3 GB is not the magic number

I will start with two informative posts made previously. They illustrate that 3 gb shouldn’t provide any substantial benefit. I’m not saying gen-tech deliberately misled us, but his results are dubious at best. Fps can vary widely in games base on numerous factors a few screens prove almost nothing. I thank Chuck 232 and Dcp for their posts on the matter.



Chuck 232 said,
“Everyone's saying TurboCache does a lot for performance. Any proof? All I see is a couple screen shots, which may or may not be affected by a whole slew of other variables. As it stands, FEAR does not see much of a benefit going from even 256MB of dedicated frame buffer to 512MB on a 7950GT (which is much more capable of pushing the amount of pixels required for the large amount of textures than a 8600M). Unless the idea here is that TurboCache is more beneficial than dedicated memory (in which case both nVidia and ATI are going down the wrong path for their flagship cards), there is something else affecting the frame rates.

Additionally, there should be minimal performance benefit of moving from 2GB to 3GB of RAM, discounting any TurboCache improvement. Here's a performance test showing some improvement going from 512MB to 1GB, but none going from 1GB to 2GB. From 2GB to 3GB should be the same.

http://www.gamespot.com/features/613...pe=tech&page=5

I would be much more convinced by a fixed benchmark (such as a timedemo of some sort). It is quite possible there could be a driver issue depressing scores with certain TurboCache configurations or texture rendering issues (missing textures, etc).”



Dcp said,
“I don't think TC can benefit a relatively high end card much either.

The 8600M GT's on board memory bandwidth is 22.4GB/s. Now the TC part..see the graph, the system components will occupy a considerable amount of the theoretically 10.6GB/s system memory bandwidth all the time but then the real bottle neck is at the 16x PCI-Express Lane, 4GB/s bi-direction and you will need to share it will the card itself too other than doing TC. So TC's real bandwidth is something less than 4GB/s. I think TC can help but it's not a day and night difference provided that its bandwidth is severely limited."
http://www.tomshardware.com/2005/02/...che/page3.html



I did some rather unprofessional tests of my own that indicate that the performance gain should be nearly non –existent. Unfortunately, I don’t own fear so I used the most demanding game available to me, supreme commander. My memory usage was at 50%-60% meaning that even in today’s most advanced games 2gb is all that is required. Even with 40% available to turbo cache it would have its theoretical maximum amount available to it. On another note vista does not quite live up to its memory hog status. It disables much of its memory intensive tasks if a program is running. It is still significantly worse than xp. I realize this is only my second post and apologize for being rather presumptuous. I just thought this issue should be more clearly addressed.

 

Well, the whole reason that 3GB is called the "magic number" is that it defies reason, but multiple tests from more than one user show that 3GB of RAM in Windows Vista x32 improves game performance in many memory intensive games. Supreme Commander is very CPU intensive, so that would not have been the best of tests. Also, it doesn't have to do with Memory Load, we know that, but it does have something to do with that exact value for some reason syncing up with the other system components better than 2 or 4GB. I am not saying that your logic is wrong, but I AM saying that tests have proven otherwise. I know Ken pretty well, and I know that he doesn't lie like that, especially when it wouldn't behoove him (why not say 4GB is the magic number to sell more RAM?).

 

tests with ram will remain inconclusive until a final bios is released. over the years weird things can happen every time a bios is released. not to mention that ken him self said he had some problems with ram. in tests that i ran 4gb had noticeable results (a few %) on 3dmark scores. i have not tried 3gb of ram but i doubt there is much difference between 4gb and 3gb. i dont see how getting 3gb at a time when ram is cheapest it ever was would be good advice. especially for anyone that ever considered using a 64bit os which will have support for all 4gb

 

Vista is not a memory hog in the sense that most people think. Vista uses superfetch to load programs that used most often so that they are able to be loaded quicker. If progam a is superfetched then program b is not, but requires more memory, then program a gets offloaded to make room for b. It's pretty smar because most people use the same thing every day. There are exceptions, but when I turn my computer on I pull up Firefox and Thunderbird everytime so it does have it's benefits.

 

I haven’t searched through all the threads, but who else demonstrated this effect. If it is the case for now, perhaps a driver update will fix how the GPU handless ram. I just can’t see how this would occur. Can you give me any links to other tests? Thanks for the civil reply rep up.

 

The last post was intended for alex. I guess I wasn’t fast enough.

 

No problem, welcome and I had to rep because of the way you presented your case. 2 sources and then made your own statement and did it in a non offensive way.

 

All the replies where civil thanks guys.

 

I’m used to far less friendly forums.

 

We keep it pretty nice here. Every now and then something will blow up but we are very friendly here. If we have heated discussions its about computers and not about each other, flame wars, blah blah... Hope you stick around.

 

This thread was the real clencher for the 3GB magic number theory for me. Sneis is a reputable guy with no reason to lie as well.

 

Well I just quickly glanced at this one, and the guild wars test was invalid. He tested the FPS in cities, which you shouldn't do. The more people in a Given area will cause the FPS to change. Since you can't control who is in a city, you should take the test outside where everything is instanced.

 

Did he do a comparison test with a stock g1s? It looked to me that he went from 4 to 3gb. The 4g version may have problems inherent to the 32 bit operating system or system limitations. That could mean that the 3gb would perform the same as a 2gb set up. I just don’t trust screen caps. Thanks for the link though.

 

His Oblivion benchs were consistent with GenTech's shown increase though, and he and I both pointed out the problems with the Guild Wars bench.

 

I ran all of my tests on Vista x64; all 4GB was detected and used by Windows as normal. I never ran stock 2GB g1s numbers but I did run the stock G1p with Xp x64.

In all reality the intent of my post was to show that the G1s although an awesome laptop, was not exactly leaps and bounds better than the G1p. With all of the numbers flying around, I think some people were expecting 50% more performance because of 3dmark tests when in all reality it just wasn't possible with the 8600m GT in the real world.

Furthermore, I tend to stick to playing games that I actually enjoy; hence providing an idea of what to expect when playing Guild Wars, CS:S and CnC3. Whether or not these games are ram/cpu/gpu intensive I wanted to give you guys an idea of how they ran and at what settings.

It was impressive to see though how well both the G1p and G1s ran Oblivion, I remember my first experience with the game it ate my XPS Gen 2 for breakfast. Also it was sad to see that CnC3 didn't quite run any better on the G1s compared to the G1p, I love that game but the only way to play it over LAN is to decrease all settings to bare min.

 

I thought that COre Duo, and Core 2 Duo CPU benefited from matched dual channel RAM. I cant remember exactly where I read it, but I have always thought that. Meaning if you want 2Gb of RAM get 2 x 1Gb RAM sticks. 3Gb is unbalanced since I have yet to see 1.5Gb RAM, and it is not offered at any of the sites I have looked at for Asus Laptops.

SO.... Would it not be a benefit to the way the system uses its RAM to get 2x 1Gb or 2x 2Gb instead of 1Gb + 2Gb? Just throwing it out there...

 

One would think, but the issue is that performance tests show differently. Against all logic 3GB even outperforms 4GB in some instances.

 

You wouldn't want to see me angry


This thread has been an interesting one so far, whatever the case may be in the end, we all win, +knowledge for everyone

 

Again with the bright happiness. Goodness, are you sure you are a forum user?

 

Sorry if it's been answered before, but is this 3GB thing happen to other brands too?

 

NO idea, it may just be with the G1S

 

when I get my S96S I will post up in comparison to XoticPCs test with my 3 Gigs

 

Also, for Gentech, look at the screenshots. For the 2GB benchmarks, there's noticeably higher graphics settings (antialiasing enabled, etc), while for the 3GB benchmarks, there's noticeably lower quality (no antialiasing, no softshadows in fear, etc).

That is the cause of the higher frame rate with 3GB.

Here's an example:

"Magic" 3GB performance gain
Oh wow, much better looking for some strange reason with 2GB but lower fps...

 

I'm sure that you realise that the first screenshot was from the F3SV, which he always used lower settings on, and the second is from the G1S, which used higher settings, right? Otherwise you would just look silly.

 

But considering he 'marketed' the 3GB to be the reason for the increased performance, I don't look silly at all. Maybe he's 'silly' for not realizing lower quality settings = higher performance.

 

But he didn't lower the settings.

What you presented were two different computer systems with the settings the same throughout the tests. The first picture was of the Asus F3SV, with a nVidia 8600M GS 256MB card, which uses slightly less powerful pixel shader processing(I believe). The second picture is with the Asus G1S, which uses a more powerful card (8600M GT 256GB) and will look (very slightly) better at the same settings, but much faster. So yeah, Ken has been in the computer business for 10 years, I'm pretty sure he understands how settings affect framerates.

 

But considering he was saying how the GS with 3GB of RAM was faster than the GT with only 2GB while not using identical settings shows ignorance. You shouldn't say one's faster than the other if you're not comparing them in equal environments. You also shouldn't say the performance increase was based on RAM when you're lowering the image quality settings.

They can both render identically visually. The only difference is how fast they render the scene. The reason for the visual difference was because with one system, he had the game configured to render without aa and softshadows (and who knows what else he disabled), while the other had them on.

 

Well, that is suspicious, (lack of soft shadows and AA at 0.) what I don't understand is why the screencaps are 4:3 aspect ratio, whereas the settings clearly state 1400x1050. screen caping does not change the aspect ratio.

 

I hadn't seen MrSneis' tests until now. I think I see what the problem is.

It's especially clear if you look at his CoH scores.

3GB of RAM is indeed a magically leap over other memory configurations, but it is very misleading because the performance with other memory configurations is extraordinarily low. We're talking about 2/3rds the performance of the Go 7700. The 3GB scores are more in line with what I'd expect the 8600M GT scores to be. So it's not that 3GB is magical in a good way, it's that other memory configurations is magical in a fashion detrimental to performance. Hence the 3GB config looks good in comparison. Compared to the G1p, obviously there's very little improvement in CoH, even with the new GPU.

Now let's look at the Oblivion numbers he posted.

The 8600 series cards have one advantage over the DX9 cards they're replacing - shader power. In shader-intensive games, such as STALKER and Oblivion, the desktop 8600GTS performs wonderfully, even outperforming cards like the X1950 Pro, which it typically underperforms in other games.

http://www.firingsquad.com/hardware/nvidia_geforce_8600_gts_preview/page8.asp

I don't see why it would be any different on the mobile side. Especially compared to the older Go 7700, the 8600M GT should have a substantial advantage - this is not reflected so much in the 4GB scores. The 3GB score ameliorates that slightly, but still nothing spectacular, given the advantage the 8600s have on the desktop.

So I reiterate: 3GB is not the magic number. In fact, it is the other memory configurations' fault for portraying the 3GB config as something of a miracle performer. In reality, the other memory configs have dismal video performance. I would hope that other laptops do not exhibit this sort of behavior as almost no laptops ship with 3GB of RAM.

AlexOnFyre, 1400x1050 is 4:3 aspect ratio. There's nothing wrong there.

In any case, Osserpse already pointed out what was wrong with GenTech's testing methodology for those FEAR screen caps. I can't believe I didn't think to actually take a look at the screenshot image quality. I just took GenTech's word at face value when they presented the image showing the settings used for the tests. (I obviously presumed they used the same settings).

 

So does FEAR run at 1680x1050?

 

Yes FEAR supports all widescreen resolutions

 

So why are the menu screen shots in widescreen when the setting says a 4:3 aspect ratio?