1GHz CPU with GeForce3 Tis and GeForce4s

[Reverend]

GeForce3 Ti200, Ti500, GeForce4 MX440 & Ti4600 shootout with 1GHz CPU at 3DPulpit

Discussions and comments welcomed here.

 

[LittlePenny]

I did enjoy your usual calibre of variety, it made things much easier to put in perspective.

Altough, I think I may have missed something, because I reached a different conclusion than you. I thought it would be nice to have a GF4 in that machine because you can max out everything(FSAA included) and still get decent frames. Whereas if you had a GF3 with a faster CPU wouldn't the frames be much worse with FSAA on?

I dunno, you guys are the experts, I just blabber.

 

[mat]

thanks for this Article (mainly because i have an Athlon 1.2 GHz and im looking for a new Card) Rev!

I have two Questions:
Did you enabled Aniso. Filtering?
Is it possible to include a TI4200 Card (when its available) in this test? (I think a Ti4600 is too fast for this CPU, but a TI4200 could be just right)

 

[pascal]

Nice article

The only question are those games software T&L?
Probably with true hardware T&L the framerate will be much higher.
I have always be intrigued by the 3Dmark2001 car chase, the "low detail" have the poly count (around 170k) that U2 and Doom3 will have, and the high detail (around 350k) is just a good extreme test will perfect CPU scallability. With the "low detail" usually the 1GHz CPU is not the limit.

Also about CPUs there are a 1.4GHz P3-S spanking the 2GHz P4.

 

[Reverend]

LittlePenny, look at the SSamSE and MP scores... a GF4 really isn't that much faster than the other cards... so, yes, "decent" framerates is with a Ti4600 but the other cards aren't too far behind this "decent" framerates. With a faster CPU and a GF3, how can framerates be lower?!? A system with a faster CPU has to be faster, AA or no AA, even with a GF3.

mat, no aniso applied since the GF4 MX440 doesn't have the various aniso options as the other cards. No, I don't think I'll have a Ti4200 but I'm on the Ti4400 list.

pascal, SSamSE and MP supports HW T&L, dunno about F1 2001 to be honest.

 

[Doomtrooper]

A review like this is very uncommon these days, most people believe everyone is running with a 2ghz machine and that us simply not the case at all. The average CPU mhz rating on the Madonion Poll was still 75% slower than 1GHZ.
This shows you really need the TOTAL package to get the speed out of modern cards. I should hope a person that can afford a $700.00 dollar video card could afford a faster system than 1ghz but there is
some people that think a video card can grant time travel, or bring you three wishes

 

[LittlePenny]

LittlePenny, look at the SSamSE and MP scores... a GF4 really isn't that much faster than the other cards... so, yes, "decent" framerates is with a Ti4600 but the other cards aren't too far behind this "decent" framerates. With a faster CPU and a GF3, how can framerates be lower?!? A system with a faster CPU has to be faster, AA or no AA, even with a GF3.

Ok, it is obvious that I am wrong, or am just explaining my thoughts poorly. Most likely I am wrong, but I am going to try again.

It is my understanding that the situation with the 1.1GHz and GF4 the 1.1GHz CPU was the limiting situation. So wouldn't you be able to enable alot of higher end video card specific features like FSAA and aniso for a relatively cheap frame hit(As shown in the SSSE chart)?

But if you notice the GF3 does drop quite a bit when going to FSAA meaning it is no longer CPU bound. So if you upgraded that CPU, I really do not think those FSAA scores for the GF3 would rise too much. Are you catching what I am saying? Or am I just terribly off? I have class now, but let me know later if I need to add to this or retract my conclusion.

thanks for reading and commenting on my thoughts

 

[Entropy]

These kinds of data is certainly worthwhile.
Useful to lots of people thinking about where to spend their money, I'm sure. Very unfortunate though that data from the new AthlonXP2000+ box wasn't included. It would provide the data for the alternative money-spending path.

This statement shows an attitude/conlusion I've seen many times with this type of comparisons:

with a 1.1GHz CPU, 1024x768x32bits + 4xAA on the GeForce4 Ti4600 is definitely playable.

This is with just under 40 fps, which was also the result in Serious Sam.

I'd turn that around and conclude that todays' more demanding games cannot be played without serious performance problems on a system such as the one you tested. Those fps scores are pretty much exactly what I get on my old old iMac in Q3, and I would hesitate to call that experience "playable". Of course, you can play with arbitrarily bad framrate. You just won't play particularly well.

Still that system was a very nice one a year ago. Games designers often get accused of being too conservative on these messageboards, but looking at your results, I wouldn't really say that critique is valid overall.

Entropy

 

[Reverend]

LittlePenny, a CPU-limited situation is bound to the chosen resolution. You cannot really generalize too much. If I had benchmarked everything at 1600x1200, things would probably ("probably" becoz I haven't tried since it would be uselessly slow) be fillrate-bound even for a GF4 Ti4600. So, yes, if you only ever game at no more than 1024x768 on a system like the one in my article, sure, you'll get less of a hit if you enable AA and/or aniso. I say "less of hit" but that doesn't mean there's definitely no hit. In most of the benchmarks in my article, the GF4 Ti4600 is the only card that more or less maintains framerates with various AA compared to no AA while the other cards gets hit bad. This means it has lots (fillrate, bandwidth) to spare compared to the other cards but only at the chosen (1024x768) rez.

The short of it is that it comes down to the chosen resolution. You will always see an increase in framerates with a faster CPU (and I mean a meaningfully faster CPU, not a 700MHz CPU compared to a 500MHz CPU). The thing is the how much of an improvement. So, for a GF3, if you use 1024x768, going from a 1GHz CPU to a ,say, 1.7GHz CPU, you WILL see an improvement. The keyword, like you said it yourself, whether you'll see improvements "rise too much" or not.

 

[LittlePenny]

You cannot really generalize too much.

Caught me there

The short of it is that it comes down to the chosen resolution. You will always see an increase in framerates with a faster CPU

I stand corrected. Perhaps I was relying on the numbers a bit too much, just going by that Serious Sam SE 1024x768 graph, it did appear that when using QCAA the GF3 and GF4MX became the limiter, because they were no longer at 49.6 like they were all before FSAA was turned on.

I guess if I learned anything, it is that it will become more difficult for you reviewers to write something everyone can appreciate, since the feature you like most or focus on isn't something I might like or want to hear about. Too much qualitative versus quantitative.

thanks Rev

 

[pascal]

Reverend:
pascal, SSamSE and MP supports HW T&L, dunno about F1 2001 to be honest.

Thanks Rev, but how can a game with low poly count (much lower than 170k, probably around 30k) using hardware T&L (with fast GPUs like GF3Ti200 and GF4Ti4600) and a fast CPU (1.1GHz) be cpu limited ???

Unless the CPU is doing more than driving the GPU or something is missing (or not optimized) in this games. The frame rates are lower than some extreme tests like the Unreal 2 performance test (using multiple passes and very high poly count) extrapolated for 1.1GHz (my guesstimate 50fps). http://www.anandtech.com/cpu/showdoc.html?i=1595&p=10

 

[Reverend]

For SSamSE, the limiting factor is the use of complex shadows (8 lights, I think). I'm pretty sure this is custom lighting done by the CPU and not the GPU but I could be wrong. Not too sure what's up with MP though.

 

[pcchen]

One possibility is, it is not CPU limited but memory limited. Even if a game engine just does a simple BSP walking every frame, the data size could be larger than the L2 cache and makes memory bandwidth (or latency) important.