Slashdot Mirror
Intel Releases "Fastest Chip Ever"
mao che minh writes "From News Factor Network: Intel has released the world's fastest chip ever. The new P4 runs at 3.06GHz, at 3 billion cycles per second. Man, and I'm still squeezing the last bit of life out of my Pentium 233!" Tom's Hardware already has a review up about it, and it looks to live up to most of the hype.
Intel's Pentium 4 3.06GHz: Hyper-Threading on Desktops
No, Intel is also releasing new ARMs and they're not the fastest chips ever. Neither is the Transmeta Crusoe.
-73, de n1ywb
www.n1ywb.com
See this article from Tom's Hardware.
Sadly this trend won't go away anytime soon. When you pack that many transistors running at ultrahigh frenquencies in a tiny package you have to pay somehow.
I agree to your points but i would like to say something. A faster CPU it would be good for multimedia usage (something that the average person would care about). For example ripping DVDs or CDs, or as you mentioned rendering. You can't tell the difference between your 600 MHz and 2 GHz PCs because of several reasons. One that comes to mind is the hard drive bottleneck The IDE drives today are just too slow for the current systems. I believe that in order to build a PC that takes full advantage of the processor power and memory bandwidth it would be expensive. but as you know The average joe hears 3.06 GHz and thinks that is faster than his current PC. He doesn't think about other limitations of the system as a whole. Which if you think about it, boils down to marketing.
We break a new GHz barrier every month?
What GHz "barrier?" It's not like 3 GHz was theoretically impossible or anything. This is just a matter of making something go slightly faster than it did yesterday.
Or is it the big round number that impresses you?
I write in my journal
Your text editor probably won't get any boost from another processor, but if you're setting up a server then there's nothing to worry about.
I have two dual processor machines and they run at 3.2G's but I don't think they would fare all that well against a single processor machine running at 3.06G since not everything can use both and it doesn't seem like all that much uses both effectively.
Actually, the hyperthreading only helps in apps that support hyperthreading. Your dual processors are hyperthreaded. So any hyperthreading app that takes advantage of the P4 will also take advantage of your dual processor setup.
I imagine two different processors would be much better than 1 hyperthreaded processor.
Also, they only mention a 25% performance increase. Dual processors running hyperthreaded apps have at least a 60% performance increase. However, I bet this P4 would beat your machine in non-hyperthreaded apps.
If you don't understand any of my sayings, come to me in private and I shall take you in my German mouth.
i've had a developer maryville board on my desk at work for the past 2 months (p4 2.8ghz). my experience with it so far hasn't been particularly impressive. i mean, it presents itself as 2 cpus to the underlying os (works w/ xp, .net rc1, and linux), but when you do something that actually taxes both cpus (make -j8 bzImage or what have you) there's a lot of thrashing and no true performance gain. i like the idea that no one program can totally lock up your cpu (netscape / q3 / X / etc), but i haven't seen any gains in day to day use.
i'm curious how oracle / msft will deal with the licensing issues that will come about from presenting virtual cpus.
-BlueLines
--BlueLines "The cost of living hasn't affected it's popularity." -anonymous
Linux supports HT. No OS as far as I know sees much benefit from it.
The difference between a "process" and a "thread" is pretty small. A thread is just a process with shared page tables, for the most part. This means that there's less overhead switching between two threads, since you don't have to flush the TLB and caches. The processor per se knows absolutely nothing about any of this - it just knows when the OS commands it to flush the TLB and the caches, and change the page table addresses.
The basic point of HT is that it's sort-of another CPU, but it's just leeching unused resources from the main CPU. So, the scheduling logic in the OS needs to understand that it's not a real CPU, and thus should be grouped with the real CPU it's associated with. Linux 2.5/2.6 will support these tweaks, with 2.4 you'll need some sort of patch currently. Without the tweaks, you still get HT, it just doesn't help much.
But really, it never helps that much. Don't expect a 2x speedup or anything, even if your system is running heavily threaded applications.
while you are of course correct, one may be ble to overclock the overclocked by using super duper cooling. the limit of overclocking is limited often by heat, so if you can get rid of more heat, you might be able to squeeze a bit more out of a given chip.
for practical purpouses you are right, though. there is absolutely no reason you would buy this chip if you wanted to overclock it.
If you had nuts on your chin, would they be chin nuts?
The most interesting part of the reviews posted are the comparison between Hyper-threaded and normal mode. These nice graphs show that in all but one case, the speed is not harmed by having HT enabled, and indeed it improved the performance by up to 20%.
This will not make a single process speed up, but will make systems seem faster, as it is rare that you are only doing one thing at a time.
Yes, kernel 2.4.something apparently supports it. The only Microsoft OS which supports hyperthreading properly is Windows XP. Windows 2000 and below doesn't utilize it.
So it's either Linux or XP, as far as I know.
-- Jim
You are confusing the length of a radio frequency wave with the frequency with which somthing processes.
however, since it's a joke, I'll allow it.
If you had nuts on your chin, would they be chin nuts?
Actually it's a little more advanced that that.
Basically, you have all these different units for doing math in the CPU, and if one thread is only using say the interger units and another thread want to do float-point math, then the processor can actually let both run at the same time by scheduling the instructions properly. Normally, is one thread was all interger and one thread was all floating point, then the CPU would have to do a context switch to be able to run one type of instruction or the other.
It's a little more complex then that, but it's more then just holding one app in the hole. But yeah, for those running FPU intensive appls like seti, while doing normal stuff should see less slowdown.
No, he isn't confusing anything at all. The magnetron in your microwave oven runs at 2.4 GHz. This chip's internal clock source runs at 3 GHz. Simple as that.
I've got 85 Dual 2.4ghz Xeons running for 2 months now with HT enabled (both Linux + Win2k), and I concur. While each box appears to have 4 CPUS if you query the OS, running even make -j5 bzImage thrashes the heck out of the systems, negating any possible performance gain.
Not first. Xeon has been shipping with hyperthreading for a while now.
Actually this chip overclocks pretty well, [H]ardOCP got it up to 3.68 GHz air cooled and 3.82 GHz water cooled. Not bad at all. http://www.hardocp.com/article.html?art=Mzg4
mini-itx form factor, integrated video/ethernet/fanless cpu - just add memory and storage. link
if you wanted to go all-out on skipping the moving parts, you could run the os on compact flash using an ide to cf adapter from pcengines.com and use a cupid case with a dc power supply. just make sure to disable writing, or you'll wear it out! use mfs or a (non-essential) extra standard hard disk for data.
If you do a google search on optimal pipeline depth you'll find some good results.
-- Erich
Slashdot reader since 1997
This thing is faster. English summary at the bottom.
FoundNews.com - get paid to blog.,
Woo hoo! I might just get some "Informative" karma over this one... :)
Here's the deal: the DOOM III demo was a debug build. If you've got it, do a "strings" on it. You'll see a bunch of debug symbols.
That means no optimizations, and tons (I mean tons) of code to make tracking down problems like memory leaks easier. That kind of build will naturally munch processor cycles like crazy.
Corroborating evidence: the alpha is very CPU-bound, which should be surprising given how the algorithms it uses for rendering eat GPUs for lunch.
All the same, with features like per-poly collision detection, I expect the final version to do much better on a 3.06GHz chip than a 1.2GHz chip.
</off-topic>
I got my Linux laptop at System76.
Read the article.
The chip has 2 cores. 1 is used for 1 cycle, the other for the next.
Thus the transistor switching rate limitation is overcome and you appear to have a dual-processor machine (aka hyperthreading).
The best of both worlds!
I believe Apple is the "first" to start making use of the video card's GPU for day-to-day stuff.
*ahem* SGI's IRIX has been doing this for more than a decade. Their systems have always amazed me - just today, in fact, I managed to get an old Onyx system working. It's got a pair of 75mhz r8k cpus and a RealityEngine2. That's not a typo - 2x75mhz. Even with such slow CPUs, the user interface is lightning quick because of how well the OS makes use of the video hardware. Granted, the r8000 was a very unusual CPU in how effecient it was per clock, but still...
If you're talking about the Athlon, the problem is much more fundamental than heat, it's a signal distribution problem. Basically, the chip is running so fast that the time it takes for a signal to get from one component to another is more than a clock cycle. This is why with the latest release of the AthlonXP, AMD had to add more layers and do more wiring optimisation to shrink the effective distances between components (closer = faster signal propagation, obviously).
The P4 is capable of handling much higher clock rates than the AthlonXP, since the NetBurst architecture isn't designed with the assumption that all signals will propagate within a single clock cycle. My rough calculations show that the P4 could probably be clocked up to about 30ghz before you hit the same signal propagation issues the Athlon is having now. Of course, there are more traditional overclocking concerns between 3ghz and 30ghz. :P
How can this gibberish be moderated 'Informative'? There's no such thing as a 'hyperthreaded' app.
Intel has a nice tutorial on the subject.
I know that of the Microsoft OS's, only the XP family supports the Hyper-Threading. I couldn't tell you if any other OS's support it.
Distilled down, the processor creates a virtual or logical second processor which assists it in using underutilized resources.
A lot of multimedia vendors would be interested in this, a lot of gaming vendors will jump at this.
'Pleasure is the Disease, Pain is the Cure' - Lilith
K7 processor manage to beat Pentium processors running at the same frequency precisely because of IPC (they get more work done per cycle)
The Raven