X3100 dual channel performance

ok i have an X61 with 1.5gb of ram, a 1x1gb stick and 1x512mb stick, this is asymmetric dual channel memory mode if im not mistaken, what i want to know is how much fps gain am i looking at if i upgrade the 512mb stick to a 1gb stick , aka FULL dual channel, i have integrated graphics so it seems to me like it would be a noticeable difference, i know there were threads on this before but i didnt see one that dealt with the x3100 integrated gpu, the max fillrate on the x3100 is 12.4gb according to wiki, so wouldnt full dual channel memory maximize the performance of the graphics

 

probably no increase....actually i know it wont help

 

sorry but could you just quickly explain why not? from other threads it looks like about 3-4%, but thats memory performance with dedicated graphics, so shouldnt it be more with integrated graphics

 

RAM's not going to help your GPU compute any faster.

 

exactly, you gpu, actually your integrated graphics is so low end the ram wont help at all, it cannot use that much ram to help it at all. unless you can get a better laptop with a better gpu it wont make a difference how much ram you have in gaming.

 

ok maybe i missed the point but i thought that integrated graphics are directly dependent on system ram speed?

 

But as Lithus said, the gpu's speed is it's own bottleneck. Once that is hit, faster ram won't make the gpu process any faster..............

 

well the specs for x3100 says "500mhz clock speed, 12.8Gb peak memory bandwidth"

doesnt that mean that the GPU can push that much?

 

And if the peak memory bandwidth of DDR2-667 is 21.3GB/sec, it already way exceeds what the x3100 can handle............

 

every site i looked at says DDR2-667 bandwidth is 5.4Gb/s...??
is this single channel speed - aka its doubled when your using dual channel mode?

 

Dude, take what we're telling you. Faster/better/more RAM is not going to noticeably improve your gaming performance. This is fact, we're not pulling this out of our backsides. I'm sorry, but you have to accept it.

 

Actually guys, while the speed of the RAM won't change much going to 2x1GB, the additional 512mb could very well speed up gaming performance, especially if the OP's using Vista. And to be clear, the speed of IGP's is tied closely to RAM speed, however in this case going to a symmetrical array won't really increase the bandwith very much at all over an asymmetrical array.

 

look this is a forum, and i have a right to be sceptical, i just dont understand why it says 12.8gb peak bandwidth for the gpu and 5.4gb peak bandwidth for the 667 ram, and how this isnt a ram bottleneck, if someone can explain that to me ill be all set, sorry i need numbers

ps. im using windows xp

 

Odin is right. It will help, especially if he is using Vista. The speed of the IGP scales extremely well with CPU speed and memory size. You guys are right in saying that merely increasing the size of the shared RAM will not increase performance, but increasing the speed via a symmetric Dual-Channel array will.

 

You know, I find it ironic how one of the primary arguments against IGPs are that they are crippled by shared memory bandwidth and yet when an opportunity presents itself to double the memory bandwidth via dual channel, suddenly it's irrelevent.

http://eshop.macsales.com/Reviews/MacBook/Testing/Memory_Benchmarks

Luckily there are some benchmarks available showing the effects of various RAM configurations from 2x256MB to 2x2048MB on MacBooks. Granted they are done by a retailer, but they do look reasonable, and they seem to be the only people willing to do so much comprehensive testing.

They only have 1 game and that's Halo and there actually is a bit of improvement going from 1.5GB to 2GB dual channel, but in this case the GMA 950 really is a flop. The GMA X3100 would benefit more from dual channel memory, but you're still likely looking at maybe a 5% improvement. I would imagine the effect of dual channel memory will increase once hardware DX9.0c acceleration improves, since the bandwidth bottleneck right now is having to shove everything to the CPU thorugh the FSB for VS work then pushing it back through the FSB to the IGP for the rest of the processing. Once the IGP does everything in it's own hardware, it'll be less FSB bound, with functions being isolated to the IGP and the RAM via the northbridge. Truthfully, the real benefit of going to 2GB dual channel isn't really for games, but for everything else as in general system performance. If you are usually Vista, 2GB of RAM is recommended anyways.

EDIT: I see that you aren't using Vista, anyhow...

 

i also read that if the data is located on just one of the ram sticks you dont use dual channel ram anyway, aka if you have 2x2Gb sticks and most of the data is on just one of the sticks you only get single channel performance?

 

ltcommander_data, normally I would agree with you, however in this case, on the 965GM chipset the OP has, symmetrical dual channel gives only a tiny bit of increased bandwith over asymmetrical dual channel, due to Intel's Flex Memory technology. You're absolutely right that increasing memory bandwith would help, but in this case going from 512mb + 1GB to 1GB + 1GB wouldn't increase bandwith to the GPU at all. There was a lengthy discussion about this in the Dual Channel RAM thread.

 

I've been wondering what the effectiveness of Flex Memory is since in theory it should be the same as dual channel for memory up to twice the smallest DIMM, but I would think there would be some additional overhead versus a pure mode. I haven't really seem benchmarks specifically addressing this though. I guess the clarifying question is whether the OP routinely uses more than 1GB of RAM, since Flex Memory will only work on the first 2x512MB. If more than 1GB of RAM is generally used than having 2GB in itself would be beneficial and having the rest of it interleaved is a nice bonus. Otherwise than it'll be wise to hold off.

I'm personally find 1x1GB satisfactory for now and am just waiting for 2GB DIMM prices to come down to get 1 of those for 3GB. Sadly, Apple has never mentioned support for Flex Memory, which the 945PM chipset supports, but would need to have activated in the EFI.

 

There are some benchmarks on it in the Dual Channel RAM thread, and the bandwith difference between Flex mode and pure Dual Channel mode is negligable. And since the GPU will share from the bottom of the RAM, it will always have pure dual channel performance (as long as your smallest dim is 512MB or so). So for the Crestline chipset at least, you really shouldn't worry as long as you have two SODIMMS.

 

I see where you are going with this from the link, but even the author's conclusion was that more ram is preferrable overall to worrying about 128-bit operation. Having recently upgraded to 3GB ram running in single channel myself, I am inclined to agree.

BTW, my memory bandwidth figure given earlier was taken from a Mac Pro and does not apply here (totally different ballpark ). The OP was correct. My bad.

 

This could be material for another research project through running 3DMark05 and 36Mark06 with different RAM configurations, but I don't have time at the moment.

I think previous posters have already identified the likely effects: (i) more RAM generally helps Vista performance and (ii) the bandwidth gain with pure dual channel is likely to be a few percent only. I know from previous testing that using one RAM module only will reduce the GPU performance.

My suggestion would be to replace the 512MB module with a 2GB module to put 3GB on board and improve vista's overall performance. This might indirectly improve the gaming performance. If you've got XP, not Vista, then 2 x 1GB would help improve performance, but not by a big margin.

If RAM manufacturers could mass produce DDR2-667 RAM with CL=4 then than should also boost performance through faster RAM operation. We should not overlook the fact that dedicated GPUs use faster RAM than standard system memory.

John

 

It says on Intel desktop board product brief PDF about Flex memory and the Flex Mode configuration. This is how it works. If you have 512MB memory in one slot and 1GB memory in other slot, the bottom half of 1GB memory would be in dual channel with the 512MB in other slot, and the top half of the 1GB memory would be single channel.

That's of course in theory. Intel's desktop 915 chipset first introduced "Flex Memory". Now let's look at benchmarks: http://www.tomshardware.com/2004/06..._of_the_pc_future_with_775_launch/page28.html

There is a performance loss, and that's not testing the memory bandwidth sensitive integrated graphics.

Now think of it guys, even if the technology advanced enough on the latest chipsets that 2x512MB on Channel A and 1x1GB on Channel B no longer has performance impact over 1x1GB on both channel, we are going further down on the performance ladder. We are gonna be relying on the lowest performance Flex Memory, which at one slot there is only 512MB, and in the other there is 1GB. It says on the E7221 chipset PDF: http://www.intel.com/design/chipsets/e7221/30447301.pdf

"Dual Channel Asymmetric-This mode is entered when both memory channels are routed on the motherboard and populated with different amounts of total memory. This configuration allows addresses to be accessed in series across the channels starting in Channel A until the end of its highest rank, then continue from the bottom of Channel B to the top of the rank. Real world applications are unlikely to make requests that alternate between addresses that sit on opposite channels with this memory organization, so in most cases, bandwidth will be limited to that of a single channel."

There will be a performance increase going from 1x512MB+1x1GB to a 1x1GB+1x1GB config. I gained TEN percent in Company of Heroes benchmark performance going from 5-6-6-18 to 5-5-5-15. You would stand to gain more than me, possibly 10-15% is the figure here.

 

If you would even bother to read the Dual Channel RAM thread you would see that everything you've said has already been addressed. The quote you've copied from Intel and the benchmark you've linked are not applicable on 965 (and some other) chipsets, as the way in which Asymmetric Dual Channel is handled has been changed on those newer chipsets. (As an aside, you also can't compare desktop technologies with laptop technologies, as laptops only have 1 DIMM per channel). The quote you should have looked at is the following:
http://download.intel.com/design/chipsets/applnots/31320702.pdf