So, do you mean investing on the Intel 6xx CPU serie is not worth the additional cost vs. the old 5xx in a 64 bit upgrade perspective but should be also better to skip the AMD 64 and going to the Opteron 800 path?
Yes, but the downside is that motherboards that support 800 series AMD processors are considerably more expensive than the 200 series AMD Opterons or EMT64 motherboards. To say nothing of the extra cost of the 800 series CPUs and the extra cost of running a 64-bit version of Windows. BTW, AMD64 has nothing to do with being Opteron 200 or 800s - there are AMD 200 series Opterons that are AMD64. nitro
Nitro, you cover software and dbases in this thread too so i thought i might add this link to this new type of real time on the fly dbase software... bit costly - but if one could find the right place... anyway... i was wondering what you thought of this... http://www.streambase.com/www/products/index.html i have no connection to these guys at all i am just trying to think of a way i could apply this to mutiple data streams to get an edge... probably nothing... hope its interesting... cj
Nice link for benckmarks, but this isn't really new. A days worth of DOB INET data pumped through our custom DB takes only a couple minutes. And that's with a single P4 1 gb ram.
What could be so much better? The big deal when we jumped from 16 to 32 bits was memory addressing, INTC had us in that stupid 16+4 memory addressing thing where every memory spot had like a dozen different possible addresses. Man, that sucked. But at 32, what's that? Like 4 gig a thread? Geez, that's plenty of RAM for today, IMO. The only thing left is math at 64 versus 32. And yeah, it'll be more precise. 64 bits is 18,446,744,073,709,551,616 while 32 is 4,294,967,296. Now folks who watch the stars, weatherdorks and folks who fold up protein might like that improved accuracy, but are most folks gonna notice an improvement from 1 in 4 billion? 4 billion is how many folks lived on the Earth a couple decades ago, for crying out loud. On the other hand, once we move to 64 bits, then max memory becomes literally "a crapload". Memory limits are a thing of the past, and maybe we can do cool things like only use 1 out of 10 possible addresses (like the post office does with street addresses). This would help us ID memory problems, which seem like the biggest problem when writing software. Dunno, but it sounds like 64 bits is no big deal. And I don't see a big need for the next gen at 128 bits.
This is a big thread so I don't expect people to read through everything that has been said previously. As I mentioned in previous posts, if you have anything but a computer with the 800 series AMD processors, unless you need the extra addressing that 64-bits gives you, it won't make much of a difference from a performance standpoint to go 64-bits. Further, if the extra addressing is all that was important, you could save yourself a ton of money by just getting a 200 series AMD64 Opteron or an INTEL EMT64 Xeon. However, in all tests that I have done and as previously stated by Tyan to me and verified by me, the 800 series Opterons seen in QUAD (or more) systems when run in 64-bit mode are _considerably_ faster than when run in 32-bit mode. As it was explained to me and posted in this thread previously, this has to do with the fact that an extra cache line gets turned on in 64-bit mode that is not present in 32-bit mode on the 800 series Opterons. This more than _doubles_ the memory bandwidth of the QUAD Opteron computer over running it in 32-bits and blows away the performance of like 64-bit systems from INTC in 32 bit or 64-bit mode. Also, perhaps you trivialize the potential for extra memory address space use that 64-bit gives you. While you may not see a reason to have 1 Terabyte of physical address space (RAM), it doesn't mean it is not useful. nitro
Actually it's not that simple. Those numbers are for integer arithmetic. The accuracy of single precision floating point operation (32bits) is nothing like 1 in 2^32. While the accuracy of a single floating point operation may be acceptable, when multiple arithmetic operations are performed rounding errors can quickly add up to cause serious problems. A simple example is an exponential moving average where errors can accumulate indefinitely. In all such cases double precision floating point (64bit) is much preferred. The new 64 bit CPUs are faster at handling doubles. I agree that for most PC users there is little to be gained at the moment in spending money on 64 bit, but the increased address space, better memory performance and improved arithmetic performance will be useful in quite a few applications.
Thanks for asking. It is running systems smoothly. My only gripe is that I am still running 32-bit software. That will change very soon when MSFT introduces Visual Studio 2005 and .NET 2.0. That it is not running in 64-bit mode is not the fault of the machine, but my attempt at adapting 64-bit was a bit premature at the time because I am running MS-Windows, and under Windows64 I couldn't even get 64-bit drivers for some of the NICs in the machine. Until that time that I go 64-bits, I am under using the machine considerably as explained in this thread. I have learned an enourmous amount from attempting to do what I want to do with it, so the time was not wasted. nitro