The CPU/Memory VERSUS GPU Debate! Is Money Being Wasted?

[Neil]

Hi Guys!

I was hoping you can help me with an experiment.

Here are my system specs:

AMD Athlon 64 X2 4600+ (Socket 939)
4GB RAM
XP Home SP2 32 BIT
XFI Xtrememusic Sound Card
8800GTS 512MB 670Mhz/1.94Ghz RAM

Sharky and I had a bit of a debate a few weeks ago about the importance of the GPU versus the whole desktop system. I think that aside from loading time, the CPU and RAM speed is overrated, and offers limited benefit to game play once you have a good GPU. He was of the opinion that fast RAM is critical, and a good CPU can make all the difference.

Using the latest version of 3DMark, can you post your system specs and your scores? I'd be interested in seeing if better GPU's out weight better desktop machines with inferior GPUs. It will be interesting to see if there is a trend.

Conveniently, Sharky can't install 3DMark for some reason, so I need your help to settle the bet.

If you don't have specs to share, post your opinion!

In fact, I'll run a poll to get things started.

Regards,
Neil

[crim3]

GPU and sound hardware ( I saw a review about how fps could rise just changing the sound card, a lot cheaper way of improvement ) are the critical components for gaming. GPU really makes the difference.

Of course, at some point, the rest of the system will become a bottleneck to the GPU, which won't be running at full throttle. So you just can't put the latest video hardware (it's a waste), but you can put one that match the rest of the system and see a great gaming performance.

Well... maybe, nowadays game physics are becoming more complex and are getting the CPU pretty busy, and dedicated hardware is still anecdotal. Now I'm not so sure about I have just voted. Does 3dMark benchmark game physics?

I have fear of what can do the latest 3dMark in my rather old system, I hope it won't get injured by the effort

[Neil]

The best I got so far was 8,776 marks from 3DMark 2006.

My specs:

Windows XP Home SP2
EVGA 8800GTS 512MB 670Mhz Core, 1.94Ghz RAM
4GB RAM DDR 400 (PC3200)
A8N SLI SE (Socket 939)
AMD Athlon 64 X2 4600+


If I'm correct, and if our GPUs are the same, most people should do much better than me because my CPU/motherboard is not top of the line. In theory, if your GPU is lower than mine, (let's say an 8800GTS 640MB, or 320MB), your CPU/memory should still blow me out of the water - but I don't think it will.

If Sharky is correct, even with your inferior GPUs, you should still do a lot better than me.

Now, I'm just interested in gaming speed. Not CPU stuff like video compression calculations, etc. 3DMark may not be the best measuring stick, but it's a start.

Let's see how things turn out.

Regards,
Neil

[Jahun]

Hmm no 3DMark here atm but there is no super clear cut answer, even 3D Mark itself is not perfect. It doesn't compare directly to all games..
It really depends on the game involved and the specific equipment.

A good GPU is paramount ofcourse.. but with recent drops in CPU and RAM prices it is a no brainer to get a nice CPU and some RAM to match. I don't believe in the added speed of DDR3 yet.. compared to the cost that is.

I work on a X2 Athlon at my job, and have a QX6700 Intel here at home (actually I have also 4 Q6600 machines too..) It is overclocked to 3.4 ghz (the other 4 to 3.6 :p) but it works alot smoother than the X2 in my usually overcrowded desktop.. I never played games on the X2, but I am convinced that your 8800 GTS would benefit from a better CPU.. You wouldn't need an extreme version ofcourse..

There should be reviews around that compare cpu speeds.. with games, while keeping GPU the same... lemme look that up..

[Neil]

Wait!

I know if the GPU is the same, there will be an improvement, no doubt.

But will a $100 increase in GPU power = $600 of CPU/memory power? That's the question.

Regards,
Neil

[LukePC1]

The crysis CPU benchmark would be a good compromise. It challenges both, CPU for the physics and GPU for all the graphics. I think the demo has the benchmark built in.
The CPU's should be all right, if you put physics to maximum and the rest to medium.

I'll try to run some benchmarks, but my system is even lower than Neils
But if the gpu is of such great importance he should get at least 2-3 times my score (I found an equivalent to mine for 65€ at ebay. The cheapest 8800 is over 200€ --> at least 3 times the price).

edit:
3D mark 06 (with a lot of open windows and stuff in the background, incredible 1684 3D marks. My system was the slowest in it's category
That might be due to the fact my card is always displayed as a 7900GT/GTO - but runs at 7900gs clock (or below that?)

NVIDIA GeForce 7900 GT, 468 MHz / 603 MHz

Main Test Results
3DMark Score 1684 3DMarks
SM 2.0 Score 529 Marks
SM 3.0 Score 696 Marks
CPU Score 1560 Marks

The result is common for 7600gs or 6600GT!

All Results with Similar System Your rank: 278 out of 278. You scored better than 0% of all similar systems.

New record

[Neil]

How do you access the Crysis benchmark? That would be a good working sample.

Regards,
Neil

P.S. Sorry about 3DMark's under appreciation of your video card. I think software gets snobby over time.

[Maggi]

Hi Guys !

I wished, it would be all that simple ...


About every game engine depends on different high performing system components, hence you'd have to look at four basic components:

• CPU
  GPU
  RAM
  Sound

The reason behind it is the concept of the game, ie. if it is stuffed with physics and AI, you need a fast CPU and RAM, if it is bound to post processing eye candy and thus requires raw shader power, a fast and powerful GPU would result in greater benefit than a CPU and/or RAM upgrade.

I could live with 3D Mark 2006, but as it was already noted before, the sound system should not be neglected at all, because a lot of systems are using their built-in onboard sound chip, which should be ok in a lot of situations, but on the flip side of the coin, they can eat up a lot of framerates when being overloaded, hence we should at least also use 3D Mark 2003, since it is the last version that offered a sound test as well.

One example about sound card performance would be my onboard 7.1 Realtek soundchip, but unfortunately, I don't have the exact numbers with me, so I'll have to look them up when I'm back home, but from the top of my head, I observed the following:

The soundchip is ok when using only 16 audio channels, resulting in an average CPU utilization of 1-2%, but as soon as I exceed those 16 channels, the chip started eating CPU cycles beyond 15% ... obviously, that'll cap the whole gaming experience.


So for a valid evaluation, you must not rely on one single benchmark and especially 3D Mark should be widely known for not representing real world's gaming experience.


Instead, we should make an agreement on more different benchmarks, in order to get a more accurate and general performance average, so I start a list of benchmarks that I'd like to suggest to give a more complete picture about what components are affecting performance:

• Synthetic Benchmarks:
  3D Mark 2006 (multicore and PS3.0 support)
  3D Mark 2003 (sound test)
  RightMark 3DSound

• Real World Benchmarks:
  Doom3 (OpenGL, stresses all components)
  Quake4 (OpenGL, modified Doom3 engine, DualCore support)
  Crysis Demo (D3D, separate CPU & GPU benchmarks in the game's Bin32 folder)

Lastly, we should also look at Mono3D vs. Stereo3D and how enabling our beloved real 3D impacts the gaming experience, depending on system configurations.

Let me know what you think about my suggestions ... 8)

Cheers,
Maggi

[Neil]

I own Doom 3 and Crysis, so both are good.

We should also turn our sound cards off so it is not an interfering factor.

How do you access the Crysis benchmark again? Is it in the demo?

Neil

[Maggi]

I own Doom 3 and Crysis, so both are good.

We should also turn our sound cards off so it is not an interfering factor.

not at all ... or do you play any game without sounds ???

How do you access the Crysis benchmark again? Is it in the demo?

read my post again ...

Crysis Demo (D3D, separate CPU & GPU benchmarks in the game's Bin32 folder)

8)

Maggi

[Neil]

Found it!

The reason I suggesting cutting out the sound card is depending on CPU/on-board sound cards (and even some on-board cards use depend on the CPU), that could hinder our results.

We are just trying to determine if one step in GPU performance is the same as hundreds of dollars of upgrades in CPU/memory as far as game performance goes.

Regards,
Neil

[Maggi]

We are just trying to determine if one step in GPU performance is the same as hundreds of dollars of upgrades in CPU/memory as far as game performance goes.

I see, but for that single purpose, you could just have a peek into Tom's Hardware's CPU and VGA Charts ...

http://www23.tomshardware.com/graphics_2007.html

http://www23.tomshardware.com/cpu_2007.html

Cheers,
Maggi

[Neil]

That won't work.

The systems are too consistent. In the case of the CPU scale, all the GPUs are the same. In the case of the GPUs, all the CPUs are the same.

When the environment is consistent, the better components will always show a measurable improvement.

In this case, inconsistency helps us. Can we show that a good GPU with a mediocre motherboard combo will outperform a fantastic motherboard combo with a less powerful GPU?

Regards,
Neil

[Maggi]

ok, now I understand ... 8)

Thanx for clearing that up !

[Freke1]

I have some experience with this subject a few years back:
(Aquamark benchmark)

The same 6800GT card in 2 different pc's:
old pc (2500Mhz) new pc (3700Mhz)
CPU 5000 CPU 10000
GPU 7500 GPU 10000

2 different (6800GT and 7900GT) cards in same new pc:
6800GT 7900GT
CPU 9500 CPU 10500
GPU 10000 GPU 12000

Basicly:
1) upgrading the CPU and RAM: doubled the fps in games
2) upgrading the graphics cards: doubled the fps in games

gear: 2500MHz, 3700MHz CPUs (single core) and 6800GT, 7900GT graphics cards.

[Neil]

Well, I don't know if the Aquamark rating will help us because it probably rates the CPU/memory speed and GPU speed as separate elements and adds them together.

If we use the Crysis example, and assume your work is correct, a faster CPU/RAM and faster GPU will double or quadruple the FPS.

I think it is more likely going to be a faster FPS with the GPU, and maybe 20% of that improvement with a faster CPU/memory (game speed improvement).

Though...I'd like to be proven either way.

Regards,
Neil

[Neil]

==============================================================
TimeDemo Play Started , (Total Frames: 2000, Recorded Time: 111.86s)
!TimeDemo Run 0 Finished.
Play Time: 79.68s, Average FPS: 25.10
Min FPS: 11.96 at frame 150, Max FPS: 35.27 at frame 864
Average Tri/Sec: -23705788, Tri/Frame: -944436
Recorded/Played Tris ratio: -0.97
!TimeDemo Run 1 Finished.
Play Time: 70.51s, Average FPS: 28.36
Min FPS: 11.96 at frame 150, Max FPS: 39.16 at frame 90
Average Tri/Sec: -26489590, Tri/Frame: -933933
Recorded/Played Tris ratio: -0.98
!TimeDemo Run 2 Finished.
Play Time: 70.54s, Average FPS: 28.35
Min FPS: 11.96 at frame 150, Max FPS: 40.09 at frame 996
Average Tri/Sec: -26479394, Tri/Frame: -933955
Recorded/Played Tris ratio: -0.98
!TimeDemo Run 3 Finished.
Play Time: 70.57s, Average FPS: 28.34
Min FPS: 11.96 at frame 150, Max FPS: 40.09 at frame 996
Average Tri/Sec: -26456612, Tri/Frame: -933518
Recorded/Played Tris ratio: -0.98
TimeDemo Play Ended, (4 Runs Performed)
==============================================================

EDIT: Here are my CPU results:

Running CPU benchmark 1
Results will depend on current system settings
Press any key to continue . . .
Running...
==============================================================
==============================================================
TimeDemo Play Started , (Total Frames: 1500, Recorded Time: 44.62s)
!TimeDemo Run 0 Finished.
Play Time: 59.20s, Average FPS: 25.34
Min FPS: 10.79 at frame 202, Max FPS: 39.09 at frame 346
Average Tri/Sec: 29315374, Tri/Frame: 1156969
Recorded/Played Tris ratio: 0.62
!TimeDemo Run 1 Finished.
Play Time: 52.24s, Average FPS: 28.71
Min FPS: 10.79 at frame 202, Max FPS: 39.67 at frame 354
Average Tri/Sec: 33319966, Tri/Frame: 1160491
Recorded/Played Tris ratio: 0.62
!TimeDemo Run 2 Finished.
Play Time: 53.57s, Average FPS: 28.00
Min FPS: 10.79 at frame 202, Max FPS: 39.67 at frame 354
Average Tri/Sec: 32475188, Tri/Frame: 1159780
Recorded/Played Tris ratio: 0.62
!TimeDemo Run 3 Finished.
Play Time: 53.27s, Average FPS: 28.16
Min FPS: 10.79 at frame 202, Max FPS: 39.67 at frame 354
Average Tri/Sec: 32772064, Tri/Frame: 1163925
Recorded/Played Tris ratio: 0.62
TimeDemo Play Ended, (4 Runs Performed)
==============================================================
Press any key to continue . . .


The above results were at 1680 X 1050
NO AA
All settings HIGH

[sharky]

hi guys! sorry for beeing late, didnt find the post.

actually the main point was a bit different:

neil asked me some thoughts about a pc, i told him that 533mhz ram are out of date and slow.

the answer was, that he preffers to put the money in a gpu, because the frequency change in the ram does not bring a high performance increasement. so i said that it is not true, because the cpu gets the data from the ram, and if the ram is slow also the cpu is slow. so even if he has a higher graphics card, i will win if i have better ram which have a higer frequency and make my cpu work at 100%...

bye

sharky

[Neil]

Sharky is correct. I widened the bet.

Let's see what the results tell us.

Regards,
Neil

[sharky]

here is my result.

on top there is teh CPU test, second is GPU test. settings: AA off, everything to high, except physics that where at very high.
it may also be important to say that my computer needs a format since most applications crash and its not clean.

[A.J.]

RAM doesn't have that big impact on performance,I recently upgraded from ddr2 800mhz to 1200mhz and I got about 1fps increase on average

At low resolutions games are usually bottlenecked by CPU, at higher resolutions by GPU. Crysis however seems to be bottlenecked by both in my experience.

So get a good CPU and GPU, but fast memory only if you're overclocking.

[LukePC1]

@ sharky
wow ca. 20fps max! If you turn on S-3D it won't be playable...
I think I cant even start at that settings

@ A.J.
For me a change from single to Dual-channel Ram didn't change much (except I've got more Ram now...)
A CPU has a L2 cache (which ich VERY high on a Core2duo --> even less impact)
I think crysis needs both parts, too. But if I you run the GPU benchmark all AI and physics are turned off, so it needs only GPU

I get bad grey and brown anomalies... I'll try to reinstall it

[Neil]

Igor, run your stats again.

I couldn't set physics to "Very High", that's a DX10 thing.

We are getting pretty close!

Regards,
Neil

[sharky]

@ AJ

well, 800 mhz should be enough for every program around at the moment.. 533 starts to be a littl ebit less than needed. i think that if my core dou quadro would be a octa core, (8) there would not be a big change since most of the cpu already is not used.

@ Lukepc1

i am running on vista, with system messed up, the benchmark claims to be running on DX10 even if the settings are set to high and not very high.

i think that already the operating system makes the result very different for both. to have an accurate result we woul dhave to have the same OS, and both a fresh install.. else its not reliable as result

[Neil]

Excuses, excuses!

Ok, are there other Crysis owners out there that can add to the mix?

You have to run the game in 1680 X 1050, no AA, everything HIGH (not higher), and post if you are using XP or Vista (XP would be better).

Regards,
Neil

[Tril]

Neil and Sharky, did you use the Crysis demo or the full game? The demo was badly optimised compared to the game. I've heard people say that their fps almost doubled when they went from the demo to the full game.

Here's my current system :

AMD Athlon 64 3000+
2x512MB RAM
Onboard soundcard
8600 GT stock freq. (core 600MHz, Memory 720MHz)

I tried 3DMark06 under Windows XP, Vista 32 bits and Vista 64 bits to see if there would be some difference in scores. I ran the test two times on every OS.

Windows XP, drivers 169.21 :
3968
3971

Vista 32 bits, drivers 169.25 :
3886
3875

Vista 64 bits, drivers 169.25 :
3884
3894


I previously ran the Crysis demo benchmark before and after I upgraded my video card.

NVIDIA drivers 169.04 beta on both cards under Windows XP 32 bits.
Resolution : 1280x960, 32 bits.
No AA and all settings on medium.

EVGA 7900 GT KO, core:500MHZ, memory:750 MHZ.
min FPS : 6.90
max FPS : 35.42
average FPS : 22.70

Gigabyte 8600 GT, core:600MHz, memory:720MHz.
min FPS : 14.58
max FPS : 37.96
average FPS : 25.25

The 7900 GT is a Dx9 card and the 8600 GT a DX10 card. I did not force the use of DX9 or DX10 on any of the cards.

In the end, the 8600 GT performed slightly better but honestly not that much. I still think it was a nice upgrade because the 7900 GT had a fan and the 8600 GT I purchased is a model that is totally fanless (uses a big heatsink instead). I'm a quiet pc guy. My perfect pc would be a totally silent pc.

I will probably post back in this thread in a week or two. I'm upgrading my pc and the only part that will be the same will be the video card.

[A.J.]

Here are my results on winxp and crysis sp demo:

XFX Geforce 8800 Ultra Extreme 650Mhz/2.2Ghz (forceware 169.21)
Core 2 Duo E6600 2.4Ghz @3.6Ghz
Transcend PC2-9600 aXeRam 2GB 1200MHz 5-5-5-15
Asus Commando
Sound Blaster X-Fi XtremeGamer Fatal1ty Pro

Beginning Run #1 on Map-island, Demo-benchmark_gpu
DX9 1680x1050, AA=No AA, Vsync=Disabled, 32 bit test, FullScreen
Demo Loops=4, Time Of Day= 9
Global Game Quality: High
==============================================================
TimeDemo Play Started , (Total Frames: 2000, Recorded Time: 111.86s)
!TimeDemo Run 0 Finished.
Play Time: 55.87s, Average FPS: 35.80
Min FPS: 25.99 at frame 143, Max FPS: 42.90 at frame 978
Average Tri/Sec: -34790288, Tri/Frame: -971868
Recorded/Played Tris ratio: -0.94
!TimeDemo Run 1 Finished.
Play Time: 52.87s, Average FPS: 37.83
Min FPS: 25.99 at frame 143, Max FPS: 43.89 at frame 1016
Average Tri/Sec: -36316536, Tri/Frame: -960057
Recorded/Played Tris ratio: -0.95
!TimeDemo Run 2 Finished.
Play Time: 52.86s, Average FPS: 37.84
Min FPS: 25.99 at frame 143, Max FPS: 44.83 at frame 1004
Average Tri/Sec: -36315988, Tri/Frame: -959814
Recorded/Played Tris ratio: -0.96
!TimeDemo Run 3 Finished.
Play Time: 52.87s, Average FPS: 37.83
Min FPS: 25.99 at frame 143, Max FPS: 44.83 at frame 1004
Average Tri/Sec: -36335372, Tri/Frame: -960524
Recorded/Played Tris ratio: -0.95
TimeDemo Play Ended, (4 Runs Performed)
==============================================================

Beginning Run #1 on Map-island, Demo-benchmark_cpu
DX9 1680x1050, AA=No AA, Vsync=Disabled, 32 bit test, FullScreen
Demo Loops=4, Time Of Day= 9
Global Game Quality: High
==============================================================
TimeDemo Play Started , (Total Frames: 1500, Recorded Time: 44.62s)
!TimeDemo Run 0 Finished.
Play Time: 44.46s, Average FPS: 33.74
Min FPS: 19.89 at frame 198, Max FPS: 46.90 at frame 374
Average Tri/Sec: 37632692, Tri/Frame: 1115432
Recorded/Played Tris ratio: 0.64
!TimeDemo Run 1 Finished.
Play Time: 41.32s, Average FPS: 36.30
Min FPS: 19.89 at frame 198, Max FPS: 46.90 at frame 374
Average Tri/Sec: 40426892, Tri/Frame: 1113676
Recorded/Played Tris ratio: 0.65
!TimeDemo Run 2 Finished.
Play Time: 41.80s, Average FPS: 35.89
Min FPS: 19.89 at frame 198, Max FPS: 46.90 at frame 374
Average Tri/Sec: 40278300, Tri/Frame: 1122350
Recorded/Played Tris ratio: 0.64
!TimeDemo Run 3 Finished.
Play Time: 41.85s, Average FPS: 35.84
Min FPS: 19.89 at frame 198, Max FPS: 46.90 at frame 374
Average Tri/Sec: 40027124, Tri/Frame: 1116766
Recorded/Played Tris ratio: 0.64
TimeDemo Play Ended, (4 Runs Performed)
==============================================================

I also ran these at default CPU speed of 2.4GHz and it didn't really make any difference (less than 1 fps), so I guess you're better off getting some kind of SLI system if you want to play crysis with these settings.

I also still have to disagree about the RAM speed, I ran some tests with crysis and 3dmark with 800mhz and 533mhz ram speeds while using the same timings and the differences were negligible, in crysis it was only about 0.2 fps. But if you're using some kind of value ram with bad timings, then yes, you're better off upgrading it, especially since the prices of memory are so low nowadays.

I don't think there's a big difference between crysis demo and full version, at least according to this article:
http://www.tomsgames.com/us/2007/11/16/ ... o_vs_full/

[Neil]

Fascinating!

I have the full version of Crysis in a box, not installed yet.

I'll have to install it later to be sure. I'm running the demo right now.

Let's keep those stats coming. As it stands, A.J,.'s performance improvement against mine seem to have more to do with his 8800 Ultra Extreme (can you get more extreme than that?!?) versus my 8800GTS 512. Just a 10FPS difference, and his CPU is about 20% faster and his memory is three to six times the speed of my RAM (his 1200Mhz DDR2 versus my 400Mhz DDR).

It's a scary question. Is a $200 increase in GPU value equal to $1,000 or more in CPU/memory/motherboard upgrades?

Regards,
Neil

[A.J.]

I forgot, my 3dmark06 score is 13754. I agree also that 3dmark doesn't reflect real-life performance well but I have a x1950 pro in my closet, I can do some 3dmark tests with it too tomorrow and see how much gpu affects 3dmark.

[Tril]

If you post results with the full Crysis game, you should install the patch first. I read that it improves performance so the results without and with the patch can't be compared directly.

Edit : I bought Crysis for Christmas but did not have time to install it yet. I did tonight and I tried the included benchmark without the patch and with the patch.

Same system as my last post in this thread.
XP, drivers 169.21, 1280x960.

Without patch, medium :
==============================================================
TimeDemo Play Started , (Total Frames: 2000, Recorded Time: 111.86s)
!TimeDemo Run 0 Finished.
Play Time: 105.41s, Average FPS: 18.97
Min FPS: 4.56 at frame 143, Max FPS: 28.62 at frame 873
Average Tri/Sec: 14181804, Tri/Frame: 747471
Recorded/Played Tris ratio: 1.23
!TimeDemo Run 1 Finished.
Play Time: 85.06s, Average FPS: 23.51
Min FPS: 4.56 at frame 143, Max FPS: 35.84 at frame 94
Average Tri/Sec: 17766754, Tri/Frame: 755602
Recorded/Played Tris ratio: 1.21
!TimeDemo Run 2 Finished.
Play Time: 85.47s, Average FPS: 23.40
Min FPS: 4.56 at frame 143, Max FPS: 35.84 at frame 94
Average Tri/Sec: 17711436, Tri/Frame: 756912
Recorded/Played Tris ratio: 1.21
!TimeDemo Run 3 Finished.
Play Time: 84.73s, Average FPS: 23.60
Min FPS: 4.56 at frame 143, Max FPS: 35.84 at frame 94
Average Tri/Sec: 17845712, Tri/Frame: 756060
Recorded/Played Tris ratio: 1.21
TimeDemo Play Ended, (4 Runs Performed)
==============================================================

With patch 1.1, medium :
==============================================================
TimeDemo Play Started , (Total Frames: 2000, Recorded Time: 111.86s)
!TimeDemo Run 0 Finished.
Play Time: 126.75s, Average FPS: 15.78
Min FPS: 1.49 at frame 138, Max FPS: 28.62 at frame 971
Average Tri/Sec: 11789289, Tri/Frame: 747121
Recorded/Played Tris ratio: 1.23
!TimeDemo Run 1 Finished.
Play Time: 85.40s, Average FPS: 23.42
Min FPS: 1.49 at frame 138, Max FPS: 35.97 at frame 91
Average Tri/Sec: 17690326, Tri/Frame: 755335
Recorded/Played Tris ratio: 1.21
!TimeDemo Run 2 Finished.
Play Time: 85.48s, Average FPS: 23.40
Min FPS: 1.49 at frame 138, Max FPS: 35.97 at frame 91
Average Tri/Sec: 17680328, Tri/Frame: 755634
Recorded/Played Tris ratio: 1.21
!TimeDemo Run 3 Finished.
Play Time: 85.61s, Average FPS: 23.36
Min FPS: 1.49 at frame 138, Max FPS: 35.97 at frame 91
Average Tri/Sec: 17677636, Tri/Frame: 756652
Recorded/Played Tris ratio: 1.21
TimeDemo Play Ended, (4 Runs Performed)
==============================================================

Without patch, high :
==============================================================
TimeDemo Play Started , (Total Frames: 2000, Recorded Time: 111.86s)
!TimeDemo Run 0 Finished.
Play Time: 264.68s, Average FPS: 7.56
Min FPS: 0.00 at frame 136, Max FPS: 14.61 at frame 98
Average Tri/Sec: 7902370, Tri/Frame: 1045801
Recorded/Played Tris ratio: 0.88
!TimeDemo Run 1 Finished.
Play Time: 216.25s, Average FPS: 9.25
Min FPS: 0.00 at frame 136, Max FPS: 14.61 at frame 98
Average Tri/Sec: 9777296, Tri/Frame: 1057176
Recorded/Played Tris ratio: 0.87
!TimeDemo Run 2 Finished.
Play Time: 211.73s, Average FPS: 9.45
Min FPS: 0.00 at frame 136, Max FPS: 14.61 at frame 98
Average Tri/Sec: 9983108, Tri/Frame: 1056866
Recorded/Played Tris ratio: 0.87
!TimeDemo Run 3 Finished.
Play Time: 210.41s, Average FPS: 9.51
Min FPS: 0.00 at frame 136, Max FPS: 14.61 at frame 98
Average Tri/Sec: 10045575, Tri/Frame: 1056863
Recorded/Played Tris ratio: 0.87
TimeDemo Play Ended, (4 Runs Performed)
==============================================================

With patch 1.1 :
==============================================================
TimeDemo Play Started , (Total Frames: 2000, Recorded Time: 111.86s)
!TimeDemo Run 0 Finished.
Play Time: 273.39s, Average FPS: 7.32
Min FPS: 0.00 at frame 136, Max FPS: 15.00 at frame 97
Average Tri/Sec: 7651366, Tri/Frame: 1045912
Recorded/Played Tris ratio: 0.88
!TimeDemo Run 1 Finished.
Play Time: 214.94s, Average FPS: 9.31
Min FPS: 0.00 at frame 136, Max FPS: 15.00 at frame 97
Average Tri/Sec: 9836517, Tri/Frame: 1057116
Recorded/Played Tris ratio: 0.87
!TimeDemo Run 2 Finished.
Play Time: 209.30s, Average FPS: 9.56
Min FPS: 0.00 at frame 136, Max FPS: 15.00 at frame 97
Average Tri/Sec: 10100402, Tri/Frame: 1057017
Recorded/Played Tris ratio: 0.87
!TimeDemo Run 3 Finished.
Play Time: 209.38s, Average FPS: 9.55
Min FPS: 0.00 at frame 136, Max FPS: 15.00 at frame 97
Average Tri/Sec: 10090431, Tri/Frame: 1056346
Recorded/Played Tris ratio: 0.87
TimeDemo Play Ended, (4 Runs Performed)
==============================================================

When I said that fps almost double in the full game compared to the demo based on other people experience, I was wrong. Same thing about without the patch and with the patch. It does not change performance much on my system.

The lastest beta drivers improve Crysis performance but I did not install them.

[Maggi]

Hi Guys,

I just wanted to drop a note, that I'm too busy to contribute these days, but I'll post back, once I get some numbers for you ...

Cheers,
Maggi

[Neil]

I'm looking forward to your results! Don't forget!

Neil

[gisabun]

Another facter that could be missed is the hard disk. Got a new top of the line hard disk or an "old" hard disk? Cache on it? Is it defragged regularly. An IDE, SATA2, or (if lucky) a 10,000 RPM SCSI?

[cirk2]

HDD are importatnt for Loading times and file movement.
Performance will increase in games that dynamicaly load content, e.g. WoW or Two Worlds.

IDE is the oldest and slowest transfair type, somebody who wants performance get a newer one.

SATA(2) is current the standart (at least im my litle world) good transfairates and compatible like an old IDE.

SCSI is an server Technology, so the HDDs are not created for being shutted down serveral time a day. So you have to get an external housing which keeps the Disks permantly online wich prodicing noise and heat.

I prefer an other Server Technic: Raid
Not with SCSI, ofcurse. 2 ore more SATA(2) on a raid controller in Raid 0. So you can Write on two HDDs n the same time -> 2x Data rate, plus Writing and Reading at the same time.
One disadwantage is there: If one hdd goes to hell every data is lost, so you have to keep an seperate backup.

[Maggi]

I'm looking forward to your results! Don't forget!

Neil

Hi Neil,

sorry for the delay, but I'm still too busy to finish gathering hard numbers.

So far, I found out that the answer to your initial question is: "it depends on where your bottleneck is"

In my case, I was doing incremental upgrades over the last few years, starting out with an Athlon XP 2400 (DDR-266) and a Matrox Parhelia 256MB

First upgrade was a Radeon 9600 non-pro with 256MB RAM -> nice boost, fair price
Second upgrade, Radeon 9600 pro with 128MB RAM -> marginal boost, used part, cheap
Third upgrade, NVidia 7800 GS -> big boost, incredibly expensive
Fourth upgrade was replacing mainboard, RAM & CPU to my actual config (X2 3880+ EE, DDR2-800) -> big boost, used parts from eBay, fairly cheap

Interestingly, the latter configuration provided an enormous boost in most situations, but there are also a few numbers, where performance actually decreased compared to the former config ... go figure !

However, it might take another week or two, until I finished benchmarking and until then my recommendation is that anybody should try to find out the weakest link in their respective hardware configuration and opt to upgrade that part.

Cheers,
Maggi

[Maggi]

Holy Mo ... time is flying these weeks !!!

Anyways, benchmarking several configurations and digging my harddrives, I found some numbers that date back to the beginning of this millennium ...

Back then I was running a Celeron300A (Mendocino) @ 504MHz, which got upgraded by a Celeron2-566@850MHz and my trusty G400 @ G400 MAX speeds got replaced by a Matrox Parhelia, before I upgraded MoBo and CPU to an ASUS A7V-333 with an Athlon XP 2400+ (Thorton) and afterwards the Parhelia got replaced by a Radeon 9600 non-Pro with 256MB, which eventually made room for a Radeon 9600 Pro with 128MB, only to be replaced by my actual nVidia 7800GS (AGP), which is still present in my actual setup of an ASRock 939Dual-VSTA, sporting an Athlon X2 3800+ EE @ 2500MHz and 2GB DDR2-833 via AM2CPU upgrade board



As you can imagine, it is quite a time consuming process to gather and align all those numbers, so please bear with me, that I'm still not done yet.


However, I just found an excellent article covering this very topic right here:

http://www.digit-life.com/articles3/vid ... tions.html

They even have an interactive chart system, which lets you compare various upgrade options and it gets very clear that for gaming purposes a gfx card upgrade results in a tremendously bigger boost of framerates than upgrading the CPU ...

Cheers,
Maggi

PS: I'm having some odd issues with the Crysis Demo, showing heavy screen corruption, unless I have the shaders set to max quality ... is that a known issue ?

[Maggi]

Hi Guys,

I'm terribly sorry that I was never able to finish my personal survey on this topic.

I'm just too busy and on those rare occasions that I have some free time on my hand, I have other/better things to do.

...

Fortunately, Tom's Hardware was covering this exact topic to a great extend and to gie you a quick summary of his findings, I'll just say that in general, upgrading the GPU offers greater performance boost than upgrading the CPU.

Have a look for yourself here:

http://www.tomshardware.com/reviews/cpu ... ,1928.html

Best regards,
Maggi

[LukePC1]

Thanks for sharing

It seems like a faster CPU helps to some degree, but it has to be balanced with the CPU. A pitty they didn't include any AMD/ATI hardware. Maybe they behave different ?

So it's good for the 3D gamer to focus on GPU, but if only some GPU's work with a special Stereo driver you could get some FPS with a better CPU - and upgrade the GPU later when it's worth it.

[Maggi]

HI Luke,

as I initially claimed, it depends on where your bottleneck is.

If you have a powerful GPU already, you might end up getting a big boost from upgrading your CPU.

I doubt that AMD/ATI hardware would result in any different conclusion.

In my case, I got a noticeable gain from upgrading my former Radeon 9600pro to my actual 7800GS when I was still running on an AthlonXP 2700+, but when I upgraded the mainboard CPU and RAM to an X2 3800+, I got another noticeable gain, simply because my newer GPU wasn't getting saturated from the older CPU.

Cheers,
Maggi

[crim3]

I've just upgraded my processor, from p4 2.4GHz to p4 3.4GHz (socket 478, yes, still) and passed the 3dmark tests and the aquamark3 before and after replacing it. I'm using a nvidia 6600GT and 2 GB of DDR 400 RAM. Driver is 162.50 and the same for stereo driver. Here are the numbers:

Code: Select all

                   p4 2.4          p4 3.4
                   ------          ------
                 2D      3D       2D     3D
3dmark01        12890   6755     15839  7462

3dmark03        7570    3255     7958   3368

3dmark05        2890    1549     2892   1563

aquamark3       46413   23068    52988  24132

It seems that a 6600 can eat simple textured polygons like nothing, so the quicker processor means a great performance rise with old graphic engines (say 3dmark01). But with more modern graphics the bottle neck is the 6600 and the values are almost the same.
These are just raw graphic power numbers. In practice, the games and simulators do more things than graphics and I can see a general fps increase. But what makes the expenditure worth is that this processor has cost me three times less than what it cost some years ago. With normal prices the upgrade is not justified (in my case! not in general)

As a side note, I have enjoyed a lot seeing the 3dmark tests in 3D, specially that one of the globus flying through the cannon in 3dmark05. Even with the extremely low fps it was a real pleasure for the eyes. It's a good idea to have it on the computer to show to family and friends how S-3D enhances the visual experience absolutely.