Information about CPU Limit

CPU Limit is an expression that in a PC if the CPU is a bit slow compared to the graphics card, it must wait for information from the CPU to work, and it will decrease the number of images per second (fps) in games.
It occurs mainly in low or medium resolution, 1280 and lower, as high resolution (1680, 1920), the graphics card has a lot of pixels to manage, and this phenomenon is fading.

Power PC CPU finds that the Limited is an important phenomenon, very marked in the middle and lower resolution, 1280 and lower, yet sensitive and high resolution, 1680 and 1920, while PC Inpact concludes that the phenomenon is not very important, declining performance in 1280 average real but is not sensitive at stake

Games or "suckers" are needed to get an idea of CPU limited, requiring just a game does not push the PC into a corner.
And also, Power PC takes into account the minimum number of frames / s, responsible for the lags.

Test procedure

So I decided to test with my 2 PCs running Windows XP (DX9) with the same screen, a 19-inch, 1280 x 1024 resolution:
- A Dell Dimension 8400 2005 (chipset motherboard i 925X) with a P4 650 at 3.4 Ghz, 2GB DDR2 533 Mhz, 1 slot PCI Express x16 v1,
- PC assembled in the spring of 2008, motherboard Asus P5KC, C2D E6850 at 3 Ghz, 2 GB DDR2 800 Mhz, 1 slot PCI Express x16 v1.

I mounted successively on 2 PC GeForce 8600 GT qualified entry-level, a 3850 HD characterizing the mid-range HD 4850 and a representative about the high end.

I've been running a test first synthetic 3DMark06, the overall score, but also results in frames per second of 4 games, because the overall score reflects the power of the CPU, thus distorting a little results.
Then I tried Crysis in high detail, without AA, the same scene.
This game is often used as a reference test configurations because it is very demanding in resources.
I played the demo of Crysis, free download, so even those who do not have the game can be tested on their PC and compare.

I found the following results:

3DMark06 Tests

GT 8600 PC P4: score: 4200; Proxycon: 15.5 fps; Firefly: 17.1 fps; Canyon: 16.4 fps; Deep Freeze: 20.2 im / s.
GT 8600 PC C2D: score: 4900; Proxycon: 16 fps; Firefly: 17.6 fps; Canyon: 16.8 fps; Deep Freeze: 20.6 im / s.

HD 3850 PC P4: score: 5700; Proxycon: 22.8 fps; Firefly: 21.0 fps; Canyon: 37.0 fps; Deep Freeze: 28.8 im / s.
HD 3850 PC C2D: score: 9500; Proxycon: 30.3 fps; Firefly: 36.1 fps; Canyon: 39.6 fps; Deep Freeze: 46.0 fps

HD 4850 PC P4: score: 5900; Proxycon: 22.8 fps; Firefly: 21.1 fps; Canyon: 43.1 fps; Deep Freeze: 29 im / s.
HD 4850 PC C2D: score: 11 500; Proxycon: 35.7 fps; Firefly: 44.7 fps; Canyon: 53.4 fps; Deep Freeze: 60.6 im / s.

It finds:

the 8600 GT is 3DMark06 score below 10% with the P4, but this is mainly CPU score, which is involved in the overall score, the results of games are equivalent (at 0.5 fps almost) with the P4 or the C2D.
It is not limited by the CPU in 2 cases.

The HD 3850 is significantly limited by the P4, the overall score is 40% lower, but again the score penalizes the P4 CPU, the results on the games being only 30% lower than with the P4 on all 4 games.

The HD 3850 is already a little CPU limited with a P4 in the order of 30%, which is quite something, even if the results are better with a 8600 GT (about + 50%) .

The HD 4850 is even more limited by the P4 because it gives the same results as the HD 3850 on 3 sets of 4.
This card offers virtually no gain in performance compared to an HD 3850, combined with a P4.

Testing Crysis playable demo

The settings of the game are:
1280, details high without Anti Aliasing,
The graphics cards are not overclockees,
Scene of sunrise over the bay, No. 19 (the objective is to destroy the jamming),
I get the following values, frames per second measured by Fraps over a period of about 360s:

GT 8600 PC P4: min: 1 fps, max: 16 fps, average: 10.8 fps;
GT 8600 PC C2D: min: 4 fps, max: 18 fps, average: 11.2 fps;

HD 3850 PC P4: min: 2 fps, max: 41 fps, average: 23.7 fps;
HD 3850 PC C2D: min: 13 fps, max: 45 fps, average: 29.3 fps;

HD 4850 PC P4: min: 3 fps, max: 48 fps, average: 24.9 fps;
HD 4850 PC C2D: min: 24 fps, max: 70 fps, average: 48.3 im / s.

It finds:

For the 8600 GT, the difference between P4 PC or PC C2D is low: 0.4 fps better on average.

For HD 3850, the difference is significant between the 2 PC, about 20% on the P4.
Despite this limitation, this card is interesting for a PC P4 compared to the 8600 GT, because we get 2 times more fps (23 instead of 11).

For HD 4850, there is virtually no point improvement over the P4 PC 3850, on a 1 image / s better!
This card gives the best of itself on the PC with C2D 3 Ghz, with an increase of 50% compared to 3850.

There is a minimum of frames per second significantly lower when the cards are on the PC P4, which is inconvenient for complex scenes.
HD 3850 allows to play Crysis on High at P4 PC with 24 fps average, but the scenes with a lot of players will be difficult to play: 2 fps instead of 13 fps on the PC C2D!

Influence of the resolution

Without larger screen, I did not measure in 1680 or 1920, or the phenomenon of CPU Limited wanes when we play in higher resolution, because the graphics card has more pixels per image to manage.

The magazine PC Update No 39 (January 2009) made measurements with Far Cry 2 in 1280, 1980 and 1920, comparing an E6400 at 2.13GHz (E2200 neighboring a current) with a E8500 at 3.16Ghz with the same graphics card, an HD 4850.
The results are as follows:
- E 6400 + HD 4850: 52 fps in 1280, 39 fps in 1680 and 21 fps in 1920
- E 8500 + HD 4850: 81 fps in 1280, 53 fps in 1680 and 25 fps in 1920

It shows that in 1280 the gap is large between the 2 processors, remains significant in 1680 and became negligible in 1920:
if we play in 1280 (19 "), a powerful CPU to take advantage of a powerful graphics card if you play in 1920 (24), a CPU input range may suffice.

Conclusion

The results speak for themselves, the CPU Limited "is important in 1280 x 1024 ...
It is still sensitive to 1680x1050, and becomes quite low in 1920x1200.

So be careful when you want to update your PC, change graphics card should be done cautiously and in the case of a very powerful graphics card, it will also change the CPU to make the most of his new purchase especially if we play in 1280 on a 19 ".

If the CPU can not be replaced by a really powerful, it will be limited in the power of the graphics card.

In particular, on a PC with a P4 as CPU, HD 3850 represents the maximum of the upgrade for the graphics card, more powerful cards cost more without significant performance gain on a 19 ".