Slashdot Mirror
High Performance DDR2 Memory Breaks 1.25GHz
TrackinYeti writes "Performance PC Memory manufacturer, Corsair recently released a new addition to their flagship Dominator line of
desktop memory,
the
TWIN2X2048-10000C5DF. This 2GB DDR2 memory kit features the company's
DHX Dual Path Heat Xchange cooling technology, support for Enhanced Performance
Profiles (EPP), it includes one of Corsair's Dominator active memory coolers,
and it's rated for operation at a currently industry leading 1.25GHz."
Lovely speed, but I wonder what all that heat output will do the ambient temperature.
Because this "story" sure does!
just imagine your Beowulf cluster equipped with THOSE.
im sorry but there must be something better to do with all that money other than spend it on hardware that will be outdated in 6 months.
"Performance $OBJECT manufacturer, $COMPANY recently released a new addition to their flagship $BRAND line of $OBJECT(s), the $MODELNUMBER. This $OBJECTDESCRIPTION features the company's $SUPERLONGFEATURENAME, support for $ANOTHERFEATURENAME ($ABBR), it includes one of $COMPANY's $OTHERPRODUCTHERE, and it's rated for operation at a currently industry leading $OWNAGESPEC."
Seriously, this sounds a lot like any other marketing gimmick ever invented. And it is just asking for a car analogy. Simply replace $COMPANY with Chevrolet, and start imagining the rest..!
Yeah, like giving it to me so I can spend it on hardware that will be outdated in 6 months.
It's sad when choosing an installation directory on your own qualifies you as an "advanced user."
If the smell is really strong it is probably some sort of infection which is causing there to be high levels of trimethylamine oxide which cause a fishy odor. Get her on a regimen of Flagyl to treat the infection.
09 F9 11 02 9D 74 E3 5B D8 41 56 C5 63 56 88 C0 is the magic number.
I bought it to do high-end video processing, virtualization, and I got a good deal on all of it (except for the processor and memory) which I bought from NewEgg. The 850watt Cooler Master Power Supply pulled so much juice that the computer wouldn't even post until I used the big thick power cable that came with the power supply.
Every few years I build up an awesome machine, just for fun.
Good security is based upon reality and common sense. Common sense is a function of having common knowledge.
I can't find registered ECC DDR2 faster than 667 MHz. Why?
I was hoping my next machine would be a quad core with 800 MHz DDR2 and ECC.
Much as my current machine is PC3200 DDR with registered ECC. No sense throttling down the relative bandwidth per core.
[Please don't waste time trying to convince me I don't need ECC.
SGIs taught me otherwise and soft error rates really are on the rise. Just answer the question thanks.]
Also commonly known as plugging it in
Lose: misplace or fail || Loose: not bound together
Hey, didn't you see the sign? You broke it, you bought it.
Two questions?
1. How relevant is it to have memory that is this fast? As I understand it, no matter how fast memory is, if there isn't enough of it, your computer has to read and write from swap space on the hard drive, and even the fastest harddrive is at least a million times slower than slow memory, since it is a matter of nanoseconds vs. milliseconds (someone might correct me on the technicalities of this). So wouldn't lots of normal speed, or even slow memory, work better than too little ultra-fast memory? (Someone should just build a system that can support 8 gigs of 30 pin SiMMs!)
2. Am I a cranky old man who isn't up on this trend of memory needing active cooling? The closest I've seen is RAMBUS with aluminum sinks built in. It seems that no matter how efficient the cooling system claims to be, active cooling is another thing that can go wrong. I would much rather have slower memory that I don't have to worry about frying, then fast memory that is dependent on a fan that may break.
So, with those things in mind, how worthwhile is this?
Hopefully I didn't put any [] around my words.
The basic structure of Dynamic RAM has not changed, it still takes about 50nS for row precharge (Tras
and 20bS column reads. All they've done is speed up the interface logic. The memory cell access is no faster.
OK, so once you've opened a row, you can read that faster, but how many operating systems are
optimized to keep the data row aligned in the system memory? You have a data request that is outside
of the row you've opened, you have to close that row and open another, 120nS penalty.
At 1.0GHz, that's 120 clock cycles.
Starman97@Gmail.com (bring it on spammers)
and yeah, BFCs (big, fast caches) are far more important than fast main memory for the majority of applications. Nevertheless, these fast memories sell really well on the enthusiast market, where most people don't really know what a cache really is.
The Raven
Your DRAM in a PC is essentially an L3 cache. Your disk, an L4. With todays CPU's hitting 90%+ L1 cache hits, and 85% L2 cache hits what they've done is double the speed of 15% of your cache misses. BFD. Net overall system performance increase is maybe 5% depending upon your application.
"TV, a medium as it is neither rare nor well done." Ernie Kovacs
No arguments on active cooling. It seems more like a hobbyist/hot rod solution than a practical one.
This isn't a new DRAM chip. This is an ad from the fan and heatsink crowd.
Excluding situations in which you might be loading a new program and/or data specifically from the hard drive, wouldn't doubling the speed of the ram technically double the speed of all of your (L1 and 2) cache misses? (Of course this also assumes you have significant amounts of ram, such that nothing is loaded into virtual memory blah blah)
No, he's saying that more memory is better than fast memory. He might not need 8GB, but it's likely that 8GB of RAM would improve system performance better than doubling the speed of the ram.
The whole assumption is that anyone needing that much performance will be butting up against disk read bottlenecks due to swap anyway.
My question to programmers is this, Swap may have made sense 30 years ago, when ram was like $8/byte and not much faster than disk anyway, but in 2007, ram is ubiquitous and MUCH faster than disk. Why do we even have swap anymore at all?
Can you be Even More Awesome?!
The active cooling on memory is just like those cheap aftermarket rear fins you see on cheap riced out wannabe cars. All show, adds nothing to go.
No really, i have know about this for at least half a month, iirc to the ram is only usuable in highend intel MB's like Asus's Striker board which retails in aust for about 700ish it self.
Every piece of hardware will be outdated in 6months. So you can either never buy computer hardware because of worrying about it being outdated or just buy the hardware that you actually need.
...and that is all I have to say about that.
http://jessta.id.au
Because it saves you from a complete system lockup due to insufficient physical memory?
If you are using swap then, yes, you have way too little memory. I have a desktop with 1.5GB of RAM, and it never swaps unless I leave an ISO in tmpfs for a few days. On the other hand, under Windows XP it would swap regularly. There is no reason for a normal computer user to be using swap regularly, and if you doing something fancy (ex. DB server) you are probably going to get a lot of memory so you will still not be using swap.
Centralization breaks the internet.
....if you really care.
But not many people need this kind of performance. OTOH Low power is more sexy.
Fanless is where it's at.
Money is the root of all evil?
That it definitely doesn't do.
I have my box freeze once in a while due to RAM exhaustion. It goes into swap, and swaps so heavily I can't even move the mouse around anymore. The problem with swap these days is that it's horribly slow compared to RAM. While memory keeps getting faster and larger, the hard disk's speed doesn't grow anywhere that fast. So once you're swapping and using a significat amount of it (just 256MB say) everything grinds to a halt.
Current OSes seem to try too hard to keep stuff running. That might have made sense in the past, but I don't want that anymore. If something goes insane and tries to get all the RAM for itself I want it to die, quickly, without freezing the machine solid for several minutes.
Yet another whiny fan to sieze up and die in six months.
How long before they put active heatsinks on mice?
It's probably dirty. To clean it, deflate her and wash with warm soapy water, then hang up to air dry.
More wattage, more cooling, more noise = more speed?
Amazing!!! Congratulations, the marketing industry has clearly found your testosterone, and billed you for it.
How about more articles on the nice low wattage small and quiet machines? Unless you can beat a Mac Mini just don't bother.
- Adam L. Beberg - The Cosm Project - http://www.mithral.com/
Not necessarily. I too have 1.5GB of RAM. Windows and all the background stuff takes some 300-400MB (MSN, AV, Firewalls etc) and several hours of BF2 boosts the size well beyond 1.3GB. I tried without a pagefile and ended up crashing a lot after longer sessions due to memory problems. Newer games probably use even more. Sure, it's a gaming rig with full details and stuff, but the "monstrous" 2GB of memory isn't just for some exotic server use, gamers can really put it to good use too.
I just bought 4 2gig Gskill sticks and the BIOS isn't stable unless the 4th slot is empty. From what I've heard[*], this is a pretty common problem with filling up all for slots. Nice speed, but tell marketing they need to deliver stability also.
t ers.memory/browse_thread/thread/9dcf22f919ada367/2 30b6421faf43861
[*] - http://groups.google.com/group/misc.forsale.compu
boycott slashdot February 10th - 17th check out: altSlashdot.org
High Performance DDR2
There's a High Performance Dance Dance Revolution 2?
My question is, why does windows use the swap file, even when I have plenty of free physical memory.
...it makes it to easy for cats to catch them.
My question to programmers is this, Swap may have made sense 30 years ago, when ram was like $8/byte and not much faster than disk anyway, but in 2007, ram is ubiquitous and MUCH faster than disk. Why do we even have swap anymore at all?
Because RAM isn't quite ubiquitous, and because people still run out of memory even when they max out their systems -- or, for that matter, when they buy as much RAM for their systems as they can afford. Remember, RAM is only relatively inexpensive, when compared to the price of RAM in the past. Disk is still much, much cheaper. And when you do run out of memory, you'd much rather have your machine slow down while swapping, thus giving you time to kill processes or allow the machine to work through the problem, rather than simply crash.
Two relevant examples:
1) A co-worker of mine is a Windows developer. He usually has at least two additional copies of Windows running in VMs on his workstation. Now that he's supporting Vista, the 4 GB of RAM in his workstation (which is as much as it will take) isn't always quite enough. He would rather spend some time swapping than have his workstation crash.
2) I admin several Linux servers at our university that are multi-user statistics/computation servers. Anyone at the university can log in and run programs. Most of the users are graduate students using Matlab, Mathematica, SAS, Stata, etc., running large data sets and using lots of RAM. We're only budgeted for a certain amount of hardware, and as a result, our machines only have 16GB of RAM each. When a few users start really large jobs, and one of them uses up what's left of available RAM, we don't want the machine crashing and taking everyone's jobs with it. Likewise, if some doofus writes a program with nasty memory leaks, swap space gives us the time we need to go in and kill the offending process; other users who've got long-running jobs don't have to lose 2 weeks worth of work because we ran out of memory.
The plural of 'anecdote' is not 'data'. --Dan Kaminsky
Very. Memory IO is very important to performance. Intel, since the P4 has been trying to push the FSB frequency higher and higher, and using dual-channels to double the speed. AMD chose instead to integrate the memory controller onto the CPU, which reduced latency, and gave them a big performance boost. Even there, the only difference between socket 478 and 939 is the later has dual-channel memory.
It's extremely unlikely your computer actively operates on GBs of data at once. You probably have a couple hundred MBs of data that is really being used, and the rest can be swapped out with very little performance penalty. So, you'd have to have EXTREMELY little RAM for that argument to hold any water.
The reason people like to get excess RAM is because of caching... The more RAM you have, the less repeated reads from the hard drive. It works for servers that are always-on, or otherwise constantly run the same programs/data, but not so much for desktops. When you boot-up, it doesn't matter how much RAM you have, everything you launch needs to be loaded from the hard drive, and again the next time you reboot. No to mention that you probably run a range of applications, and open a large set of different files, that couldn't be cached in 8GBs of RAM (that's not even 1 entire DVD, for example).
Well you're definitely old. Even my very old DDR266 ran too hot, and really needed a heat-spreader installed. Even that only works because I already have good airflow in my case, and if my case fan goes out, my RAM will fry.
Whatever you may think about modern (hot) computer components, nothing made since your Tandy can operate without fans. Even the earliest PCs absolutely required a PSU fan. And if you want to go to extremes, get an incredibly low-end computer that can be run fanless, you're again at the mercy of one of the vents possibly being blocked, and quickly overheating the system. So, complaining that a fan on your RAM could potentially go out is nonsense, since the rest of your computer will burn up if some OTHER fan goes out.
Slashdot gets worse every day... Pipedot: News for nerds, without the corporate slant
I am _guessing_ that if the OS developers know that certain OS memory block uses are infrequent and access times to them are not critical, they can quite safely cache these data in swap, leaving more of the precious RAM resource available for application use.
But, I'm no OS developer.
My dad's method is to stick to win95 until forced to 98, XP until Vista, ect. Right now he is using 98.
The government can't save you.
By default, it lets processes overcommit memory. That means you can malloc more than there actually is. This is done with the expectation that programs allocate extra memory they don't actually use. Problem is that an excessive allocation succeeds, but then the system can't satisfy it, so it has to kill some random process.
Do this: This will turn off overcommit completely. When some program tries to request too much, malloc simply fails. No random processes get killed. The program that tired to allocate the memory is given a chance to handle the failure, unlike what happens with the default setting.
Ask and ye shall recieve: http://www.gadgets-reviews.com/index.php?page=post &id=166
ok, its not really a heatsink, but it is active.
11 was a racehorse
12 was 12
1111 Race
12112
At work my laptop runs Windows XP with Outlook, Firefox, and Eclipse at the same time and rarely use much more than half a GB
That is only true if you work on small Eclipse projects. Make that a few huge projects, and run a tomcat server on it and you'll burst the 1.4Gig in no time... What you run, is exactly what I run at work, but my usage is around 1.4Gig.... On bootup, Task Manager reports about 200Meg uses, and I'm fairly certain that there is no crap on it.
Ahhh...the great dumpster continuum. Many a free computer will be found there. -- sowth (748135)
"Why do we even have swap anymore at all?"
Swap allows for more effecient use of memory, which isn't as ubiquitous as you seem to think (we're constantly finding the 16G limit on sensibly priced/available hardware somewhat annoying) -- if you're pushing the limits of physical memory, swap allows for least used pages to be written out to disk to leave expensive and scarce physical memory available for things that actually need it. Go ahead - allocate and dirty a whole bunch of memory for a bit and force your nearest *ix box to dip into swap. You'll probably still find hundreds of MB swapped out days later because it's just been sat there doing nothing.
On-disk binaries and other mmapped files also make use of the semantics of swap; when you run an app, the system basically uses the on-disk executable and libraries as read-only swap, and any pages not being used can be discarded because they're "swapped". Similarly userspace apps commonly mmap's data files into their address space, basically using them as app-specific swap files, with the VMM taking care of caching data and flushing dirty pages in a manner hopefully appropriate to the system memory load. Swap space is exactly the same thing; it just happens to use anonymous space on an OS-wide file for general purpose use. If it's not being used, it's not hurting, and if it is being used, wouldn't you rather a few unused pages get written to disk instead of your app just outright failing? If not, well, it's your system, feel free to turn it off.
The memory companies seem to be fighting the Ghz wars of yesteryear. They release these "performance" products that boast tighter timings and higher clocks, that don't translate into significant real-world performance gains because the bottlenecks usually lie elsewhere, like the northbridge or on-CPU memory controller. Corsair strikes me as a big marketing machine with just a few uber-hyped products. Truth is, in my experiences I've seen more Corsair memory cause problems than the generic stuff, mostly because they often employ weird timings that are misdetected or even unsupported by the motherboard. The fact is that their target market is a bunch of Red Bull chugging gamer types, that don't know squat and think 1% is significant. They remind me of a certain subclass of audiophiles, people who have been caught in the sticky web of disinformation that's out there... people who will fight you to the death over the quality of their hand-made oxygen-free triple-plated phase-aligned one-way audio cables.
I can tell you quite honestly that if I had to plunk down an extra 200$ on my PC, I'd get the cheap ram and bump the CPU up a few hundred MHz. Specially tuned memory is for specially tuned applications, you know, like a real-time zillion-core supercomputer!
-Billco, Fnarg.com
That is exactly my point. About 15% of all memory access are L1 & L2 cache misses. You double the speed of only about 15% of your memory access. Taken overall your total system performance increases less than 10%. For the expense it is not really worth it.
"TV, a medium as it is neither rare nor well done." Ernie Kovacs
What ? Fan ?
So I won't come with corsair's typical LEDs / LCD display / Lava Lamp ?
Damn !
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
Say you want to load something to the video card. ALL of that has to go through the front side bus. Cache isn't going to help whatsoever. A faster FSB will increase throughput to the vid card though.
Gotcha. I was just confused on the 15% number, as I'm one hell of a poor man's EE. In that I'm not an EE at all.
Who modded this informative? It's totally wrong!
A fridge needs COLD. The gas is compressed, creating heat it has to get rid of (hence the radiator at the back). Then it's evaporated to create the cooling (to keep stuff cold). It has to get rid of heat, it sure doesn't need any extra heat!
And about needing less cooling during winter, that's at best negated by requiring much more AC during summer.
Besides "not needing heat during winter" is a lame excuse for using energy-wasting non-efficient electronics (like people used to say for the old Netburst junk). It doesn't quite justify buying it in the first place. It's not really saving you any money either (you just pay more on your electricity bill instead). And ow you also need more fans, better cooling/airflow/heat sinks and all that.
Well ok, the memory this fast is right now, but faster memory in general isn't. There's a reason why Intel and AMD use faster memory with newer generation processors. Though there may be rather large latencies reading from it, you still have to do so in the end. Big though your cache may be, it isn't going to hold everything. I mean a big L2 cache these days is 4MB. Often a single executable is bigger than that, never mind the associated data it wants. So you need main memory.
Well, one thing when you have fast memory, and a large cache, is you can try to make up for the latencies by batching operations together. You read large amounts of data in at one, perhaps even reading ahead. The quicker that read gets done, the better, thus RAM speed does matter.
There's no one thing that is absolutely the most important. You find that just increasing cache size only helps to a point. Generally, that point is whatever the size is on the higher end processors of the era. They aren't stupid, they pick the size for a reason. Likewise just increasing a CPU's speed (and by extension it's cache's since they are on the same chip) is worthless if the main memory can't keep up. So while this memory may not be useful for this generation of chips, it may not be long before it is. Used to be 800MHZ ram was for overclockers and such only. Now the Core 2 Duo likes it. Doesn't require it, but it is what is recommended for best performance (supports 533, 667 and 800).
A little math helps a lot.
Let's assume 1 cycle L1 and a 2 cycle access to a 64-byte/line L2 cache. So an L1 cache miss costs 8 cycles. So if you had 90% hit rate on L1, and 100% hit on L2, your processor will spend 8*0.1/(8*0.1+1*0.9) or 47% of its time waiting on cache misses. (Hyperthreading helps this a bit because the other thread might be able to work at this point). To simplify the next step, we'll say that this averages to an even 2 cycles per access.
Now lets say that the miss rate on L2 is 15% and that a line fill costs you 100 cycles. So the processor spends 100*0.15/(100*0.15+2*0.85) or 90% of its time waiting for main memory. I know some people don't want to hear it, but this is VERY VERY TYPICAL for most applications. Now double the memory speed. You haven't really changed the access time, but you have improved the time to get the second eight words chunk. So now maybe you have cut that 100 cycles to 75 cycles. You've increased the speed of the application by 0.9*25% or 22.5%
Eric
Support SETI@home
8 GB of that RAM,
... -and I might be set for Supreme Commander ^_^
Quad core CPU
and
a very fast GPU
urd
You probably have a couple hundred MBs of data that is really being used, and the rest can be swapped out with very little performance penalty.
All the more reason to make boards that take 1 gig of -really- fast RAM, and 32 (or more) gigs of slower/cheaper RAM for... times when you don't need that speed. Eliminate swap disk completely.
"If anything can go wrong, it will." - Murphy
Well, when you're talking about 1GB+ of RAM, try comparing PC-100 SDRAM to PC2-6400 and above, or hell, even DDR-400 for a more accurate comparison between low-end and high-end RAM from one generation to another. The onset of DDR brought about a big difference in system performance, and while DDR2, at the PC2-4200/4300 level, isn't that big a jump, the PC2-6400 level and higher are very much so. I've put together a few PC's with both PC2-4200 and PC2-6400 on otherwise-identical hardware, and system responsiveness is way higher with PC2-6400.
I suppose, though, if you're still using Windows 98, you're not too concerned with that, and the same goes for if you're just using the OS to write letters and play Solitaire. However, if you're planning to do any gaming at all, or any other system-intensive task like a render, encoding, and so on, you're looking at much faster performance for any given task, plus greater system stability and responsiveness while it's working. Of course, the type, brand, and tier of the RAM you're buying is also a concern here, but even OEM parts can blow the doors off of yesteryear's RAM - And to compare it to, say, 30-pin SIMMs wouldn't be a comparison at all. =P
It's all about the bandwidth - the input and the output - and not about the amount, these days. Back in the good old days, when people had the choice between 32MB and 64MB, RAM capacity actually meant something. Nowadays, we have the choice between 1GB and 2GB, and the only real difference between the two, to the layman, is that unless you want to run the latest and greatest in electronic entertainment software, work with CAD, run a server, or want Vista to run smokin' fast (4GB, please), there is no difference. To the end user, capacity is no longer an issue, since most non-Vista desktops run quite happily within 512MB, let alone the multi-gigabyte range. For gamers, 1-2GB (usually 2GB) is usually plenty, with 4GB being the current ceiling of pseudo-sanity (on XP, anyway).
It is true that the capacity would be an issue if we were talking about loading the OS into RAM or something like that, but as long as you're over 1GB, you should rarely be hitting swap in the first place. In the case of this particular RAM, you're looking at 2GB of ultra-fast RAM that would, for almost any standard usage, suffice with regard to capacity. I say standard, since again, servers and graphics workstations may have much more memory installed - and need it - than the average desktop PC. That said, most servers and workstations also require faster memory, and especially on the server side, where reaction time and concurrent processing is important.
Screw the rules, I have green hair!
That's fine, EXCEPT:
32GBs of slow RAM is going to take ridiculous amounts of motherboard real-estate, and added cost even for those of us who don't want it.
There's not much reason to put it on the motherboard, as SCSI or SATA would be fast enough for old, slow RAM.
32GBs of RAM isn't going to be cheap, no matter how slow, unless you do it on a small scale, so existing supplies of old/used RAM don't start running in short supply.
Slower/older RAM isn't significantly cheaper to produce, I'm afraid. I would know; I still check-up on prices of PC133 for older systems. Right now, 512MBs of PC5400 is $30, while PC133 is $40. Supply/demand applies, but you're still not going to see much savings.
Slashdot gets worse every day... Pipedot: News for nerds, without the corporate slant
Except you've forgotten that those 15% of the memory accesses take 90% of the time.
Support SETI@home
A few days ago I was running a numerical program that has a working set size of about 50MB. There is no way all these data can fit in the L2 cache, and memory bandwidth is crucial to its performance.
Damn. I just read something about "High Performance DDR2", and immediately thought of Dance Dance Revolution - imagine my disappointment when it turned out to be about memory instead.
butter the donkey
"Why do we even have swap anymore at all??"
You don't sound like a programmer at all, nor have worked with memory intensive applications. Programs eat up memory very quickly especially if you're multi-tasking or working on large images or video files, swapping is absolutely necessary for a whole host of applications. While you may think ram is "ubiquitous" and "large", it is actually tiny to the exploding amount of content. A 20 Minute Xvid encoded video is 170Megabytes, an uncompressed DVD is 4.3 Gigabytes, and lets not even get started with HD DVD and other high definition formats. For the most part most people are cheap and get what is cost effective. 512MB to 1GH is the current standard, not that much when you consider the code and other asset bloat of modern programs and games.
Kind of sad how memory speed has gone nowhere while CPU speed has raced ahead. The latency on memory is still 1990's numbers.
I am not a programmer. But I just don't see why augmenting my ram on disk does anything useful. All it really does is make my system appear to have more ram than it does, while causing the disk to thrash if the "extra" ram actually gets used for anything. If i need some kind of buffer space, then why not partition off a portion of the existing ram for that, add a return value to malloc for 'allocated from buffer area' or some such so I can get the buffer without sacrificing speed or HDD wear, and most importantly, not doing these things without warning the programs at all.
Your examples are silly, too. An "uncompressed" DVD may be 4.3 GB, but there is no reason to load the entire thing into ram at once, and anything that you put on the disk is going to be read *slower* than reading it from the DVD and decompressing it on the fly not only because of the vast difference in speed between the disk speed and the ram speed, but also because you have to write it before reading it. How much do you have to have decompressed ahead of time anyway? I assume it depends on processor and ram availability: if you're wasting cycles waiting on virtual memory thrashing, you might need to have a bigger buffer...
Video games shouldn't ever touch on swap space. They should be made to degrade or fail gracefully if memory is low: fps is far more important than graphics quality for anything more intense than sudoku.
If there isn't enough ram, programs should either deal with it intelligently or fail. After all, the people best able to optimize a given program for limited memory are the programmers themselves, not developers of some other project.
Can you be Even More Awesome?!
Why not just have the OS deny the memory request and freeze or kill the program that bumped it over the limit? A system lockup is something that happens on memory exhaustion only because that's how it was designed to operate. My crappy OS swaps everything to disk if I leave it on overnight, resulting in horrible performance in the morning. It wasn't doing anything overnight, but XP just likes to spontaniously swap stuff that isn't used, freeing memory for future uses at the penalty of slowing everything I already have open.
Learn to love Alaska