Do Newer Processors Equate to Slower Applications?

[brad jones]

I've posted my newest article on DevX (which is a site I also oversee like VBForums). If you like the article, please consider a Digg or DZone vote (links at bottom of article). If you don't like it, let me know that as well!

Thanks.

Brad!

Do Newer Processors Equate to Slower Applications?

While the promise has been that with new computers come more power and faster overall performance, it's likely that the speed is about to hit a brick wall. Even better, individual core speeds might be slowing down.

http://www.devx.com/enterprise/Artic...k=DXRSS_LATEST

[Slaine]

Interesting article Brad.

I have posted it to slashdot - thinking it might interest the folks there - hope you don't mind.

[Hack]

Very interesting Brad, and I like the "cars and lanes" analogy.

[RhinoBull]

So, what's your point - I really didn't get it.
If it's about modern processors being slower or at least not so fast I find this opinion amusing but I do respect it. Sorry.
The actual thing you should be looking at is the entire package - processor by itself doesn't mean much until it is part of your system.
You could have 3Gh Pentium 4 and say 256 MB of RAM and older but still compatible base board (or mother board if you will) - whould that combination be productive? I don't think so.
If you add another 1GB (or so) of RAM you'll be saying "god, it's flying now!).
Analogy with cars doesn't fit me either - it depends who's driving it.
You can have two pro racers racing each other at say 150 MPH with full confidence and at the same time and on the same road someone going 30 MPH with his hands all shaking up...
Two cores are better than one and they are also much more productive.
You will notice it immediately if you do lots of graphics, video, etc type work that does require intense processor involvement.
Besides, the article sounded like "anti-promotional" for modern computers so some home users may misinterpret it I'm afraid.

Best regards.

[brad jones]

So, what's your point - I really didn't get it.

I agree. I don't think you got it....


Deluting the article down to its core point:

Current processors are getting slower in single core speed, so if your application is CPU-intensive, then you can't continue to build single threaded (processed) applications or your application is likely to run slower.

Obviously, if your application is constricted due to other issues (such as I/O), then a slower CPU might not matter.

[Slaine]

I think Brads point was exactly "having to look at the whole package" but folding the developer into that package.

As system architects we have to be aware that we can no longer rely on our threads having free reign of the system - instead we have to find a way to leverage our solutions into a world of parallelism in order to take full advantage of upcoming hardware.

[Hack]

What happens when the speed of a processor can no longer go any higher? How many developers rely on the additional boosts from new computers? Do you rely on the additional boost?

This really caught my eye as I've known developers who have done just that.

Of course, I'VE never done that. *whistles and looks away*

[RhinoBull]

I think Brads point was exactly "having to look at the whole package" but folding the developer into that package...

I didn't get that idea so that was why I replied.
Perhaps Brad would refrase few things to make it clear enough for everyone.

...As system architects we have to be aware that we can no longer rely on our threads having free reign of the system - instead we have to find a way to leverage our solutions into a world of parallelism in order to take full advantage of upcoming hardware.

Sorry pal but that sounds like it was quoted from the computer science book.

[Slaine]

Sorry pal but that sounds like it was quoted from the computer science book.

Perhaps I am in the wrong profession then.

[thegreatone]

Brad, i'd like to add that as a new CPU owner, (Core 2 Duo E6300 [bottom of the range]) it easily outstrips my old CPU (Pentium D 805), the new architecture is simply faster per core than ever before.

For example, in clock speeds we shall say my E6300 @ 1.86Ghz (per core) vs my Pentium D 805 which was heavily overclocked to 3.8Ghz per core, was much much faster, which is true, even if i make windows run the application on one core only, with nothing else on it.
The new architecture means that even the slowest most bottom end CPU in the Core2 range (The E4300 i do believe) can outperform a P4 @ 3.2Ghz. Whereas with the higher chips like the X6800 (And indeed my Core 2 E6300 @ 3.15Ghz heavily overclocked) outstrip the old generation chips so much, they might as well have been running on a TI-83 calculator.

What i'm saying is, clock speed doesn't always mean real-speed. The world record for computing Pi to 1m places was held for a long time by a severaly overclocked P4 @ almost 7.3Ghz. This all changed when the new Core2's came into action, the same speeds can eb attained at a mere 4Ghz with the new architecture.

Basically, a Core2 based processor with it's new design will run an application much faster than a Netburst (think P4, Celeron D) based processor ever could.

However, i must conceed, multi-core working programs should indeed be written, and i severely wish they were. I know for a fact the new Quad cores would be unreal in power with it. As would my Dual-3.2Ghz E6300.

That's pretty much all i have to say. Apart from agreeing with Rhinobulls point about RAM

[brad jones]

Basically, a Core2 based processor with it's new design will run an application much faster than a Netburst (think P4, Celeron D) based processor ever could.

Find a single CPU intensive application and run it on the older and newer processor adnd see what it does. Most applications are more I/O bound with bursts of CPU-intensive processes. As such, the new architecture will better use the CPU and will give better 'performance'. But, if you find an application that is CPU-intensive without being I/O (or otherwise) bound, then I think you will find that the slower speed limit will slow the app.

Brad!

[thegreatone]

Find a single CPU intensive application and run it on the older and newer processor adnd see what it does. Most applications are more I/O bound with bursts of CPU-intensive processes. As such, the new architecture will better use the CPU and will give better 'performance'. But, if you find an application that is CPU-intensive without being I/O (or otherwise) bound, then I think you will find that the slower speed limit will slow the app.

Brad!

Give me an application to test, and i'll prove Core based is much more efficient than Netburst, and as such, it's not all to do with more Cores, it's about the architecture as well, which is why the AMD processors were for a long time considered faster than their equivalents, the Intel P4's, whilst running at a lower clock speed. (i'll have to borrow a friends P4 for this test mind )

The more efficient the pipeline in a CPU the faster, in theory it will execute the code it's meant to, just as you'd suspect. Which is why newer more efficient architectures will have a lower clock speed. Intel know for a fact most of their Core processors will do almost 50% more on average than the speed they sell them at, even the lowest of the low (like my E6300 was). So, i don't think it's about them being less powerful, or producing more heat, my Core2 is as toasty as a P4 was at the same clock speeds, but the performance difference is phenomenal.

I don't know how else to put this Brad, but the newer CPUs on the market are simply more efficient and as such don't have the kind of clock speed you'd expect. Whereas the performance chips they produce do (again think Intel Core 2 Extreme X6800's).

Thanks.

[Shaggy Hiker]

I don't think that in any way contradicts what Brad had said. After all, he was really talking about trends in the future, and some are noticeably true. Where are the 5 and 6 GHz processors? We aren't seeing increases in processor clock speed, instead, we are seeing increases in efficiencies, caches, memory speed, bus speed, etc. The curve, which was with us for the 80s and 90s, is leveling off already. The difference between the 8086, 286,386,486, etc. was always HUGE, and readily obvious to even a casual user. Now, you have to find just the right program even to test the difference between systems. Partly that is because most of the programs we use today run at instantaneous speed (from our perspective, of course), so an increase in speed is meaningless.

Consider that Word will underline, or correct, words as you type them. This means that things are happening between keystrokes, which was pretty much impossible back in the days of the 8086. Word on a faster computer would be about the same as Word on a 1GHz computer, as long as there was sufficient memory to avoid HD access.

This makes it somewhat difficult to compare processors well. I have an application that takes about four days to run on a 2.8GHz Intel D with 4GB of memory. It takes about the same length of time on a 2.8GHZ Pentium IV, and about the same on a 1.8GHZ AMD Athlon. Unfortunately, the test is not a good one, because, while the program is almost exclusively processor dependent with VERY little IO, the program is nonlinnear, and no two runs will ever go at the same speed. Still, since the time to finish is about four days, and I have run it on all three computers, with none of them appearing much faster, I would have to say....very little.


Oh yeah, let me add the point: If we are seeing a leveling off of the speed of each core, then new features will slow an app unless it can be distributed across cores, which was the point of the article. The efficiencies are coming from the support around the processor, and there is still somewhat to be gained there, but for a truly all-in-the-processor process, those gains will top out, and may have already.

[superbovine]

There is always http://en.wikipedia.org/wiki/Quantum_computer

[brad jones]

Oh yeah, let me add the point: If we are seeing a leveling off of the speed of each core, then new features will slow an app unless it can be distributed across cores, which was the point of the article. The efficiencies are coming from the support around the processor, and there is still somewhat to be gained there, but for a truly all-in-the-processor process, those gains will top out, and may have already.

Exactly. Cache and other external efficiencies [processor architecture] will continue to be tweaked for additional speed. I don't dispute nor do I disagree with that fact. According to comments I've been told from Intel people, there is still a little room for cache to be tweaked. The architeture is likely in the same condition. But even with those tweaks, there is still a limit on how fast the cores are going to go once you get there.

[Shaggy Hiker]

If I know Intel, they will tweak as much cache as they can possibly tweak.

[thegreatone]

Exactly. Cache and other external efficiencies [processor architecture] will continue to be tweaked for additional speed. I don't dispute nor do I disagree with that fact. According to comments I've been told from Intel people, there is still a little room for cache to be tweaked. The architeture is likely in the same condition. But even with those tweaks, there is still a limit on how fast the cores are going to go once you get there.

Very true, i see your point now. Mine was based on them not needing to up the speed at all... But eventually if it were to continue they'd have too.

Still, die shrinking will yield a tony bit more speed. As stated.