Slashdot Mirror
Tom's Reviews Kryotech's 1000MHz PC
GenBradly writes "
Tom's Hardware gives a review of Kryotech's newest model, the SuperG. They claim this new unit allows you to stablely overclock your Athlon to 1 GHz. Though the unit weighs 70 lbs, it's much quieter than previous [Kryotech] models, only slightly quieter than elevator music. It may be a bit expensive but I would do a lot for a 1 GHz computer." Lots of pictures, lots of specs, lots of tests.
Alpha's are running in the 800+MHz range (at least, the newer 21264s are). But they aren't the fastest, MHz for MHz. In fact, the per-clock execution speed of alphas is quite poor for integer and moderate for floating-point. MHz for MHz, MIPS are the fastest normal processors (See the top 500 list to see some absolutely insane hitachi machines with only a few moderately-clocked processors which are probably the fastest in existence). But yes, Alphas do run at insane clock speeds and have for some time.
This is like having a second refrigerator (a small one, but a refrigerator nonetheless) in your apartment/house/cardboard box. Add to this the habit of most geeks (like me) keeping their computer on 25 hours a day, and you get the idea...
I'm sure that we won't have to wait for 1GHz processors from both Intel and AMD for more than 6 months...
--
Yeah, I know. That was one of the biggest mistakes with the 386. Intel knew that FPU performance was important, and they were redesigning the FPU anyway (AFAIK, the 387 was incompatible with the 287). They should have abandoned the stack based FPU at that point.
People always talk about how great Intel's FPU speeds are, but that's only relative to other x86 chips. And now they're not even top there.
/peter
If you do the same thing at home, it's overclocked.
Actually, hard drive speed have been increasing (thus the need for faster interfaces - U2W SCSI / ATA66) One important factor in hard drive speed is the spindle speed of the hard drive. I have a few 5400 and 7200 rpm hard drives. The 5400 rpm hard drives are kinda warm after a few hours of use, but the 7200 rpm drives gets hot after a few hours of use. Now a 10000 rpm (currently the fastest afaik) hard drives... most of them require fans to keep them cool or at least some decent air flow. If you try to go any faster... say... 12000 rpm or 14000 rpm, umm... well... I think the hard drive will probably need it's own refrigiration device. =)
The importance of hard drive speed also depends on what you do. If you are running a webserver or maybe a database where you constantly need to access the data, then hard drive speed is very important. But if you are say... playing quake, then hard drive speeds are not that important because the hard drive is only used to load up the game into memory, once all the data is in memory, then the performance is mostly determined by the cpu speed and the video card.
_______________________________________________
There is no statute of limitation on stupidity.
Seeing as intel and AMD ar both working on 1 ghz chips, I wonder how many people actually buy these...
:)
Also, wonder how fast these things will be able to run a chip build to run at these currently insane speeds?
bash: ispell: command not found
This sig left intentionally blank.
Seriously. I've got a 300 MHz AMD, and the CPU speed suits me just fine. Meanwhile, my multiple 7200 RPM SCSI drives, NVidia TNT video card, and six expansion cards are all generating heat like crazy. I've got a temperature probe on the video processor, and it hovers around 110 degrees Fahrenheit with the window open. I've also got five fans in this beast, so it isn't lacking in air flow.
I want a Kryotech system that cools the whole case. Not chilling the CPU to sub-zero, but cooling the entire case to 50 degrees or so. I would be willing to pay serious money for such a system. Lower overall temperature prolongs life, reduces failures and errors, and can even improve performance.
dragonhawk@iname.microsoft.com
I do not like Microsoft. Remove them from my email address.
Does that mean I'd have to let my wife win and get rid of my 3640 sucking down 30 amps from the special twistlock plug we had wired?
The Open Genera Alpha implementation listed in the third reply is available. There is some talk of getting things moving again. I for one would like to see it come back to the land of the living. It's been the best rapid prototyping environment I've ever worked in.
Geez, most trolls at least try.
It is most amazing to me how shocked we are all at the idea of a $2500 computer. I remember when I bought by 'badass' P120 with 32 rocking megs of RAM, a 1.7 gig HD and a shitty 17" monitor that blew out 15 days after purchase for $2600. All that was one of the cheapest ones out of the old favorite Computer Shopper. Now we are amazed at a similar price. Kinda weird.
About a year ago I replaced the motherboard in my server with a dual PII-400 board. The board+CPUs+RAM ran me just a little bit more than a thousand (I kept everything else - disk, monitor, etc...)
That was a year ago. Although I haven't checked the prices lately, I would expect a dual PIII-500 to run for about the same. According to this article, the complete system runs about $2500, so I'd be surprised if a complete dual PIII-500 server can't be had for the same amount of money.
Well, I got curious, and looked up VA Linux's prices. They quote $3600 for a dual PIII-500, but that's with a separate SCSI controller, and an 18 gig SCSI disk. That's a thousand bucks right there. I wasn't able to find what Kryotech puts in this machine, I'd be very surprised if they bundle something similar. Also, VA Linux usually comes out a little bit on a pricey side, and almost always you can get a better deal on the individual components elsewhere.
If you're looking for a cheapest way to boost the power of your existing system, it'll definitely be cheaper just getting a motherboard+CPU+RAM, and reusing the rest of your equipment, instead of buying a complete system like that. You'll probably get better performance too in a multi-tasking environment. A single 1 GHZ CPU would will come out on top only in situations that involve linear processing. Try benchmarking 'make -j 4', for example. A single CPU won't buy you much there.
--
Hey, wait a second. Isn't this what the Java folks keep waiting for? :-)/2
While there is some value to a simple objective test of how fast Quake3 will run on the best system that can be put together, what Tom offers is comparisons between different processors. If these comparisons largely depend on the graphics card (and possibly its drivers) instead of the processor, then you are not getting the whole picture.
-
We cannot reason ourselves out of our basic irrationality. All we can do is learn the art of being irrational in a reasonable way.
There is no "one true bottleneck" that limits computer speed. Some apps are limited by disk I/O. Some apps are limited by memory bus speed. Some are limited by network speed. Some are limited by level 2 cache speed. Others are limited by CPU speed. It all depends on the app. Very few games are limited by disk I/O. Very few static page web servers are limited by CPU speed (or so I would hope...). And so on.
If a computer were only as fast as its (sp) slowest component, then we wouldn't have any need for L1 or L2 or disk cache, would we? Engineers have been working for years to ensure that computers aren't brought to their digital knees by their slowest components.
:), and continues to be.
Sure, engineers have been trying to work around these problems for years, in order to minimize the effect of the slow components on overall performance. For instance, many of the design tricks for CPU design involve trying to mask memory latency (ie caches, load bypassing, load forwarding, etc). But the fact is that you can only do so much to minimize these effects. They still present a bottleneck, and still impact performance. That's why Intel is pushing Rambus, and Via et al. are pushing PC133 SDRAM - because memory is still a significant bottleneck.
And caches as we know them today are only effective because much of the data used in present apps has a high degree of locality - ie the same few pieces of data get used repeatedly, so caches make sense. But the trend is heading towards more and more multimedia stuff. And multimedia (for example, video or audio) tends to have streaming data types - you process the data once, and don't use it again. Thus caches aren't all that effective for a lot of multimedia processing. In order to gain significant performance, memory bandwidth/latency problems need to be addressed.
Of course, one problem is that I/O doesn't get the glory that, say, CPU design does, so you don't have as many people wanting to work on it. I/O has been the neglected child of computer engineering since the days of Seymour Cray (who had some great quote that I can't seem to remember
Sure, it would be nice if everything was as fast as the CPU, but that ain't gonna happen
Maybe not. But if you improved the memory speeds by 10%, I'd bet you get better performance than by increasing CPU clock by 10%.
Increasing the processor speed can only gain you so much. Problem is, for a lot of apps, half of the work that needs to be done is I/O, the other half computation. So you get diminishing returns from increasing CPU speeds, because that 50% of the work from I/O isn't going any faster. Even if your CPU was 10^10000 GHz, if you can't get the data to the CPU fast enough, it doesn't matter.
True, but if anything those results will be slanted in Intel's favor. It's been widely speculated that Quake 3 - or any FP intensive app - won't be as "optimized" for the Athlon's parallel FPU architecture. So the fact that it still beats a Coppermine handily, to means, means it's either fast, or very fast. Either way, I want one :)
And again, I say, so what? He's testing real world performance using real world setups. The bottom line is that Quake 3 runs faster on an Athlon with a GeForce as opposed to a Pentium. I don't really care why it does that, I only care that it does.
--
"Some people say that I proved if you get a C average, you can end up being successful in life."
I think there is a world market for maybe five personal web logs.
You would think that the most respected source for this sort of tests on the Internet could get it right for once.
I agree. I have read Tom's on and off for a while, and am continually annoyed at their seeming lack of concern for anything but games. It is a fact that people do things with computers besides play games.
I have e-mailed Tom's several times asking them to use more standard benchmarks, each with no response or acknowledgement. Does anyone know of another, better benchmark site? One that compares heterogenous hardware, on unices, with more reasonable benchmarks? Anyone want to start a sitle like Tom's for Linux users? I mean, hardware vendors will often send you free "evaluation" hardware to play with. That could be a pretty good incentive. ;)
--Bob
1^2=1; (-1)^2=1; 1^2=(-1)^2; 1=-1; 1=0.
We are always hearing about the new speeds of the processors and how much better/faster they are than the previous processor. However, with every new CPU and the accompanying motherboards the same system bus remains: PCI, ISA (being phazed out), and AGP. I understand that AGP is continually getting faster (1,2, and now 4x), but PCI and ISA aren't getting any quicker. Esp. with the new wave of gigabit ethernet cards hitting the market, even if you assumed that the CPU and disks could/would push a full gigabit, when does the PCI bus hit the wall? Are there new busses in the works for the system internals, or will everything eventually be powered by USB2 or IEEE1394?
-R
Well the benchmarks still give only a 33% improvement over the trusty 550Mhz Celeron. A A dual Celeron 550 would still give you more MIPS. For encoding massive amounts of mp3s and rendering video in parallel the dual Celeron 550 has held its mark longer than any configuration since its time.
Seriously, though, this is neat, but not for what it is. Rather, if a slow processor can be overclocked to 1 GHz, by supercooling, then upgrading the processor to something that naturally does 1 GHz, and THEN applying the same cooling, should give you an AWESOME speed!
Besides, new processors aren't cheap. They can cost as much as much as a new computer, with the previous generation of processor on board. The cooling might work out cheaper, in the short term, for the same performance.
It's a small world and it smells funny; I'd buy another if it wasn't for the money; Take back what I paid (SoM)
In true Tom style, he goes ahead and sticks a geometry accelerating Nvidia GeForce card in the computer he uses to test the 3d performance of the processor (effectively testing the card rather than the processor). Tom has been nutorious for always choosing setups that create bottlenecks in the wrong places, in fact, his 3d-card tests are usually the best places to look for processor performance, and vice versa...
You would think that the most respected source for this sort of tests on the Internet could get it right for once.
Also, since this is Slashdot we are about to get a hundred posts saying something along the lines of: Processor speed doesn't matter, its X (replace x for "harddisk read/write","bus speed","cache memory" etc). Don't believe them. Yes, there are applications where these things matter more (specifically server activity in most cases), but for 90% of us the only applications where speed is an issue at all any more are the 3d apps where it is all about processor speed (except in the case of geometry acceleration as covered above).
-
We cannot reason ourselves out of our basic irrationality. All we can do is learn the art of being irrational in a reasonable way.
With that sort of system, are they completely sure of correctness and stability?
For instance, is it still inadvisable to run computational tasks where the arithmetic *must* be reproducible on other platforms, and where the system must be able to keep this up for, oh, a week while a job completes?
Only the dead have seen the end of war.
3DNow is floating point. It is, however, hampered in part by the FPU architecture, because it uses the FPU stack for it's registers, IIRC. That means that as soon as you need to do something that 3DNow doesn't do, things get messy. Or so I believe, I haven't looked closely enough at it to be certain of this.
3DNow also only supports single precision. I don't know what "guaranteed accuracy" means here. I seem to recall that the iterative square root and reciprocal algorithms can give you 23 bits of precision, though.
This is all based on reading the 3DNow spec for the K6-2 a long time ago, so it might be wrong or it might not apply to the Athlon.
http://www.cpureview.com/ is a site alot like you're talking about. It doesn't usually have server type statistics, but they do show how long a kernel takes to compile :)
plus, the site admin reads slashdot.
I'd just like to thank Kryotech for naming it after me. I'll even let them buy my url www.SuperG.com for a cool million.
This is a joke, albeit a poor one. Do NOT send any death threats, or intellectual property/copyright infringement lawyers.
AFAIK, the two reliability factors cancel out - the refridgeration helps the reliability, whilst the overclocking stresses it. Now if you were to overlock without refridgerating, as many people do, then I think you are playing with fire.
Like another poster stated, I'm not very impressed with overlclocking in general - I mean if you were talking about 100%+ performance improvements then I think it would be worth the trouble, but if you're only getting 10-30% improvement over a normal system then I'd forget about it and go buy a multi-processor system or cluster.
Donte Alistair Anderson Roberts - hi son!
Karma: Chameleon
Right now, it isn't the speed of the CPU that is slowing down computers... Nor is it the Bus (Most of the time). It's the HD access speed. HotlineHQ, a compny I used to work for, is a great example of this. The slowest part of the whole chain on our webserver is the hard drive (Ultra-wide SCSI II). and the server is only a 266 (Granted, it runs normally at 75-90% capacity). Hard drive manufacturers keep insisting on Larger drives, not faster ones. So we have this enormous riverbed for information to flow through, but a dam up the stream is preventing the full usage.
-- "Our job is not to make the incredible possible. Our job is to make the impossible credible."- Jerry Olivieri
Great, now I can take my motherboard out of the freezer and pound that vodka I was using get the cooling for those last few Mhz!
I obviously expressed myself somewhat unclearly. Alphas are some of the slowest chips when you compare them to other CPUs running at the same clockspeed. What I meant to say was that Alphas have usually had the fastest clocks among the mainstream CPUs.
But it looks like x86s are overtaking them. This is quite an accomplishment since an x86 can do quite a bit more work per cycle than an Alpha.
/peter