

rag doll - 19.8

Sound tests

no sound - 37.9 fps

24 sound - 34.5

60 sound - 32.4

Hmm. I can explain about the sound - my mobo came with the MCP-T sound chip, which was significantly more powerful than the standard MCP chip found on most NForce2 mobo's. And your CPU tests are roughly in line. But your vid card is far better, and you have twice the system memory.

This is where Sandra can help you out - even the freeware version can analyze critical parts of your system, and see if everything is running the way it should be....

[Jack Carr]

Thank for posting your results. I did download and run Sandra and the most significant thing that it told me was that I had the page file size set to high. I lowered that value. It also told me that I was FSB bound. Nothing I can do about that. At least nothing that I know of without a motherboard upgrade.

Thanks again for the links to the benchmark software. I had fun toying around with it. I'll see what else I can tweak to improve my score. I'm a bit afraid to start overclocking components though, not only because I have no idea how to but also because its additional stress on the parts.

My motherboard has the MCP-T NForce2 chipset as well, but I am using a CReative Labs Soundblaster gamer 5.1 card.

[Redwolf]

Never buy hardware based on Sandra results.

The memory bandwidth benchmarks in Sandra have absolutely no resemblance to any kind of real-world performance except maybe a few cryptography applications (e.g. if you crack other people's passwords all day).

That doesn't mean you neccessarily are not memory bandwidth bound, it doesn't allow you to tell either way.

If you want to know how memory-bound you are, go to the BIOS and set your memory clock to something lower than you already have. I think you most likely run at 166 MHz right now. Set it to 133 MHz and compare the performance. If your computer still feels about the same way, and computes CM turns in about the same time, you were not memory-bound in first place.

AMD K7 and K8 CPUs generally don't care too much about memory speed for real-world performance.

[Jack Carr]

This is a bit better. I turned off Anti-Aliasing and Anistropic filtering and shut off my anti-virus (DUH!).

The results:

Benchmark Settings

Program Version 3DMark03 Revision 6 Build 0

Resolution 1024x768@32 bit

Texture Filtering Optimal

Pixel Processing / Antialiasing None

Post Processing false

Vertex Shaders Optimal

Main Test Results

3DMark Score 10914 3DMarks

CPU Score 729 CPUMarks

Detailed Test Results

Game Tests

GT1 - Wings of Fury 227.6 fps

GT2 - Battle of Proxycon 91.8 fps

GT3 - Troll's Lair 75.1 fps

GT4 - Mother Nature 59.9 fps

CPU Tests

CPU Test 1 81.4 fps

CPU Test 2 12.9 fps

Feature Tests

Fill Rate (Single-Texturing) 3089.1 MTexels/s

Fill Rate (Multi-Texturing) 5679.6 MTexels/s

Vertex Shader 30.7 fps

Pixel Shader 2.0 164.1 fps

Ragtroll 50.6 fps

Sound Tests

No sounds 43.6 fps

24 sounds 37.3 fps

60 sounds Not Supported

[TufenHuden]

EVGA 6800GS 256mb about $215.00 USC

AGP 8x or in PCI-E-only 12 pipelines/

with rivaturner you can unlock the other 4....

[Hpt. Lisse]

Much better, Jack...

Hey T, the way it falls down is this:

While limited to 12 pipes, the GS was manufactered on a smaller die process (110 for the NV42, vs. 130 for NV41), hence a big increse in core speed. Becasue of the new manufactering process, I don't think the GS can be "unlocked" - those 4 pipes may not exist on the NV42 chip. But I'm unsure of this - if anyone knows for sure, please post.

Here's a good article on it.

And look at the price difference to a 6800GT-

eVGA 6800GS (425MHz) - $190

XFX 6800GS (485MHz) - $205

eVGA 6800GT (350MHz) - $245

BFG 6800GT (370MHz) - $280

By that time, you're knocking on the door of a 7800GT... so I imagine the XFX 6800GS would probably outperform a standard GT, cost less and allow the move to PCI-E to be less painful on the wallet.

[Redwolf]

The unlock pipes trick also didn't work on any PCIe 6800 (no letters) (which is nv41), it only worked on the AGP version (nv40).

Before getting the 6800GT I'd upgrade 50 bucks and get a 7800GT instead - if you have PCIe.

[ January 19, 2006, 01:06 PM: Message edited by: Redwolf ]

[TufenHuden]

EVGA 6800GS 256mb AGP one has

DDR-3 ram and can be unlocked-NV40

chip....

[zdenka]

i recently upgraded from an xp3000/ti4600/1 gig of very cheap ram to an amd64 3700 san diego/7800gt/1 gig geil dual channel/asus a8n sli premium.

3d mark 03 score went from 1870 to 15390.i can play sturmovik with all settings maxed and get perfectly adequate frame rates but......... in combat mission i havent seen any difference what so ever.large maps with 10k+ forces still take ages to load,although i dont get any stuttering.

[Schrullenhaft]

The 'loading' of maps isn't a function of the video, it's a function of your hard drive (primarily). The 'loading' screen occurs at a point where textures are being read from the hard drive and loaded into video memory, etc. The hard drive is the slow part in this equation (compared to the memory, bus and videocard/memory).

[Hpt. Lisse]

Yeah, if you can scroll 'huge' maps full of high-res models without noticeable stuttering, then I wouldn't call that "no difference", 'cause your previous setup wouldn't do that...

[zdenka]

i must admit that i was unaware that the speed of the hard drive affected the game loading that much.but there again,i am a noob with regards the finer points of pc's.

Hpt.lisse, i did say large and not huge.