Another title could be “Everything you should know about the MX but nobody took the time to tell you”.
The real title should be “The variables affecting the frame rate of Quake 3 timedemo demo001 using the GeForce MX video card including, CPU limiting, memory bus, memory speed/ns, core/memory clock, drivers, and motherboard fsb, including the variable of individual motherboard performance with an explanation of the two basic versions of the MX”.
I got some emails with questions about CPU limiting and frame rates with Quake 3 timedemo. I replied saying I would put something together and submit it to Joe for consideration.
If you’re going to write an article about something, you have to find out what you don’t know. Getting reference data was like digging a bottomless pit: The more I dug the more I uncovered. It turned from interesting to boderline fascination.
I found memory chips from 7.5/ns to 3.5/ns.
I found core/memory clocks from 175/143 to 250/260.
I found Memory capacity from 16Mb to 64Mb.
All the retail boxes have a big MX on them. To me, it seemed like some were an obvious deception to prey on the uninformed, short on cash, wanting a decent performing value video card.
Another thing I found: There is a reference design for a MX 100 with a 32bit bus. I couldn’t find one, but it wouldn’t suprise me if one shows up in the market. There is also a 128Mb sample card but I can’t imagine it making its way to the market.
There is no shortage of MX reviews with everything from “This card is a phenominal overclocker” to “I couldn’t get one extra Mhz”. Another said, “The extra 32mb of memory in the MX400 didn’t make a bit of difference”.
The original MX was a winner from day one – it put new meaning into the buzz word “Value Card”. It also put some serious hurt on the VooDoo cards helping their demise. It seems every manufacturer and some I never heard of are trying to capitalize on the popularity of the MX.
There are two versions of the MX in the market: One with a 128/64 bit memory bus SDR/DDR and the other a 64 bit SDR memory bus. Everything else is just a variation of the two.
The one with the 64 bit bus is the kind you put into a box you’re building for Grandma to email the kids. I seriously doubt its value even for a casual gamer. I even found one with 64mb of memory. Everyone knows 64mb is better than 32mb right? Wrong, not with a 64bit bus – just an advertising gimmick to get your hard earned money.
DDR is another one. With a 64 bit bus used with DDR and two rendering pipelines, the MX cannot realisticaly take advantage of DDR memory. The NV11 core with good cooling can hit high clock speeds but this is not the determining factor. Memory speed in nanoseconds is what makes the difference in performance. Higher numbers/ns are slower and lower numbers/ns are faster.
There are several varibles affecting the performance of the MX and we’ll take at look at them one at a time.
Relevant System Data
- Duron 750 Unlocked
- FSB/Memory 100/PC133
- 384mb Memory
- Original Det. 5.32 Drivers
- 5.5ns SDRram
- Default Core/Memory Clock 175/183
- DirectX 8
- OS Win 4-10-98
- VSync Off
- Sound Off
The effects of CPU limiting on the MX in Quake timedemo demo001 can can be easily illustrated with the following chart.
As you can see, the first 250MHz gives a 17fps increase. The next 250 MHz gives a smaller gain of 10.4 fps. Higher CPU speeds over 1 GHz yield little in fps. Higher resolutions will lower the difference but it is still there.
The next chart shows the gain from fsb speed and overclocking the core/memory.
As you can see, the higher motherboard fsb yields an additional 8 fps. The Quake monster has a hunger for memory and that includes the motherboard bus. A good DDR motherboard with DDRram will give an additional increase. Overclocking the core and memory gives another 6 fps. Going from 128MB to 384MB system memory picks up a few fps. As you can see there are a number of variables determining the frame rate in Quake timedemo.
The effect of overclocking the card can be seen better in higher resolutions.
In 800×600 I got a 6 fps increase; in 1024×768 I get a 12 fps increase.
Drivers can also make a difference. Trying different drivers is a tedious install/uninstall process. Beta drivers can screw things up to the point you have to take them out in safe mode – a real pain. I was going to put something on drivers together but the install/unistall and doing the coolbits hack with every new driver wore me down. I’m going to continue to work with drivers at my leisure
Another variable is the coMBination of video card, motherboard, and operating system. A card may work great with one combination and not as well with another. One brand motherboard may yield higher fps than another with the same card and CPU all things being equal.
When I first got Quake, trying to find out how to run time demo with the Sound Off was a bit of a chore. Sound On can affect the frame rate because your also including your sound card in the test. In some cases it will make no difference. In others, you’re frame rate could be off five or ten frames.
Go to demos and hit the tilde key. This will bring down the console. Enter the following.
- timedemo 1 (Hit enter)
- s_initsound 0 (enter)
- snd_soundrestart (enter)
Quake will restart with the sound off. Click “play demo” and timedemo will run with the sound off. You can make changes in system setup and still run timedemo with sound off. In Win 98SE, I could just click exit and everthing would return to normal.
In Win 4-10-98, Quake would crash when I went to exit if I didn’t reverse the process by typing:
- s_initsound 1 (hit enter)
- snd_soundrestart (enter)
One last set of benchmarks with every last frame I could squeeze out.
Relevant system data:
- ASUS A7V133
- Duron @ 1.02GHz
- Fsb 145
- 384MB Corsair PC 150
- Prophet 2 MX
- Detonator 12.40 Drivers
- Vsync Off
- Agp Aperture Size 128MB
- Agp 4X
- Fast Writes Enabled
- Core/Memory Clock 215/208
- Mobo Via 4.31 Drivers
- Win 98SE
- DirectX 7
- Sound off
Most maunfacturers have several MX cards to choose from. This comes from the marketing end of the business. They try to appeal to different segments of the market. If they perceive one segment will buy a card with an extra 32MB memory, they will make one. If they perceive a fan on the heatsink will sell to another segment, they will make one (even if it’s on a MX200).
A few recognize the performance segment of the market with fast memory, a good heatsink and fan on the core. Pay your money and make your choice. After the GF3 MX hits the market, there may be some very good bargins in older MX cards if you are willing to do your homework.
With very few exceptions, marketing could care less about performance. All they care about is selling video cards. It’s their job and some do it very well. If deceiving the uninformed is part of it, they will do it. Business is business and the bottom line is profit. MX buyer beware.
After reading this I’m sure you can understand why the MX reviews give a wide variety of results. Mix in “Stroligo’s Law” and it’s even more difficult to get meaningful data. If I compared all the MX cards with all their variations, this would have been a book instead of an article. This is just an attempt to shed some light on the many flavors of the MX. Some are bitter and some are sweet.