Discussion in 'Article Discussion' started by WilHarris, 15 Feb 2006.
Command rate is the kicker, because everytime a command is issued, whether you have to wait a single clock or twice as long becomes more of an effect when you load the ram with plenty of I/O. If you have lower timings, then more commands can be issued which makes it more of a problem.
Ive gotta say, using 1gig on this lappy is getting to be a problem now when I run half a dozen apps at once. 2gig on my main PC is a nice buffer, even though it is "slow" 533MHz 4-4-4-12-1T DDR2. 4 DIMMs on an Intel machine is fine
I love the fact you've shown that 120MHz extra memory bandwidth and all you get is a single frame per second more. LOL. Although those games that really make use of large amounts of ram, like BF2 do show 5fps difference, but it is only 5fps and it' still maximum detail.
I totally agree with your conclusion that getting another pair of 512s is more worth while, and possibly a bigger cooler (if you need it) to overclock a touch more to negate the few fps loss here or there.
A very informative and well-written article; this should help to cut down on some of the confusion in HW&OC about this subject (especially command rate) and certainly dispel a lot of myths about memory performance .
There are a few typos:
"assuming something about his breaks [brakes]"
All the "prechange"s should be "precharge"s - result of automatic spell correction?
"1T or 2T clocks" - should be "1 or 2 clocks?". I thought T was suffixed to the option in order to distinguish it as a time (I might well be wrong here, though).
thanks, well spotted
Well, i upgraded from 756MB (an odd number i know) to 2GB recently and i can't say i noticed any difference in gaming performance, i.e. frame rate. However, the overall computing experience was massively improved - minimum load times between levels in games, being able to alt/tab out of games to desktop in an instant, running itunes, photoshop, outlook, FF, messenger, DoD:S, etc. all at once with no performance hit. The single best upgrade i can remember having made. Such a pity it was horribly unstable and I've had to RMA it (Crucial Ballistix - anyone else had problems?), it's been 4 weeks now since i sent them back and i'm still waiting to here anything.
More RAM always = better unless your machine cant run it
The differences in frame rate seem minimal, but when playing games like FEAR, for instance, with a Gig of 3200 XL, I still get 'stalling' when running the game on the highest texture setting.
The frame rate is fine -- it's when you're walking down a corridor, you can tell when the scene ahead is being loaded, as the whole thing slows up as the pages are being transfered to the HDD.
So there are other, more noticable reasons to go for more than 1GB of memory.
There are some small differences, but unless you've got a high end video card, you're not likely to notice the difference. Ultimately, the computing experience is so much better with 2GB of memory, but our article was designed to dispell some of the rumours that 2T = bad.
The only game that really makes a big difference in memory size is Battlefield 2 - you simply cannot run the game with all details set to maximum without hitching with only 1GB of memory.
That's the thing - you get more hitching and frame rate drops with less memory, as more textures and such have to be loaded into memory each time you go into a new room.
well as far as gaming goes i wouldnt expect to see much of a difference concidering the card being used has a full 512 on it already. for people still using older systems with 128mb 6800s lets say, the upgrade from 2x512 to 2x1024 will DEFINETLY see a boost in gaming performance. people with 6800's still have a fairly competitive card that still cost within $250-400 or that they dont want to upgrade their motherboard too if they still have AGP. that is to say untill recently wotht eh 7800 GS (probly nto entirly worth the price of an upgrade) but i guarantee you will NOTICE an improvement . I have a 6800 with 256mb.... ill let you know how well the upgrade went once i get my new board, had to RMA a gigabyte... though i guess it woulnt be a direct comparrison concidering i had an AMD 2700+ before and now i have the opteron 148.... but anyway. untill next time kids
Very well written and presented article. I found it to be incredibly useful. Shame there wasnt a few other brands offered just for fun but I can imagine the time invested for these 4 setups already. Thanks for another great article!
Hey, one more thing, I've found that machines don't like it as much when you have chips of different size and timings, I think it has more to do with the timings.
Anyway, I've found that if you turn interleaving off [via the BIOS], it becomes <i>STABLE, but</i>... it becomes unbearably slow while gaming... anybody know why that is?
Bank interleaving means that the memory controller can have two or more pages of RAM (in different banks) active and write to them both subsequently, i.e. sending a write to one page and then to another without waiting for the first to finish. Without interleaving, a bank page must be closed each time another is to be opened. Another advantage of this system is that the two banks can alternate their refresh and R/W cycles so that access is continuous without breaks for row refreshing.
It improves performance, especially in memory-intensive applications where many reads and writes are being made to different areas of RAM - this is a situation typical of a computer game. I wouldn't have thought it would be unbearably slow, but it would definitely reduce performance. It's probably because the game's high amount of access to large strings etc. that would not fit in the L1/L2 caches mean that they're more RAM intensive.
Indeed so - many older chipsets and CPUs had a maximum cacheable limit of only 64MB, which meant that anything above the 64MB address was in fact not loaded into L1/L2. Since Windows loads in from the top down, performance was completely crippled as a result.
4,194,304MB is the limit of a 32-bit address system, and while memory controller addressing has far exceeded that, the subsystems of Windows XP - or any 32-bit OS, for that matter - have not. It will conk out at about 3.5GB due to PCI MMIO occupied address space around that area. The memory hole here is created in order to give addresses to adapter/peripheral ROMs. Since this address range is mapped to the memory on these devices, it is not addressable as normal RAM and subsequently any installed memory in this range is useless. If you are using a 64-bit OS and a supporting BIOS, this area can be remapped to somewhere above 4GB to give you the full amount of memory; usually, this option is called "Remap Memory I/O Hole", "MMIO Remapping", or "S/W Memory Hole Remapping".
I think it was meant this way to advise people who currently have 2x512MB to still get 2 more 512MB modules despite having to run a 2T command rate.
Memory management can become problematical with increased amounts of RAM-- Win 9x OS and associated apps only saw a performance boost in memory up to 1GB, after that the performance actually decreased because of management overhead (max allowable memory was 4GB, IIRC).
I foresee a similar problem with XP; where that sweet spot will be is unknown, I'd hazard a guess that it would be around 4GB, but I have neither the time nor equipment to test that theory.
Regardless, with the current state of gaming, memory increases above 2GB will not return any significant improvements, IMO.
Last page, should read 4x512MB over 2x1024MB (or 2x1GB).
Very good article, made the whole memory stuff a bit clearer to me.
After reading that I'm glad I didn't bother going for some expensive stuff and went with simple corsair value select. All I'm after really is real world performance. Maybe when I'm rich I'll go for the expensive stuff
XP (not 64) can handle upto 4gig, however most motherboards can only address 3.8gig of ram(even if they say 4 gig)
also, XP can only address 2gig per thread so even if you have 4 gig(which i do) you can only use 2gig per thread thus its useless unless the program is multithreaded
will A64's (ie my Opteron 170) use a full 4gb?
It will if you use Windows XP 64-bit and enable the relevant setting in your BIOS to enable MMIO hole remapping. Keep in mind, however, that XP x64 has very poor driver support and the benefits of 4GB of RAM are debatable.
I think 4gb setups will start coming in with vista, with it's memory hogging 3d interface. Running 1gb here and haven't yet run out of memory in any games, but my graphics card holds me back from the high settings. It's a big improvnment over 512mb though where I would have to wait for 2mins to get back into css after alt+tabbing out.
I don't understand why the slower 400Mhz XMS4000, but with lower timings, was beaten by the faster 520Mhz XMS4000, but with higher timings, in all four tests...
...I heard that you should always opt for lower timings over slight-higher speeds - isn't this true?
And also, you didn't mention anything about which of the four memory formats you should look for if you intend to overclock...
Separate names with a comma.