Game Benchmark: Grand Theft Auto IV

ice got 512mb more than enough.... YAY IM FIRST TO COMMENT XD

It's not entirely right to compare 3 different manufacturers... Especially that 2GB version is definitely not a "reference design"... Ideally - Toms should have just asked one manufacturer to provide 3 versions of the card with same GPU just with different RAM installed. Sapphire does offer this.

And where is the same set of benchmarks for GTXes?
Surely the "theory" does not only apply here... Have a look at the 5870 tests - 1.8GB SLI nvidia configuration is doing worse than a pair of Radeons... So clearly the theory is not everything - The memory controllers used on the cards are important, the bandwidth and the amount is important, the drivers are important, and even internal GPU architecture is important to answer the question in the title of the article... Not to mention that you can only compare these things only when you eliminate other bottlenecks... Now when we talk SLI / CrossFire - it's yet another pair of shoes... Title should read:
"how much RAM do you need for a single Radeon 4000 series in your i7 OCed system?"

Ok... I take it back... You are right... I haven't read the whole thing - the title should be: "Why buying more than 1GB card is pointless". Then it does all make sense.

good report well done
off course i knew this all along :0
when you max out everything on 1900x1200 or 2500x1600 every little bit of hareware helps

Really this article needed to address Crossfire & SLI modes as these setups need a lot of RAM since only the Primary card's RAM is used. This therefore needs to be big to cope with the extra processing power that is requiring more data.

Another thoughtful test from toms hardware, this is useful info rather than just the same old tests that everyone else has. Yes, it would be better to include nvidia and multi-gpu configs in every conceiveable config, but you can't expect everything. It's enough to say that 1gb is enough for vast majority of the time.

I expect they will update this is a few years

In Tom's original review of the 4870 1GB you compared it to the 4870 512MB and concluded that the additional RAM was making a performance difference at normal gaming resolutions. These results somewhat agree but they seem to be lacking in comparison to the original review.

In particular one of the games that showed a big difference was GRID and there were others. Interestingly you haven't used any of those games that showed a difference before with this review. What I don't understand is why you would try and bias buyers to lower RAM cards, that doesn't make any sense to me, especially when you have already run such a review and concluded the opposite of today's review.

seems its best to have 1 GB of ram.

The answer is MOAR!!!

this is a reply to the comment about only the Primary cards memory being used in Crossfire and SLI, this is not correct, what happens is that the same data needs to be duplicated for each GPU (thats what Dual GPU cards need 2 sets of Memory, 1 for each GPU, if only the "Primary" card/GPU needed Memory a Dual GPU card would not need Memory for both, its not like you need to choose the GPU when running in single GPU Mode, not only do you not get that choice but if you could it could still be avoided and making the card cheaper by only allowing the GPU with the Memory to run when single GPU mode is selected. if only primary Card/GPU needed memory even in Dual GPU mode a Dual GPU card would only need 1 set of Memory).

that means running in SLI is no more "stressful" on the Memory than running in single GPU Mode as the GPU's are fed by the same Memory in normal or SLI/Crossfire Mode.

That means if you have 3 GTX 285 IGB's in Tri SLi all 3 GB's are being used but are running as dedicated 1GB for each GPU (GPU from card 1 cant access Memory on cards 2 and 3,it can onyl access its own Memory, thats what i mean by dedicated). Since the data needs to be duplicated for each GPU any games/programmes are restricted to only using the amount of Memory each GPU has dedicated to it, in the case of the GTX 285 1Gb that would be 1GB.

Its not always the Primary GPU that dictates the amount of Memory games/apps can access, if you use a GTX 285 2GB as the primary card in a Tri SLi set-up with 2 GTX 285 1GB games/apps will be resticted to using the lowest amount of dedicated GPU Memory, in this example that would be 1GB.

That is an extremely helpful post, mont - thanks for clearing that up and helping me to understand.

There is a discussion where the argument was made that running a 2 gig videocard on a 3 gig system would not give reliable results as it would need more ram to make use of the video memory.

http://www.scorched3d.co.uk/phpBB3 [...] 195#p56195

Have there been tests done that would support this argument or can anyone explain how main ram would limit the amount of video memory ,or would not?

The first limitation is that a 32bit OS can't address more than 3.5-4Gb of RAM, including (I think) VRAM.

The second would be that the CPU uses RAM to store game information before it passes it on to the Graphics card, so if the VRAM is using up address space, then the available RAM for the CPU is much less than it should be. This means that the CPU can't provide game information (eg. The location of a tree) for the GPU to render quick enough and the game is unplayable.

Since this test is done on a 64bit OS its not an argument.

Would the 2Gig card require 2Gig or more of dedicated ram to function up to its fullest potential? Could make sense, altough i have to be skeptical as i dont hear why it needs this. Would it mean that more than 3 gig in this test could have given different results?

I have read that all the data stored in the dedicated GPU Memory is also stored in system Memory but thats something i am not 100% educated about so cant say if thats true or not.

Very nice article, well done! Just to augment, I think it would be interesting to see how memory would affect Crysis with AA on. Can you post a couple of extra benchmarks for the 1680 and 1920 resolutions?

The RAM & VRAM relationship discussed above would be interesting if we could have some clarification.

GTA4 is an awfully coded game, just another badly ported console game. It does make an excellent benchmark though

+1 to the above.

Here's some clarification on the RAM matter (I removed the pics):

"Note the gap in the memory address range from page 9F0000 to page 100000, and another gap from DFE6D000 to FFFFFFFF (4GB). However, when I boot that system with 64-bit Vista, all 4GB show up as available and you can see how Windows uses the remaining 500MB of RAM that are above the 4GB boundary:

What's occupying the holes below 4GB? The Device Manager can answer that question. To check, launch "devmgmt.msc", select Resources by Connection in the View Menu, and expand the Memory node. On my laptop, the primary consumer of mapped device memory is, unsurprisingly, the video card, which consumes 256MB in the range E0000000-EFFFFFFF:"

To summarize, the graphics card will use a certain amount of address space (but not physical RAM space). But, the space required DOES NOT equal the amount of VRAM the card has on it. It can do, or it could be anywhere from 1/2 the amount to 1/8 depending on the setup. SLI uses comparatively less than a single card with the same VRAM total.
Apparently there's a feature called 'memory remapping' or some such which helps with this, but I don't know anything about that.

So the card uses some address space. In 32bit, this causes some of the physical RAM to be pushed above the 4Gb address space boundry, making it unusable for all intents and purposes. This isn't a problem in 64bit, because (as it says above) the remaining physical RAM that wouldn't be addressed by 32bit is addressed above the previous 4Gb limit. It might become a problem when VRAM is around 200Gb and RAM is 2Tb (the current limit of Windows 7 64bit).

(Sorry for the spelling - annoying American spell checker)