Any computer scientists (or anyone else) who can help me with this question?

I hope this is the right forum for this.When reading about supercomputers, the processing speed always seems to be measured in flops, kiloflops, megaflops, gigaflops, teraflops, etc. I've found a number of sites telling me that FLOPS = Floating Point Operations per second, but I can't seem to find anything to give me some comparative idea just how fast this is.I'm interested in seeing how much faster current supercomputers are than the average modern PC with, say, a 2 - 3 GHz processor. Could somebody give me an idea what the conversion rate is between flops and hertz? Or alternatively, point me to a URL that explains this?Thanks in advance!

You won't find any direct conversion between Flops and Herz.Even in closely-related processors the relationship isn't perfect - a Pentium 3 does a bit more than a Pentium 4 for the equivalent clock speed, but a Pentium 4 can be driven faster. AMD processors are generally faster than Intel for the same clock speed so AMD labels their processors with the clock speed of an equivalent Pentium - while the processor runs slower.If you are comparing very similar procesors then the clock speed will be a reasonable guide FLOPS are better if you want to compare the performance of quite different processors for the sorts of uses supercomputers get put to. It probably isn't that useful to home users because home users don't often do lots of floating point calculations.

It's not really the forum for this. But I'll answer anyway.As PaulK points out there is no direct conversion, but here's a site with some numbers:http://computational-battery.org/Maskinvare/Flops.html

PaulK, thank you for your help! I appreciate your explanation, even if I am more confused now than I was before, heh heh.But seriously, I'm kind of lost on where to go from here. Perhaps the best question I can ask, at this point would be...Is there any meaningful way, in general terms, to compare the processing speed of a supercomputer to the average PC? Or is it entirely dependent on the specific computers and/or components in question? It doesn't have to be precise, by the way. I would be quite happy just to be able to estimate.Thanks again!

Yes I know. I apologize. I understand the purpose of the forum and I generally prefer just to lurk, as I'm not much of a debater. Unfortunately I haven't been able to find anything on this and there are some very learned people here, so I knew some of the posters on the forum would be able to help. Again, I'm sorry. It's not my intention to clutter the forum with irrelevant topics.Also, thank you for the link! I'll check it out and hopefully it'll make things a little clearer. I appreciate your help!

The problem I think you're having is that "computing" isn't just some general task - different computer configurations are optimized for different things. Supercomputers aren't sitting there running Windows NT or a screensaver, they're running processes optimized for massively parallel processing supercomputers.It's not easy to compare. Your average supercomputer has 32 or 64 CPU's, usually chips that you're probably familiar with. That doesn't make them universally 32 times faster than your desktop for every single computation.Benchmarking (comparing computer performance) is a lot more complicated than it sounds.

A lot of supercomputers use "domestic" processors or relatives these days - especially if you include PowerPC and POWER. Although in the case of Intel processors they often use the Xeon server variants.So you could look at Home - | TOP500 which lists the top 500 supercomputers and has a lot of information about them.There are a lot of overheads coordinating processors so your home computer will do better than 1/1000th of the performance of a thounsand-processor equivalent On the most recent list number 500 has 128 Xeon P4 processors - and number 4 has 2500 3.06 MHz Xeon P4. And number 1 is 4 times faster than that at floating-point operations.(I don't have time to check but I think that the Rpeak values are the theoretical maximum so dividing that by the number of processors should get the rating for a single processor - the Rmax figure erpresents the top speed in "real" use).

Hey Crash! Whew, no kidding! I hope I didn't give the impression that I know a great deal about this. Just to make it clear (as if you couldn't already tell), I don't! I'm fairly new to the idea of supercomputers and obviously need to do some more research. I'd just assumed that there would be a simple conversion one could do to compare processing speeds. Heh, clearly it's not that simple.Oh well, thanks for your help Crash! I appreciate it!

PaulK, hoo boy! I'm in over my head! I'm not particularly knowledgeable on this subject and I didn't realize it was so complex. If I had I would have done much more research first. Now that I know, I have some reading to do. Ah yes, I've been there! I intend to go back and check it out in more detail, though. There seems to be a lot there to keep me busy. Great! Thanks! I'll have to remember that after I do a bit more reading. Once I get my head around their functioning a little better, I'll try it. Thanks again, PaulK!

FLOP is exactly has you describe it, Floating Point processes per second. The reason supercomputers are measure this way is because they are typically used only for crunching numbers.You need to know about the ALU that is designed in these processors, and how much money the company had to create the ALU. If they used certain logic gates as apposed to other logic gates. When your talking about making 1,000+ processors, the choice is narrowed down quite a bit. For example, it may be cheaper to use 'X' amount of XOR gates instead of 'Y' amount of NAND gates. But if you have 8 gates for a device instead of 10 its going to be quicker, but not necesarilly cheaper.Most of these supercomputers are custom desgined. They typically aren't the same CPU's you can buy off the shelf. They will have an entire team of Electrical and Computer Systems engineers who design specific memory modules/registers/buffers etc that suit the needs of the processes. These computers are usually VERY basic at the core, because they are typically not used for a 'multitasking' GUIs. They usually have only a few applications (IE. Add, Subtract, Load, Store, Multiply, Divide etc).So comparing the speed of a supercomputer to a "P4" is rather complex. You could desgin a 8bit CPU that can 'add' faster than a 64bit or 128bit CPU. It all boils down to how many gates the voltage has to travel through, and how fast the gates internal clocks are.The computer is only as fast as its fastest components =) Using a 1mhz IC chip combined with a 200mhz IC chip is going to give you bad results, and A LOT of these supercomputers are designed like this(for money reasons). It costs a whole lot more to buy 200mhz IC's than it does to buy 1mhz. (Just a hypothetical example, its a little more complex than this)Cheers,
Dextar

Hey dextar! Nice avatar! Yes, I realize that now. I originally thought it would be a simple matter of converting flops to hertz, but I can see now that there's far more to it. I'm doing a little research to familiarize myself with things. Heh, it's a slow process but at least I'm heading in the right direction.Thanks for your help dextar!

quote:

---

Yes, I realize that now. I originally thought it would be a simple matter of converting flops to hertz, but I can see now that there's far more to it. I'm doing a little research to familiarize myself with things. Heh, it's a slow process but at least I'm heading in the right direction.

​

Thanks for your help dextar!

---

Just read up on the different parts of a processor, and you can deduct a lot from that.No need to thank me, I love talking computers, and its not to often to have a conversation above laymen level =)Cheers,
Dextar

Hey thanks for the suggestion! I'll do that!It's going slow. I'm only learning in small steps, at the moment. But I'll see what I can find on the components of a processor and hopefully learn a thing or two from that. *Chuckle* I thank everyone that helps me out. It's just one of my various weird traits, I guess. I'm always genuinely appreciative of any help that I'm given. This will probably sound quite stupid but sometimes I worry that my unrelenting gratitude may come across as insincere or patronizing. Unfortunately I'm more or less stuck with it, since I just don't feel right about accepting someone's help without thanking them for it. Weird, huh? Heh heh, I'm afraid you'll probably go right over my head if you want to talk computers in more than laymen's terms with me. I can operate a PC as well as any average user, but that's pretty much the extent of my computational genius.Computer science really isn't my field at all, I'm afraid. I'm more well versed in other areas like physics, astronomy, etc. Mind you, I'm no expert in these either. I consider myself to be a reasonably well read layman, but compared to some of the people that post on this forum, I know nothing!Anyway, not to go against your wishes or anything, but thanks again dextar!