VOGONS


First post, by framebuffer

User metadata
Rank Newbie
Rank
Newbie

OK silly title apart... 😁

I got this nice DELL C840 with a Pentium 4m 1.40GHz and a GeForce4 MX440 Go

dell-c840.jpg

And yesterday I clean installed Windows XP SP3 and drivers and did a quick 3DMark 2001SE, but the result surprised me

dell_c840_3dmark01se2.png

Basically it scored as a GeForce2 GTS paired with a Pentium4 HT 3.46GHz or a GeForce2 Ultra paired with a Pentium III Tualatin 1.40GHz

How is this possible?
The card is not even a full MX440 Go (2pp/4tmu 220/440MHz 64MB) but a 32MB version with RAM clocked at 200MHz.

https://framebuffer.io

Reply 1 of 5, by Namrok

User metadata
Rank Member
Rank
Member

Oh wow, what a blast from the past. Pretty sure that was the laptop I was sent off to college with in 2002. When the Doom 3 tech demo was leaked, I made a valiant attempt to play it on that thing.

More on topic, I found scattered benchmarks actually putting a Geforce 440 Go in the 5000 range in 3DMark2001. I think utilizing a better CPU, and more VRAM. And speaking of VRAM, I can't find a 440 version listed that only had 32mb of VRAM. But who knows.

Reply 2 of 5, by framebuffer

User metadata
Rank Newbie
Rank
Newbie
Namrok wrote on 2021-04-14, 15:52:

Oh wow, what a blast from the past. Pretty sure that was the laptop I was sent off to college with in 2002. When the Doom 3 tech demo was leaked, I made a valiant attempt to play it on that thing.

Oh I remember, I had a GeForce3 Ti200 and didn't went much better on it, was an early and very heavy release

Namrok wrote on 2021-04-14, 15:52:

More on topic, I found scattered benchmarks actually putting a Geforce 440 Go in the 5000 range in 3DMark2001. I think utilizing a better CPU, and more VRAM. And speaking of VRAM, I can't find a 440 version listed that only had 32mb of VRAM. But who knows.

Sure, you can score in the 5000 range with higher clocks, but what I don't fully understand about the C840 result is

1. If a 2pp/4tmu 220/400 (440Go) + Northwood 1.40GHz and 4pp/8tmu 250/460 (GF2 Ultra) + Tualatin 1.40GHz have the same score, this means that the Northwood is way much faster than a Tualatin at the same frequency! (?)
2. If a 2pp/4tmu 220/400 (440Go) + Northwood 1.40GHz and 4pp/8tmu 200/366 (GF2 GTS) + Pentium4HT 3.46GHz have the same score, this can only mean that Celsius cards are so much memory bandwidth limited that a CPU with almost triple the frequency can not do so much difference to make up for the 74MHz difference in RAM speed?
3. Isn't point 2. a contradiction of point 1. ?

https://framebuffer.io

Reply 3 of 5, by gerry

User metadata
Rank Member
Rank
Member

benchmark software is always intriguing and throws up unexpected things

for actual real experienced graphical performance in software, ie. in a game, so much depends on the interplay between CPU, RAM, FSB & Motherboard, storage, drivers, OS, Directx / opengl version, programming optimisations as well as GPU, GPU ram (amount & speed) and so on.

I'm impressed that benchmarks are often a good guide to how things will perform given all that variability, it would be interesting to challenge this machine to a game you know works well in a similarly scoring machine with apparently high specs

Reply 4 of 5, by auron

User metadata
Rank Member
Rank
Member

about 1:1 clock-for-clock vs. a p3 sounds about right for a p4 from what i've seen on 3dm2001, especially if this is a DDR system. it's a synthetic bench probably optimized towards sse2 and showing the newest platform in the best light.

and just something for consideration: GPU vendors were known to cheat in these synthetics and the scope of that might differ depending on driver versions. best would be not to overthink these results and try to run some actual game benchmarks.