What is wrong with 3dmark06?

When I compare notebooks, I mainly look at the 3dmark06 score taking into account the cpu score and the Vantage GPU score. There seems to be a contradiction between these two here.

http://forum.notebookreview.com/attachment.php?attachmentid=33246&d=1240450275

and

The 1st is a P7811 overclocked and it got 11k 3dmark06 with a weaker cpu than the i5. However, the GPU is obviously weaker than the overclocked 5830. What causes the gateway to get a higher 3dmark06 score?

 

The HD5830 in the Envy, if I remember right, is 128-bit bus with GDDR3. So it is substantially weaker than the 9800m GTS.

 

The Vantage GPU score suggests otherwise. An OCed 9800m gts gets abt 4,500 at best.

 

Isn't Vantage more affected by CPU speed than 06?

 

No. In fact, 3DMark Vantage is one of the most CPU-independent benchmarks out there. There's practically no difference between my Vantage GPU score whether I use a P7350 or a T9600. In 3DMark06, the difference is 1000 points, out of which just 400 is the difference in the CPU scores.

 

Thanks for clarifying.

Interestingly, NotebookCheck shows a similar discrepancy between the 06 and Vantage scores between the two cards.

HD5830
06: 7410
Vantage: 4528

9800m GTS
06: 9585
Vantage: 4060

In both this case and NotebookCheck's, the Nvidia does better in 06 but the ATI does better in Vantage. The difference looks to be significant enough in both cases that we can't completely discount it to CPU differences.

So what do we believe now? That 06 favors different types of GPU power than Vantage (say, number of shaders vs. memory clock), or that newer cards do better in Vantage than in 06? Or are the drivers to blame?

 

hi, does hard disk 3dmark06 take hard drive into consideration? i have just built a pc and i have the built in g41 chipset on it(intel 4500) but my hard drive is 250gb ide. I also have 2gb ram and 2.8ghz core2duo. Do you think the score is being affected by the hard drive? I getting 700ish on 3dmark06.

 

3DMark06 is outdated and you shouldnt pay much heed to it anymore.

 

I think that's the source of the discrepancy. The 5830 has a lot more processing power than the 9800 GTS, but only half the memory bandwidth (roughly). Performance in any benchmark or game will depend on whether which one is more heavily stressed so it's not surprising that the results can vary widely.

IMHO, it is unwise to purchase any of these 58XX cards with a 128-bit bus and GDDR3. The desktop 5750 and 5770 which these mobile cards are based on are bottlenecked by memory bandwidth even with GDDR5, but with GDDR3 the extent of this effect becomes truly brutal.

 

This makes intuitive sense, at least to me. I'm definitely curious to see some gaming benchmarks now. I want to see which 3dmark is more representative in this case.

It might be too soon to start calling 06 outdated and useless.

 

This makes comparing the traditional 256 bit cards to the newer 128 bit cards very difficult.

 

I also find vantage too kind to SLI setups. It doubles the scores of single 260m and 280m.

 

u can't expect everything to be perfect in this world... BTW nothing is and not even the best program

 

Not that much. For the cost of upgrading to 256bits, I don't think perf increase would be worth it.

 

So my point is if card A is better than card B, it should do better than card B in all benches. The differences here is significant 1,000 3dmark less with a stronger CPU but 1,000 more vantage gpu score.

 

Unwise? My Envy is quite fast. And you won't get any machine nearly as fast as this that is also good on battery as well as being thin and light. I can also overclock the heck out of this with very little extra heat... default is 500/800, this runs perfectly stable without artifacts at 600/1100, which REALLY helps with the limited memory bandwidth.

This isn't as fast as the 9800, or maybe even the 280, but it's got DX11, and it is VERY fast especially considering that it's so portable.

If you're looking for a purely gaming laptop with no concern for battery life, the Envy isn't what you want. But if you're looking for a laptop that'll do a lot of things quite well while still playing games very respectably (I'll come near 10K 3DMarks OC'd while warm), I wouldn't say the 5830 is unwise.

 

Not much (they're still pretty decent cards), but enough to make it a bottleneck and enough to make them inferior to the high end of the previous generation (which sold at the same price when these cards were released):

http://www.anandtech.com/video/showdoc.aspx?i=3658&p=14

My point was that you simply can't cut the memory bandwidth on those cards by a factor of 2 (as HP and MSI and various others have done for the mobile versions). That is their weakest point and doing so more or less reduces them to mid-range cards.

 

I was being slightly imprecise: it is unwise if your intention is to get a high end card that is fairly close to the most powerful mobile single GPU solutions (which the GDDR5 variety of the 5800s are). If you only want something competitive with cards from a generation or two ago, then it's fine.

Does the Envy really have such good battery life? All the reviews I could find list the low battery life as a weakness, but they appear to be talking about the version from December 2009 (pre-Arrandale, pre-mobile 5800).

 

3dMark06 *is* outdated. 3dMark06 is MUCH more CPU intensive than Vantage. This is because 3dMark06 defaults to 1280x1024, and the resolution you are, the more likely CPU factors come into play.

Put it simply... at lower resolutions, a GPU might be only utilizing 50% of its total ability. Thus it now depends on the CPU to feed it more data faster, otherwise it will have units idling, and thus the CPU becomes the limiting factor.

Crank up the resolution to say... 2560 x 1600. Now the GPU is far and away the most stressed unit, and so the CPU factor is minimized.

A real world version of this is benchmarking a 5870 desktop. Look em up - a 3dMark06 run at default on a stock Core 2 dual-core, and some people get 12k-14k. Now put that same puppy in a 4.0GHz+ Core i7 quad, and we're talking 20k+ no problem. And that's all because of CPU.

Now run Vantage in Extreme settings mode... and the performance of the GPU will not differ a lot.

Vantage is also DX10 which means that newer cards, more well optimized for DX10/11, will run better. This is likelier to translate to newer games as well, which will be better run on new hardware.

Hence despite a 5830 running benchmarks slower than a 9800M, it will easily beat it in game performance, especially in newer games.

Ah, but you're making a mistake here - comparing performance based on specs across different architectures does not work (especially across different brands).

The simple answer many people gave for why these cards aren't as fast as their specs indicate, is because they are hampered by memory bandwidth. This was true when the RV870 was released - the 5870 is theoretically 2 x 4890's on a single GPU without the issues of CrossFire scaling. However, the 5870 only matches the 4870X2.

The first thing people said was... well it must be memory bandwidth! 256-bit just doesn't cut it!...

Well, not quite. People did memory scaling tests, and found that performance does not scale nearly as much with memory overclocking as core overclocking. So memory was not the limitation - in fact, look at the beyond3d thread on R8xx architecture, they go in to a lot of detail about what could be issues - this includes scheduler/dispatcher/other components in the architecture that were changed.

In fact, a great example is the Anandtech review you linked - memory bandwidth is the easy culprit. But how about this: despite being the same clocked, having the same # of units, but being under CrossFire, the 5770 beats the 5870.

Yes, the 5870 has the same number of SP's, TMU's, RBE's, and is 256-bit with the same clocks and none of the CF problems, but the 5770 CF's beat the 5870. So it's more than just memory bandwidth issues at play here...

And let's not forget that memory bandwidth affects the different brands of cards differently. See the performance of G92 vs. G80 at high AA and memory use. Or the performance of GT200 cards vs. RV770... despite having less RAM and less bandwidth, the 4850 and 4870's matched the GT200 cards at high AA and memory scenarios.

A 5850 with 128-bit + GDDR3 is a bit lame, but the 5830, 57xx's, and the 5650 have done far better than the memory bandwidth would suggest.

In fact, look at http://www.notebookcheck.net/ATI-Mobility-Radeon-HD-5650-Graphics-Card-Review.24035.0.html

This review was clocked at 600 core, 800 memory, and uses only DDR3, not GDDR3. It loses in 3dMark06 to the GTS250M and the 9700M GTS, barely edges out the 4670. In Vantage, it is even with the GTS250M, just under the 9700M GTS, and beats the 4670 soundly.

However, in real world benchmarks, the 5650 outperforms the 320SP 675MHz clocked 4670 (vs. 400SP, 600Mhz clocked 5650) and even GDDR5 variants of the GTS 250M, despite losing in synthetic benchmarks. And in some games, the 5650 can match the GTX 260M, despite being way inferior in memory bandwidth.

And in some benchmarks we've been seeing form users, the 5830 and the 5730 perform far better in real world use than the 3dMark scores suggest. And some have benched the 5830 near the GTX 260M, which isn't bad for a 128-bit card with GDDR3

So the moral of the story is simple... memory bandwidth isn't everything, and synthetic benchmarks can be very misleading.

Given that most laptops these days are in the 1366 x 768 to 1600 x 1050 resolution range, the 128-bit + GDDR3 will not be a limiting factor. Yes getting something like that for 1920 x 1080 will be pushing it, but for the native resolution on most notebooks, it does just fine.

 

That's if you can get a 4870 for $140. When I bought my 5770, I paid 160 canadian. A 4870 was $199.99. (The 5770 is now 140~200 though). And the 5850 was running 300~350, making 2x 5770 a better deal.