I have not found any software that can reliably differentiate between memory bus widths.

With that said you should test your cards again, especially the MX 440SE looks wrong. Try running powerstrip after installation and make sure AGP is working correctly (DiME transfers). Also force Vsync off and re-apply Quake III settings after launching it with a new card.

SDR 333MHz is not a thing

AGP 4x vs 8x is irrelevant for this card

I would think riva tuner is reliable at detecting the card bus/memory

also check the memory chips, manually if in doubt.

That's interesting, could you please upload a picture of the cards?
The SE version is a stripped down version of a budget card, so no surprise it runs so bad.

lost77 wrote on 2021-02-13, 17:51:

I have not found any software that can reliably differentiate between memory bus widths.

With that said you should test your cards again, especially the MX 440SE looks wrong. Try running powerstrip after installation and make sure AGP is working correctly (DiME transfers). Also force Vsync off and re-apply Quake III settings after launching it with a new card.

Before running benchmarks I typically turn vSync turned off in the nvidia panel (as I have noticed that swapping cards turns if back to "application controlled"
So it is turned off, and I do see the q3 demoo01 go above 60fps (albeit rarely) during the run of the slowest MX440SE card.

However, It only seems to be with quake3 that I am seeing this performance hit ....
Also did the Unreal Tournament demo (getting 80-90 fps in the cityintro at 1024x768 on all 3 cards) and the Expendable benchmark (90+ fps at 1024x768 on all 3 cards)

So went to look for the quake 3 settings and noticed that Color Depth and Texture Detail were set to "Default".
I'm guessing "Default" here means whatever your windows display adapter is set to (16 bit or 32 bit colors).

I noticed that the MX440SE had its display settings set to 32bit. After I set it to 16 bit the fps moved from 42fps to 74fps.

So now I get :

• MX440SE - GPU 250MHz - 64MB SDR 128bit @ 333MHz : 74fps
• MX440 - GPU 230MHz - 64MB DDR 64bit @ 261MHz : 98fps
• MX440 - GPU 274MHz - 128MB DDR 128bit @ 513MHz : 158fps

But still a bit gap don't you think between these 3 cards in Q3 ? (especially the one hitting 158fps). Don't understand why it's only in Q3.
Other interesting benchmarks I can try ? (I prefer to run game benchmarks)

SPBHM wrote on 2021-02-13, 18:16:

SDR 333MHz is not a thing […]
Show full quote

SDR 333MHz is not a thing

AGP 4x vs 8x is irrelevant for this card

I would think riva tuner is reliable at detecting the card bus/memory

also check the memory chips, manually if in doubt.

Found some references to the 333MHz clock here :
http://www.vgamuseum.info/index.php/component … orce4-mx-440-se

And also saw it in PowerStrip.

RetroSpector78 wrote on 2021-02-13, 21:52:

Found some references to the 333MHz clock here : http://www.vgamuseum.info/index.php/component … orce4-mx-440-se […]
Show full quote

SPBHM wrote on 2021-02-13, 18:16:

SDR 333MHz is not a thing […]
Show full quote

SDR 333MHz is not a thing

AGP 4x vs 8x is irrelevant for this card

I would think riva tuner is reliable at detecting the card bus/memory

also check the memory chips, manually if in doubt.

Found some references to the 333MHz clock here :
http://www.vgamuseum.info/index.php/component … orce4-mx-440-se

And also saw it in PowerStrip.

check the memory chips and google the code to know what exactly is the ram installed if possible (and in general 8 chips = 128bits, 4 =64 with these cards) my guess is that it's actually a SDR card that it will end in -6 and run at 166MHz, and perhaps riva tuner will give you the right number, the link you posted shows a GPUz screenshot, I found that for old cards GPUz is at times giving bad readings (I think it gives twice the real memory bandwidth for my FX5900 also for example), so if it's based on that it could be wrong,

if your card actually had 5.2GB/s ram and the NV17 chip it wouldn't be that much slower than the other 2, so something is clearly wrong with that info

Throwing in some data which might be interesting, or not. I previously posted about terrible results from my MX440 PCI in another topic of yours: Direct3D gaming with a PCI videocard on an AMD K6-2 platform on windows 98se

This weekend I've dropped it into a PIII rig, and now I'm seeing some performance (1987 vs 6223 in 3Dmark99). I've just grabbed the stats:

1                No-name low profile MX440 PCI
2
3Bus             PCI
4GPU Clock       146MHz
5Mem             64MB
6MemType         DDR
7MemBus          64bit
8RealClock       79MHz
9EffectiveClock  159MHz
10Bandwidth       1272MB/s
11Q3 1024 dem1.   *

Pretty ordinary compared to any of yours!

Unfortunately I haven't been able to log FPS from the Q3 demo as it's not appearing in the console at the end. But I'm watching it fluctuate between 40-90 and would guess around 70 as an average. (I realise the differences in our systems make this comparison meaningless anyway).

I have not found any software that can reliably differentiate between memory bus widths.

RivaTuner.

Unfortunately I haven't been able to log FPS from the Q3 demo as it's not appearing in the console at the end

timedemo 1

Found some references to the 333MHz clock here :

They've just copypasted this info from another wrong source. SDR SDRAM/SGRAM clocks officially never exceeded 250 Mhz on video cards. Some utilities "interpret" 128-bit SDR 166 Mhz memory as DDR 333 Mhz memory, because MX 440 officially never had SDR memory listed.

The Serpent Rider wrote on 2021-02-14, 02:17:

timedemo 1

Thanks, had my procedure all wrong.

Pierre32 wrote on 2021-02-14, 00:33:

guess around 70 as an average.

70.9 (arsey bastard!)

Pierre32 wrote on 2021-02-14, 00:33:

70.9 (arsey bastard!)

Q3 timedemo ? On what CPU / resolution / card ?

RetroSpector78 wrote on 2021-02-14, 10:33:

Pierre32 wrote on 2021-02-14, 00:33:

70.9 (arsey bastard!)

Q3 timedemo ? On what CPU / resolution / card ?

Yep. PIII 1GHz, 1024x768, card:

Pierre32 wrote on 2021-02-14, 10:55:

Yep. PIII 1GHz, 1024x768, card:

IMG_20210214_214538.jpg
IMG_20210214_214556.jpg

On my P3 550MHz with my Prolink MVGA-NV617GA MX440 (128bit with 128MB DDR mem) I am getting 59fps at 1024x768x16bit 😀

Weird thing however is that on all resolutions (640 x480, 800x600, 1024x768) and all color depths (16bit, 32bit) the timedemo always finishes with 59fps.

I do have vsync off and I can see during the demo that it goes way beyond 60fps from time to time. But on all resolutions it averages out at 59fps.

SPBHM wrote on 2021-02-13, 23:46:

check the memory chips and google the code to know what exactly is the ram installed if possible (and in general 8 chips = 128bits, 4 =64 with these cards) my guess is that it's actually a SDR card that it will end in -6 and run at 166MHz, and perhaps riva tuner will give you the right number, the link you posted shows a GPUz screenshot, I found that for old cards GPUz is at times giving bad readings (I think it gives twice the real memory bandwidth for my FX5900 also for example), so if it's based on that it could be wrong,

if your card actually had 5.2GB/s ram and the NV17 chip it wouldn't be that much slower than the other 2, so something is clearly wrong with that info

Looked up the part numbers :

• VANV17L1PS MMX440SE : V-Data VDS6616A4A-6 : DDR Synchronous DRAM (1M X 16 Bit X 4 Banks)
• Prolink MX440-8x : Samsung K4d28163hd-tc40 : "128mbit DDR SDRAM" (2M x 16Bit x 4 Banks)
• No-name MX440-8x-03913 : V-Data VDD8616A8A-6B : Synchronous DRAM(1M X 16 Bit X 4 Banks)

Is it purely the amount of chips you have on the board that determine if you have a 64bit or 128bit bus ? (Because Samsung mentions it explicitly in their datasheet)
Would removing 4 of the chips in the Prolink turn it into a 64bit 64mb card ?

And does that mean that the ANV17L1PS MMX440SE that was detected as an 128 bit SDR card is in fact an 128 bit (8 chips) DDR (ram chip datasheet) card ?

RetroSpector78 wrote on 2021-02-14, 17:06:

Looked up the part numbers : […]
Show full quote

SPBHM wrote on 2021-02-13, 23:46:

check the memory chips and google the code to know what exactly is the ram installed if possible (and in general 8 chips = 128bits, 4 =64 with these cards) my guess is that it's actually a SDR card that it will end in -6 and run at 166MHz, and perhaps riva tuner will give you the right number, the link you posted shows a GPUz screenshot, I found that for old cards GPUz is at times giving bad readings (I think it gives twice the real memory bandwidth for my FX5900 also for example), so if it's based on that it could be wrong,

if your card actually had 5.2GB/s ram and the NV17 chip it wouldn't be that much slower than the other 2, so something is clearly wrong with that info

Looked up the part numbers :

• VANV17L1PS MMX440SE : V-Data VDS6616A4A-6 : DDR Synchronous DRAM (1M X 16 Bit X 4 Banks)
• Prolink MX440-8x : Samsung K4d28163hd-tc40 : "128mbit DDR SDRAM" (2M x 16Bit x 4 Banks)
• No-name MX440-8x-03913 : V-Data VDD8616A8A-6B : Synchronous DRAM(1M X 16 Bit X 4 Banks)

Is it purely the amount of chips you have on the board that determine if you have a 64bit or 128bit bus ? (Because Samsung mentions it explicitly in their datasheet)
Would removing 4 of the chips in the Prolink turn it into a 64bit 64mb card ?

And does that mean that the ANV17L1PS MMX440SE that was detected as an 128 bit SDR card is in fact an 128 bit (8 chips) DDR (ram chip datasheet) card ?

VDS6616A4A-6 is SDRAM, not DDR
if the card has 8 chips that's 8x16bit chips and since it's SDRAM the "effective clock" is the real clock which is 166MHz, so the real bandwidth is 2.7GB/s

if you have 16bits chips which is typically the case with these cards AFAIK
yes, 4x16 = 64bits, 8x16 = 128bits

SDR memories are not double clocked like DDR ones and some versions of GPUZ forgot about it. Thats why it shows 333MHz SDR memories, but in reality its only 166MHz. Thus also bandwidth is wrong and info on vgamuseum site too as owner of cards didn't realised, that its bug in GPUZ. You might try also HWinfo software for checking of cards. Btw dont think that anyone used SDR memory at clocks over 200MHz, so if its over this number, you can be sure that its detected like DDR and you need divide number by 2.