3Dmark is completely borked

R

Rangers

Legend
Okay I just got a 7600GT, and now back in the land of living video cards (from a 9800 non-pro), I followed my longstanding new graphic card tradition of downloading and running all the 3Dmarks to add to my ORB history with my new equipment.

Anyway I noticed, 05 and 06 in particular are completely awful. Let me count the ways.

Graphics: Just besides the test, I used to like 3Dmark around 01-03 because it was simply pretty, and some of the best graphics at any given time around. 3Dmark at that time was cutting edge graphics. Now, it's hideous compared to the top games of the day. The FPS for example, in 05, besides running at a snail's pace, is just ugly compared to todays top games at any speed. About the only pretty demo now is the firefly one. They've last all sense of artistry, and all technical competence as well. Any of myriad current FPS blow away that proxyconn garbage in looks, while running at perhaps 100FPS on my same rig that gets 20 or something on proxyconn. So, 3Dmark now features both massive artistic (ugly at any speed) and technical (technically unimpressive games at EXTREMELY slow frame rates) incompetence.

But what really bugs me is the CPU tests, especially in 06, I've got a A64 3000+. 2.0 GHZ. This still I regard as a decent CPU for playing games, especially with my video card. I know it's a little dated but really, the fastest A64's are still only clocked at 2.6, 2.8 if you're lucky, and are the same basic architecture as mine. The fastest S754 I can get is a 2.4 GHZ, just 20% faster, really not worth it too me. And dual core, is basically not even used on games right now. My point is in real terms my CPU is not that far behind the top Athlons, (and until very recently Intels as well).

Yet 3Dmark 06 CPU tests, it attempts to play another ugly turret scene labeled as a CPU test, I can only assume it is software rendering this hence why it is a CPU test. I literally bounce between 0-1 FPS for the entire excruciatingly long test, which is repeated twice. While my card can at least get ~15+ FPS on the 06 game tests, so they dont take forever, somehow my CPU is just garbage. Even that isn't the worst by far, the worst is that actually getting through the CPU tests is an awful experience and basically resolves me to never run 3Dmark 06 again unless under duress or a much faster CPU. I mean seriously, it's torture.

Now this also offends me because it doesn't even make sense, the test is unbalanced, the 3000+ is both a fine CPU for my card and in fact, plenty for most all games of today as well, yet judging by 3Dmark, it both lags my card greatly, and is a bottom line piece of crap that gets 0.5 FPS on their test. Which is completely unrealistic.

Now, I know 3Dmark gets a lot of deserved flack in general, but unlike others I see the value in it. It's a good basic number to compare video cards with. For example, to try to compare my old 9800 to my new card and get a rough idea how much faster the new is, 3Dmark does in fact work as intended imo. And trying to find any current game benchmarks of a 9800 is impossible around the web, so for example, the quickest way to get a basic idea is check Tom's VGA charts for the 3Dmark score of each card. I think it might also be a decent way to compare different specific aspects like fill rate and all that junk, though I dont do that.

So I'm not a 3Dmark hater per se, but the 05 and especially 06 are just flat borked, because of that CPU test problem. Besides broken, they're ugly to boot. Rant done.
________
Problems with paxil
 
Last edited by a moderator:
sorry boy, but your cpu is way WAY slower then only 20% then the fastest (single core) amd cpu
 
hey69 said:
sorry boy, but your cpu is way WAY slower then only 20% then the fastest (single core) amd cpu
Click to expand...

Right I said 20% slower than the fastest S754 (what I have).

Fastest 754 AMD is 3700+, which is only 2.4 ghz (plus some extra cache, but extra cache hardly helps A64's at all)

It looks on a cursory examination of newegg that even the fastest AMD AM2 dual core is 2.8 GHZ. Considering a second core is essentially useless in current games, that puts it maybe 50% faster than me (counting more cache, etc). And that's absolute high end. Most "high end" AMD's floating around will be more like 2.4-2.6, with the same basic architecture as mine, and a useless for games second core.

I'm guessing 3Dmark 06 CPU tests highly utilizes that second core? That's not currently realistic though, maybe in the future.
________
Recall paxil
 
Last edited by a moderator:
Rangers said:
Right I said 20% slower than the fastest S754 (what I have).
Click to expand...
Do you really believe FutureMark designed their test just to test the S754 platform? I'm not a big fan of 05 or 06 either, but I know it's because my system ain't up to the CPU tests that I get annoyed at least.
 
Rangers said:
Yet 3Dmark 06 CPU tests, it attempts to play another ugly turret scene labeled as a CPU test, I can only assume it is software rendering this hence why it is a CPU test. I literally bounce between 0-1 FPS for the entire excruciatingly long test, which is repeated twice.
Click to expand...
No software rendering takes place - it's capped to a maximum of 2fps to remove the fps dependency from the graphics card. It's the same for everyone.

Now this also offends me because it doesn't even make sense, the test is unbalanced, the 3000+ is both a fine CPU for my card and in fact, plenty for most all games of today as well, yet judging by 3Dmark, it both lags my card greatly, and is a bottom line piece of crap that gets 0.5 FPS on their test. Which is completely unrealistic.
Click to expand...
Why is it unrealistic when it's not marketed as being a direct performance indicator of a current 3D game? 3DMark's flaws aside, it is simply a benchmarking tool that gives your PC a certain load and produces a score based on it.
 
... I thought 3D Mark was to look at the pretty pictures and then try to show you dick is bigger than other's people's dicks for about 6 months, till the next GPU comes out and your dick magically shrinks just by having an "old" GPU...
 
london-boy said:
... I thought 3D Mark was to look at the pretty pictures and then try to show you dick is bigger than other's people's dicks for about 6 months, till the next GPU comes out and your dick magically shrinks just by having an "old" GPU...
Click to expand...
No, that's what post count is for.
yep.gif
 
You? Look at me LB! I had a decent sized post count here once instead of this scrawny, limp thing. :???:

/me goes over to EB where I still have a 15K+ post count to feel a bit more manly.
 
No software rendering takes place - it's capped to a maximum of 2fps to remove the fps dependency from the graphics card. It's the same for everyone.
Click to expand...

I don't know about 06, but 05 and all previous versions definetely didn't have this cap. Pretty sure 05's software tests just offloaded the vertex load onto the cpu, while the rest of the stuff was still done by the gpu. (though it does look like software rendering, but perhaps they just lowered everything down as much as possible to remove the gpu dependency)
 
Yes, that's right but the OP was specifically referring to the CPU test in 3DMark06. In 05, the rendering method is precisely as you've described it (software VS, hardware for everything else) but the resolution is 640 x 480.
 
Neeyik said:
Yes, that's right but the OP was specifically referring to the CPU test in 3DMark06. In 05, the rendering method is precisely as you've described it (software VS, hardware for everything else) but the resolution is 640 x 480.
Click to expand...

So what does 06 do? Software everything?
 
Actually I really like the water effect in Canyon with the beast gliding below the waves, the shadows in Deep Freeze have an ehtereal beauty as well .. it does really look cold.
 
It is really a good thing to read the Whitepapers we have released for both 3DMark05 and 3DMark06. They contain some good information in terms of how the benchmarks work.

Here's something about the CPU tests in 3DMark05:
As in the previous 3DMark version, the CPU test runs game tests in low resolution using software vertex processing and disabled post-processing. This decreases the graphics card workload, and makes the test result reflect above all the CPU’s performance rendering 3D scenes and performing other 3D game related tasks like performing matrix calculations. The CPU test also uses fixed frame rendering to further ensure the workload stays the same for all systems.

In 3DMark05, an additional workload typical for the CPU in 3D games has been added. The CPU not only calculates the vertex shaders; it also continuously calculates the flight path of the air ship. The air ship actually flies the same path every time, in order to keep the workload the same between different systems, but the calculations are performed as if it would intelligently steer according to the canyon shape and other obstacles like the sea monster jumping up from the water. The path finding algorithm used is D* Lite (http://www.cc.gatech.edu/fac/Sven.Koenig/).

The D* Lite algorithm is implemented in a different thread than the rendering. This is a forward looking approach, since CPU technology is clearly is moving towards either virtual or physical dual core technology. So far only few high-end gaming systems have two separate processors. A single virtual or physical dual core processor is on the other hand becoming a more common choice for high-end gaming systems. The CPU test is therefore run in two threads and not more than that. DirectX software vertex shading is actually also threaded, which may bring further advantage to systems with a virtual or physical dual core processor. Professional reviewers can disable the other CPU or the other core of a virtual or physical dual core system, and compare the results.

The second CPU test runs game test 1 in 640x480 resolution, 2 fps fixed frame mode, software vertex shaders forced and post-processing disabled. This is just like CPU test 1, except that there is no AI calculations is CPU test 2.
Click to expand...

And here's about 3DMark06's CPU tests:
The 3DMark06 CPU tests consist of a game scene with a maze of canyons, and 87 fast-moving game units ("bots", speeder bikes and hovering tanks). The speeders attempt to navigate to a goal position at a castle at the other end of the canyon system, all the while avoiding the defending tanks, and collisions with other speeders. The tanks attempt to hunt down the speeders, and shoot them.

The game scene yields three types of load in the CPU tests: game logic, physics and path finding AI. The game logic, including the graphics engine operation, runs in a single main thread that also drives the other two tasks. The physics simulation runs in a single separate thread, and is synched with the main thread at each physics step. The path finding AI runs in a number of worker threads (the number of threads is scaled with available processors), and is synched with the main thread at set intervals, generally some multiple of the physics step interval.

The CPU tests are run in fixed frame rate (2fps) to make a more equal CPU load for all systems. The resolution is locked to 640x480 and the tests use no dynamic shadows to decrease the graphics performance influence on the result.

The physics load consists of simulating the game world with its 87 units and their rigid bodies, at 20 ms physics step. Some physics operations like traces into collision meshes, and some overhead, is also included in the main thread load. The path finding AI load consists of unit path requests in a dynamic path finding grid, where each moving unit also represents a moving obstacle of certain radius on the grid. The resulting paths are synched back to the units at intervals ranging from 200 ms to 600 ms. The complexity of individual path request fulfilments varies greatly, as we use a dynamic re-planning algorithm that is often able to re-use state from previous searches. The length of the requested paths and the grid obstacle configuration also contribute to the path finding variance.

The path finding algorithm used is D* Lite (http://www.cc.gatech.edu/fac/Sven.Koenig/) and the physics are being computed using the AGEIA PhysX library (http://www.ageia.com).

There are two CPU tests (1 and 2) in 3DMark06. The tests differ in certain parameters affecting the type of game-like load that is run on the CPU. The main settings and differences between the tests are as follows:

[font=Arial,Arial]Test 1[/font]:
Higher path finding task complexity
Tighter AI synchronization intervals
Duration 40 frames
Fixed Frame Rate 2 frames per second
Shader Profile 2_0
Resolution 640x480

[font=Arial,Arial]Test 2[/font]:
Lower path finding task complexity
Laxer AI synchronization intervals
Duration 60 frames
Fixed Frame Rate 2 frames per second
Shader Profile 2_0
Resolution 640x480

The differences result in Test 2 having a larger relative physics load (around 13%) than the relative physics load exhibited in Test 1 (around 8%), and conversely a lower relative AI load.

Both CPU tests are forward looking, since CPU technology is clearly moving towards either virtual or physical dual core technology. The CPU tests are mainly designed for dual systems, either with an on-chip solution or with two separate CPUs, but are an excellent CPU test for single core processors as well. Professional reviewers can disable the other CPU or the other core of a virtual or physical dual core system, and compare the results.
Click to expand...

Here are direct links to both Whitepapers for those interested:

3DMark06: http://www.futuremark.com/companyinfo/pressroom/companypdfs/3DMark06_Whitepaper_v1_0_2.pdf
3DMark05: http://www.futuremark.com/companyinfo/pressroom/companypdfs/3DMark05_Whitepaper_v1_1.pdf

Cheers,

Nick
 
You must log in or register to reply here.

Similar threads

R
3DMark - Solar Bay Ray Tracing test available now
Replies
7
Views
2K
Lurkmass
L
Cyan
Clair Obscur: Expedition 33 (PS5, PC, Xbox). Turn based RPG.
Replies
1
Views
702
eastmen
E
homerdog
Intel Raptor Lake voltage problem
2 3 4
Replies
72
Views
4K
Albuquerque
Albuquerque
T
GPU Stability testing for Nvidia cards - PowerSweep
Replies
0
Views
271
T2098
T
Unknown Soldier
3DMark AMD FSR Feature Test
Replies
1
Views
1K
Putas
P
Back
Top