One Last Hope...

[Hlafordlaes]

The setup, as you see, is nearly impossible to tweak, but I have done what I can:

Tyan PIII mobo - latest Award BIOS, allows some Northbridge tweaking

VIA Apollo Pro 133A - all 4in1 tested and using best version; using WPCRSET for further Northbridge tweaks; VLatency tested but not good on WinXPSP2. Board not stable at AGP4x, and Fast Writes are impossible on this chipset. (Soon will be looking into RT's Northbridge tweaks to reduce to one tool).

NVIDIA FX5500 - Have successfully used RivaTuner to try out all the NV34 and NV34B-based models; got Quadro emulation to work as per all the stickies here; nice improvement for OpenGL, but I stick with FX5500 for gaming. There is nothing else to unlock on the FX5500 according to what I read in these forums. All performance settings tweaked for "best" DirectX FPS.

Now, before you post and commiserate with one with such a lowly setup, there is one tweak I have tried and which gives great performance boosts in DirectX, but which crashes the Aquamark3 test when switching to the second scene (High Particle Count).

The Tweak: Force Pixel Shading to v1.4 and/or force Vertex Shading to v1.1

The Result: Stunning speed in first test scene, then machine does a hard reboot when the High Particle scene is initiating. (Same happens in MystV cutscenes if video setting is above minimum, regardless of my Pixel and Vertex settings.)

My Guess: My vintage 250W PSU gives up completely when the Video card takes a deep breath; or AGP is not configured well at a low level (in spite of my conservative AGP2x, SBA on, FW off setting).

The Question: I have read in some forums that a poor PSU can be the cause, and even one post claiming small FPS improvements for every ~50W increase (diminishing returns over ~510W).

Any thoughts? Any RT command line or Power User tweaks you might suggest I explore? Do I buy a new PSU?

- Hlafordlaes

[clokkevi]

Please excuse me for not being able to help you out..
I just noticed that you wrote that you got Quadro (NV34GL?) working.
Do you remember which driver version you patched and used?
Which OpenGL app was it that got improved performance when you tested it,
and was it running 100% stable over a long period without any freezes?

The reason I ask, is that while NV31 based FX 5600 is verified to work 100% stable
with patched 42.51 drivers as a Quadro FX 500 / NV31GL,
I have not yet got any reports about the same for NV34 cards.
At least the NV34 based FX 5200 is reported to freeze/crash while testing in SPECviewperf
using patched 45.28 / 42.51 drivers (the latter with NV34 id's added to nv4_disp.inf)

[Hlafordlaes]

Follow the forum stickie on this topic to the letter, and use something like DriverCleaner Pro before installing any new drivers (follow the program's Readme to see how-to). The stickie identifies which drivers for which chipsets, etc

PS Performance is marginally better as for gaming and 3d when using the card as a GForce, not a Quadro, at least on the NV34 series.

PPS My undestanding is that if you are making a different chipset into an FX500 you will definitely get errors - choose a Quadro version that corresponds to your chipset, as per the stickie.

[clokkevi]

Thank you for your quick reply..!

When I in July 2004 first wrote GeForce 4/FX -> Quadro 4/FX : The complete how-to
I wrote that GeForce 4, GeForce FX 5600 (NV31) and GeForce FX 5200 (NV34)
should use 42.51, all other GeForce FX 5xxx should use 45.28.

Then, not long ago, I got posts about people stating that the FX 5200 soft-modded to NV34GL / Quadro FX 500
was not stable, it freezed, crashed, had weird artifacts etc.
So I then removed GeForce FX 5200 from the list, and instead wrote that
"GeForce FX5200 (NV34) can not be modified.

(But FX 5600 soft-modded to NV31GL was stable with 42.51, though)

So when you wrote that you was successful in modding your FX 5500, also a NV34,
I was just wondering if you had used something other than 42.51 (or 45.28)
and also which program you had tested it with, and for how long.
Basically, if you had "proof" (as in SPECviewperf results) that NV34, at least FX 5500,
really is stable enough to be used with Maya, Solidworks, etc..

Since I only have a FX 5900 and a 6800, I can not test these other cards myself,
so I'll just add/remove NV34 cards from the list based on others experiences.

[Hlafordlaes]

Well, I certainly didn't realize with whom I was speaking! Imagine me refering an author to his own stickie. Whadda way to get started here!

Anyway, yes, I used the referenced 45.28 drivers originally as indicated for the FX5500 (mine is a 256Mb model if that is of any import). But right now I am on the highest RT-database release (as I understand it to be), which is 82.10.

Currently I switch between this and the 5200 Ultra for testing, but in spite of lots of nailbiting, there are never important performance diffs, just a fancier XP-Help style menu as a legacy of the Ultra install.

I don't switch to Quadro now on this driver, of course. When I tested using 45.28, I only went thru the Acquamark3 test, with a slightly lower score than my FX efforts. I was tempted to "stay there" with an FX 500, but as an incipient speed-junkie (if you can use my gear and the word "speed" in any proximity), I decided to go back to pure FX tweaking.

Now, I might be tempted into re-installing a Quadro driver and downloading/doing the SPECviewperf test... if you might take a momentto consider my original question and give me your best shot or guess (ie, do I have a PSU problem if after many hours of stability, I (a) force 1.4 shaders or (b) use a high-res setting on MystV cut-scenes and get a hard reboot as a reward?)

[clokkevi]

OK.. unfortunately is the Aquamark3 test only using DirectX modes, so even if your 45.28-patched FX5500 was stable in that,
it could have been that it had got issues if it had been running in OpenGL instead..

If you got the spare time with noting to do one day, yes, feel free to try
SPECviewperf test in Quadro FX 500 and GeForce FX 5500 modes..
..but if you don't, it's no big deal, I'm sure there will appear other NV34 users here..!

About your original question:
I tried that Aquamark3 test to see what it was about, and I must admit, I was impressed!
That "Details Pass 01.txt" is cotaining everything for each single frame rendered..
..I can understand all the fuzz about it being a very good benchmark now!

I also tried to set "Pixel shader settings: force 1.4" and "Vertex shader settings: force 1.1" in RivaTuner,
but as I expected the speed to increase, I got a surprise:

With default settings, I got
AvgFPSRender: 87.362198
AvgTrianglesPerSecond: 17861802
MinTrianglesPerSecond: 3024769
AquamarkScoreRender: 8735
AquamarkScoreSimulation: 9246
AquamarkScore: 59339

With force ps1.4 & vs1.1, I got
AvgFPSRender: 77.003754
AvgTrianglesPerSecond: 17984764
MinTrianglesPerSecond: 5046422
AquamarkScoreRender: 7700
AquamarkScoreSimulation: 9020
AquamarkScore: 53975

So the "Triscore" was decreased, but the AvgTrianglesPerSecond was approx. the same.

Your score was increased by doing this, mine was decreased.
I guess the reason is the poor perfomance of Shader Model 2.0 in DirectX 9.0 on *all* the FX cards, and
"The NV30, NV31, and NV34 also were handicapped because they contained a mixture of DirectX 7 fixed-function T&L units, DirectX 8 integer pixel shaders, and DirectX 9 floating point pixel shaders."
In contrast, my NV40 may be more optimized for SM2.0 than it is for SM1.x,
and so the performance drops when forcing SM1.x

Then, about your theory...:
It is, like, when using ps1.4 & vs1.1 in Aquamark3, the pixel / texture / vertex units are,
maybe for the first time in their life, performing at their maximum rate, all of them at once?
And when this happens, the card is suddenly using.. how much? 15W-20W?
I didn't find anything for FX5500, but the FX5700 at 425/550 MHz uses 25W.

My guess is that, unless there is something physically wrong with a certain unit on the card,
your lack of power theory is plausible.
I would have tried to borrow a 350W+ PSU from a friend for a couple of hours.
If the Aquamark3 tests then goes as they should, I would have bought a new PSU.
If it still fails, I would have instead have taken the FX5500 to the friend and tried it together with his PC..!

[Hlafordlaes]

OK, a deal is a deal. You answered my query, now I gotta do the test.

I have a hitch there, tho. When I go to the site, I find a 410MB download I should do without any accelerators. With or without these, my humble connection would day the better part of a day. Tried looking for just one GL test from the suite (Maya), but it looked like the download required owning a licensed copy.

Know of any bittorrent links to the suite? Or another test that would indicate the mod is operative?

Hlafordlaes

PS - When I tested the Quadro version the first time, I did also use a real simple OpenGL test called ParticleFury145, which showed lotsa improvement, and I did get the professional menus to open up properly. Still, I want to see if I crash in the SPECviewperf or equiv tests.

[loop29]

Could you pls post a graphics subsystem diagnostic report with mod applied? Should give us an idea if it is working like it should. On the other hand you might use the open gl extension viewer from realtech VR if all opengl extensions are available. Another option is to use a opengl application package like maya, I always got instant freeze with FX cards when the mod didn´t worked.

regards

[clokkevi]

ftp://spec.ir.miami.edu/dist/gpc/opc/viewperf/
ftp://spec.unipv.it/dist/gpc/opc/viewperf/
ftp://ftpsv1.u-aizu.ac.jp/pub/spec/d.../opc/viewperf/
are mirror sites where you may use a standard FTP client, and they supports resuming etc.

Also, there is no need to take 8.1 at 434.911.502 bytes...
specviewperf612is01.exe at 55.863.930 is equally good.

What is interesting, is to see:
* Will it run all the tests without freezing in Quadro 500 FX mode?
* What are the results in Quadro 500 FX mode, compared to GeForce FX5500 mode?

[Hlafordlaes]

Originally Posted by clokkevi

ftp://spec.ir.miami.edu/dist/gpc/opc/viewperf/
ftp://spec.unipv.it/dist/gpc/opc/viewperf/
ftp://ftpsv1.u-aizu.ac.jp/pub/spec/d.../opc/viewperf/

What is interesting, is to see:
* Will it run all the tests without freezing in Quadro 500 FX mode?
* What are the results in Quadro 500 FX mode, compared to GeForce FX5500 mode?

I did the Quadro mod as before and tried the Specviewperf and realtech VR tests, and got screen freezes in both quite near the start of either test. ("Who was that masked man?") No sense reporting numbers; none were to be had.

I am now sadly reduced to the realization that NVStrap can't do it for me (ie open an avenue to better or alternative performance), since the NV34 is such a loser. Well, it was fun pretending!

My only recourse is to start getting a grip on the DirectX and other mods available in Rivatuner's Power User section. Tho I have read all the docs and faqs, and pay attention to the right-click help msgs, there is a ton of stuff in RT that I just don't know how to use.

Any links to documentation (RT or non-RT) on some of the more arcane settings? Your earlier post mentioned the mish-mash of DirectX functions on the FX cards; perhaps I might tweak more finely with this in mind with access to some in-depth info.

Hlafordlaes

[PS: I am now in the market for a video card upgrade. I drool at the recent Bliss 7800 AGP, but cannot afford. Any older models still on the market that are emminently tweakable with RT?)