Catch up on stories from the past week (and beyond) at the Slashdot story archive

 



Forgot your password?
typodupeerror
Check out the new SourceForge HTML5 internet speed test! No Flash necessary and runs on all devices. ×

High Performance DDR2 Memory Breaks 1.25GHz 104

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."
This discussion has been archived. No new comments can be posted.

High Performance DDR2 Memory Breaks 1.25GHz

Comments Filter:
  • by scgops ( 598104 ) on Saturday March 24, 2007 @11:34AM (#18471195)
    Lovely speed, but I wonder what all that heat output will do the ambient temperature.
    • by klingens ( 147173 ) on Saturday March 24, 2007 @11:38AM (#18471231)
      Watercooling for memory is only a question of when, not if :)
      • Re: (Score:1, Informative)

        by Anonymous Coward
        Yes, it comes with A/C, but you may not need it. I'm running an 800Mhz version of this memory and they're not even warm to the touch.

        Watercooling for memory is already available, check products from Koolance.
      • Re: (Score:3, Informative)

        by Unique2 ( 325687 )
        Actually, it's a question that has already been answered [ocztechnology.com].
    • Re: (Score:2, Insightful)

      by cyrtainne ( 1078481 )
      They better have some good case fans. Doesn't seem to be very efficient with all of that heat output. You could probably heat one room with that system - a lot of electricity being turned into heat instead of processing power - I wonder how much is lost.
      • by Rallion ( 711805 )

        You could probably heat one room with that system - a lot of electricity being turned into heat instead of processing power - I wonder how much is lost.

        Well, I hate to be annoying, but electricity -> processing power is not a conversion that happens, because, obviously, 'processing power' is not a form of energy. ALL the electricity that your box consumes turns into heat, noise, and some light. In terms of energy, the processing is just a byproduct of the to-heat conversion.

    • by isorox ( 205688 )
      Dunno, but my boiler's on the fritz and I'm thinking about overclocking my PC to keep me warm, this mory can only help!
    • Re: (Score:3, Informative)

      by master_p ( 608214 )
      Actually, having the PC open from one side, you don't need a heater in the cold winter. And in the summer, you can use the heat to powerup a small refrigerator, so you can have your beers near to the computer.
      • And in the summer, you can use the heat to powerup a small refrigerator

        Can you explain? I don't understand: a refrigerator outputs heat to cool the inside of the refrigerator. That's why the "radiator" at the back of your fridge becomes hot. Inputting heat isn't going to help a refrigerator at all.

        • Re: (Score:3, Insightful)

          by JRHelgeson ( 576325 )
          There are propane powered refrigerators that heat a mixture of primarily Ammonia. After the boiler it goes to a condenser where the liquid is held under pressure. The evaporator drops the pressure, and with pressure drop the temp drops as well providing ice cold temps so you can have a propane powered fridge or freezer.

          http://www.propanerefrigerators.com/how.html [propanerefrigerators.com]
          • Re: (Score:2, Interesting)

            by jawtheshark ( 198669 ) *

            Hmmm, I wasn't aware of that technique.... Still, according to the article the heat need to be dissipated from the absorber. I don't know if you could use a CPU to heat the ammonia mixture, since the CPU needs to be cooled. I guess that using it as a heat source isn't sufficient to cool it.

            A new market for cases with built-in refrigerator? ;-) If it would work, I think someone would have done it by now...

    • (Title) come with an air conditioner

      What ? Fan ?
      So I won't come with corsair's typical LEDs / LCD display / Lava Lamp ?
      Damn !
  • by Anonymous Coward on Saturday March 24, 2007 @12:02PM (#18471401)
    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.]
    • by LarsG ( 31008 )
      I can't find registered ECC DDR2 faster than 667 MHz. Why?

      ECC is pretty much only used on servers, and server types are generally more concerned about stability, uptime and low heat than cutting edge speed.

      Anyway, looks like Kingston has some PC6400 DDR2. The part numbers I found were KVR800D2E5 and KVR800D2D8P5. Their website seems to have problems, so can't provide direct links.
    • Re: (Score:3, Insightful)

      by DaleGlass ( 1068434 )
      Indeed. I've been badly burned with memory problems before, so now I only buy ECC.

      I had a particularly nasty incident. My firewall had been running for months without problems, until one day it crashed. I thought oh well, maybe it hit an obscure kernel bug. Rebooted it. Several days later it crashed again. Rebooted it again. The next time it crashed but didn't boot again as it had corrupted its disk, and I had a really fun day reinstalling it with no internet connection.

      My current box takes DDR2 800. I was
      • What motherboard are you currently using? I never seem to be sure whether regular mobos support ECC or not. My ASUS A8V has ECC options in the BIOS but somehow I doubt that reflects what the board & chipset can really handle.
    • http://www.pricegrabber.com/p__Kingston_2GB_PC2_6 4 00_800MHz_240_pin_Registered_ECC_Parity_Dual_Rank_ x8_CL5_DIMM_Kit,__29356668/pid=kingston/type=2 [pricegrabber.com]

      Kingston 2GB PC2-6400 800MHz 240-pin Registered ECC Parity Dual Rank x8 CL5 DIMM Kit

      Price Range: $293.70 - $539.99 from 11 Sellers
    • by amorsen ( 7485 ) <benny+slashdot@amorsen.dk> on Saturday March 24, 2007 @02:55PM (#18472803)
      [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.]


      Can I convince you that you don't need registered RAM? It isn't the ECC that is killing speed, it's the buffers.
      • I've done some amount of experimentation on this. A lot of times, if a system requires registered RAM, there's usually nothing you can do, it needs registered RAM. Registered RAM in a system that doesn't require it also doesn't seem to accept it either.

        Oddly, a chipset that will accept ECC memory on the data sheet might actually be implemented onto a board in such a way that the board won't accept ECC memory. Mixing ECC and non-ECC and registered and non-registered type memory doesn't seem to work either
        • by amorsen ( 7485 )
          Bit flips used to be a big problem. Redesign of the individual memory cells as well as picking non-radioactive casing (duh!) has helped a lot.
  • by Anonymous Coward
    Hey, didn't you see the sign? You broke it, you bought it.
  • by Glowing Fish ( 155236 ) on Saturday March 24, 2007 @12:20PM (#18471535) Homepage
    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?
    • How relevant is it to have memory that is this fast?

      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.

      if there isn't enough of it, your computer has to read

      • 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.
        • All the more reason to make boards that take 1 gig of -really- fast RAM, and 32 (or more) gigs of slower/cheaper RAM

          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 d

    • 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 sho

  • pointless (Score:5, Informative)

    by starman97 ( 29863 ) on Saturday March 24, 2007 @12:21PM (#18471541)
    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.
    • 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.
      • BFCs (big, fast caches) are far more important than fast main memory for the majority of applications.
        So why are multi-core chips taking over the home market? Why not just use all that chip real estate for more cache instead of a second core?
        • Re: (Score:3, Interesting)

          by Anpheus ( 908711 )
          Because there are two ways to beat the latency problem. One is to get the information closer to the chip (cache) and the other is to allow the chip to do something else, or several something elses at the same time (multiple cores, multiple execution units of various kinds, etc.) If you haven't noticed, the latest Intel chips sport ridiculously large caches (they're up to 12-16MiB now) and have 4 cores.

          So... they're still using a lot of real estate for cache.
          • Re: (Score:3, Funny)

            by timeOday ( 582209 )
            I associate oversized caches with Intel's "Extreme" line of processors, which (for some reason) offered virtually no performance boost.
          • Re: (Score:3, Informative)

            by Jeff DeMaagd ( 2015 )
            The Itaniums are up to 24MB, but the x86 quad cores are "only" up to 8MB total. The next generation Core processors will probably go up to 16.
      • 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 f
    • by Ramble ( 940291 )

      For one thing, increasing the clock speed would decrease the cycle time for one clock, therefore lowering latency.

      And another thing, lots of peope buy Dominator and other high-performance brands because of the overclocking potential. Faster RAM will help you clock your processor to higher speeds. Period.

      • by Rich0 ( 548339 )
        Faster RAM will help you clock your processor to higher speeds. Period.

        Well, that is mostly because of locked frequency multipliers. The only way to increase CPU speed is to speed up the bus, and that means speeding up your RAM/etc. It is really an artificial limitation - if you could adjust the multiplier you could just overclock the CPU - and just let it wait longer for memory fetches if necessary.

        Sure, faster RAM is better than slower RAM - the question is whether this is the best place to spend your m
    • Interleaving can help out there.
  • 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.
    • 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)
      • 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.
        • 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.
        • by SETIGuy ( 33768 )
          Except you've forgotten that those 15% of the memory accesses take 90% of the time.
    • 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.
    • 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.

      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*

    • 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.
      • by badhack ( 557341 )
        Unless your program was written by a moron your dataset was probably worked on in chunks, thereby reducing the impact of cache misses. In additional, memory is fetched and stored in the cache in blocks further reducing the impact of cache misses. The effect of cache misses becomes smallish. I would be more worried about 6-16 cycle latency of L2 cache HITS.
  • by Animats ( 122034 ) on Saturday March 24, 2007 @12:39PM (#18471683) Homepage

    This isn't a new DRAM chip. This is an ad from the fan and heatsink crowd.

  • 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.
  • ....if you really care.
    But not many people need this kind of performance. OTOH Low power is more sexy.
    Fanless [logicsupply.com] is where it's at.
  • by yeremein ( 678037 ) on Saturday March 24, 2007 @01:28PM (#18472107)
    Yet another whiny fan to sieze up and die in six months.

    How long before they put active heatsinks on mice?

  • 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.

  • 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.

    [*] - http://groups.google.com/group/misc.forsale.comput ers.memory/browse_thread/thread/9dcf22f919ada367/2 30b6421faf43861 [google.com]
  • High Performance DDR2

    There's a High Performance Dance Dance Revolution 2?
  • 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 b
  • 8 GB of that RAM,

                      Quad core CPU
                  and
                                    a very fast GPU ... -and I might be set for Supreme Commander ^_^
  • by asninn ( 1071320 )
    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.
  • Kind of sad how memory speed has gone nowhere while CPU speed has raced ahead. The latency on memory is still 1990's numbers.

There's no sense in being precise when you don't even know what you're talking about. -- John von Neumann

Working...