How fast is fast?

[John E]

Last week I was the (slightly embarrassed) owner of an ageing Abit motherboard with a 1GHz Pentium III and a measly, 128MB of RAM. It wasn't a great system but 1GHz wasn't tooooo bad and I couldn't find any justification to replace it. Until Friday.....

On Friday it became clear that my motherboard had passed away sometime on Thursday night, so Friday was spent ringing every local dealer I could think of to get myself a new board. I'd always had a hankering to try an AMD chip so I'm now the (much prouder) owner of a large telephone bill, an even larger credit card bill and a new Gigabyte motherboard, with onboard LAN, Firewire, Memory Card reader, blah, blah blah - oh, and onboard RAID (which I haven't installed yet - but will). The new board supports a 2.2GHz Athlon XP chip (although I bought the 2.0GHz one 'cos it was cheap) and with the spare cash I managed to buy four times the amount of RAM (512MB) and DDR RAM at that!!

The weekend was spent installing my new purchase, loading up the drivers, tweaking the performance etc. So what's the performance like? Well, frankly, it's difficult to tell the difference between the new board and the old. It's nice to have the new features and maybe RAID will help when I get around to installing it - but I'm real glad the old board went belly-up. If I'd spent all this money in the hope of getting improved performance I'd be sorely disappointed. :(

Does anyone else agree that PC performance seems to have peaked?

[CBasicNet]

What OS are you running?

[John E]

I've got a triple-boot system with Windows XP Home (main OS) with occasional forays into Win98 and NT, on separate partitions. So far, I've only updated the XP partition successfully. The new board's drivers trashed my Win98 partition (I'm still waiting for Gigabyte to reply on that one) and I daren't tamper with WinNT yet, because of what happened to 98....

[Yves M]

Well, I upgraded from a dual Celeron 500 to a dual Althon 2000+. From 192MB of RAM to 1GB, 16GB HD to 240GB in RAID0, from a Geforce2Mx to a Ti4400... I can tell the difference :D

Seriously though, many applications just take a long time to load, so upgrading your harddisk(s) is going to help make it much faster, if they are a bit on the slow side. Also the memory should be a plus too. But it's true that even if it's twice as fast, you wouldn't notice that much.

[John E]

Interesting you should say that Yves,

My previous upgrade (i.e. the one before the one I just did) was from a 486 running, I think, at about 400MHz (or maybe 333MHz) to my 1.0GHz Pentium III - and I also noticed a significant difference. But there's been no noticeable difference since I went to 2.0GHz with DDR RAM....

I need to do some proper tests with compile times maybe. And some of the apps I write use signal processing (sample rate conversion etc) for which I've got some timings from the previous board, so I could easily compare - but from a user perception, there really hasn't been anything where I could honestly say that something has improved. :(

As I said before, I'm just glad that I was forced into this upgrade and didn't buy it voluntarily!!

[Yves M]

Hum, I doubt that your 486 was running at 400 Mhz ;) probably more like 40 ;)

But anyways, the thing is that you don't really notice if it's taking 5 seconds or 10 seconds (best case, if it's twice as fast). The things where a faster CPU really does make a difference is in games, cpu-intensive applications (like running scientific apps, raytracing, photoshop filters...) and that's pretty much it. A fast hard disk on the other hand makes programs load faster and important for developers makes compilation faster (the cpu helps a bit as well). The point of the RAM is that you can keep lots of apps open at the same time and not expect any delays.

Actually the point of having a dual system for me is based on the same rationale. When I'm running a compilation in the background, I can check my mail or even run VB on the other processor without the whole system slowing down. It's also a must for running VMWare ;)

[John E]

Hi Yves,

No, it was definitely either 333Mhz or 400Mhz because I've usually ended up doubling or tripling my processor speed with each upgrade. However, you're right that it might not have been a 486. Now that I think about it, it could have been one of the early pentiums - either a PII or an MMX.

I tried some timings yesterday, comparing the time taken to run a memory intensive process (audio sample rate conversion) against some times that I already had for the previous board. The new board is definitely faster, but not by a big margin.

[Yves M]

Ha there is the fudge factor with AMD and Intel of course. If you bought an Athlon 2000+, it runs internally only at 1666 Mhz. This compares equally to a Pentium 4 2 Ghz. But that's because the P4 is slower on the same clock-speed as an Athlon. This is however not true for the P3. So indeed your new processor is only theoretically 66% faster than the old one. Then take into account whether your software is optimized for the P3 (SSE) or not.

It's a minefield ;)

[Amn]

The topic on whether hardware progress is really making progress ;) is so **** complicated, because so many factors are involved. I ll try to list them here:

1. Average modern processor doesnt utilise at least 50% of its potential on daily basis, since software usually keeps up with it only years after. This involves, 3D Now! , MMX, SSE, SSE-II, instruction-prefetch mechanisms, cache, co processor, DMA support, and multithreading techniques. What I am saying is, that there are millions of transistors inside a PIII, P4 or Athlon processors that just sit there. For now. This means, they CAN do it, but DONT do it, because NOONE TOLD THEM TO DO IT ;)

2. People are slow. We can hardly sense a difference between 100 milliseconds and 200 milliseconds, and that is about a million average instructions for a CPU either injected or dropped out.
This means that CPU design may actually do things up to 10 times faster, but in case we are talking about moving 1 Kb of memory with the say, MOVSD instruction, we wont notice it, until there is like MOVSD going on continuously like database archiving.
Then we begin to notice. So its a trickt thing...these measurements.

3. There are always bottlenecks in computers involved. If you have an Ahtlon Processor which can push data at 1.6Gb second, and a PCI video vard which cant go higher than accept data at 500Mbs/second, you wont notice a difference. The same rule applies to a RAID controller with only one disk or a UDMA controller with the harddisk being physically slow (although UDMA enabled).

All these factors are crucial.

The ideal system is a system that is balanced on all of its buses, so each component is talking to all its neighbours at an average speed of the system (mathematical average). But since computers dont do a thing without CPU running an image in memory, software is needed that is capable of utilising every function there is for its task.

Another thing: Have you guys heard of hardware viruses ? Those that are able to burn up a CPU just by making it run some instructions ? This is because when conveyered, only a fraction of the CPU chips live up to the absolute physical silicon quality that they were designed for. Its not a secret that Intel and AMD filter out chips in hundreds because of malfunction on the conveyer line. But some end up in the middle - not the bestm but will do for average user, as engineers run typical tests for a typical user to ensure the processor is suited for daily desktop operation.
But those viruses push it beyong normal operation, without any hard tricks, just by issuing those parts of the processor that usuallt dont run. Silicon gets hotter, and eventually burns.
These things exist,, oh.. I am sure you heard of them anyway ;)

[galathaea]

To make yavmens! Intel's marketing got the best of them for the P4s, and so they added a bunch of pipeline levels so they could increase the clock speed. Unfortunately, it also decreased their throughput of operations on the same register set, so that an identical clock speed P4 runs slower (actually quite a bit slower...) than the P3 version. This is one of the reasons AMD is able to advertise such a large conversions equivalence for their slower clock speed processors (in addition to their more efficient internal instruction set, etc...). So your upgrade may not be as astounding as you had hoped.

To amn in particular: I've got a couple of those viruses on my computer now. I grabbed them off some hacker sites when I was studying ring0 programming, to try and learn what limits I should set for myself. But most of what I needed to program came no where near the exotic instruction sequences of those little devils. I've never had the guts to run any of them...

Anything I else I had to add has already been skillfully covered by others :).

[John E]

All the above is true and only serves to confirm my suspicion that processor performance has "peaked" and isn't getting (perceivably) any better. That isn't to say that raw processing power has peaked though. As Amn points out, processors are capable of things which the software has yet to catch up with. But in the meantime, AMD and Intel need to carry on selling chips. Therefore, step forward the marketing departments. I think I realised that going from 1GHz to 2GHz mightn't be noticeable - but I did think that if I added DDR RAM into the equation I might notice some kind of improvement....:mad: Amn made an interesting quote....
Originally posted by Amn
2. People are slow. We can hardly sense a difference between 100 milliseconds and 200 millisecondsI know this is off topic - but did you know that in World War II, bomb disposal experts were taught that if there was an unexploded bomb around, the best place to be was right on top of it! That was the only way that armies could get anyone to do the job. The reasoning was that because the human nervous system is comparatively slow, if the bomb blew up, you'd be dead before you could feel anything.... Fortunately, for the military, nobody ever lived to deny it!

[CBasicNet]

Is your old system previously running on WinXP?

WinXP is a resource and CPU hog.

[Amn]

No WinXP is in fact one **** of a good job when it comes to resources. I am usually running Adobe Photoshop, Macromedia Flash, Outlook, Internet Explorer and Visual Studio .NET on it, and unlike Windows 2000, it is remarkably responsive. I can actually hear it pagefile my virtual memory when i switch my tasks ;)

I was never happier than with XP. It has some issues, but generally its a big leap forward. The thing is a marvel of that very "processor utilisation" at least to some measureable extent. Never tried Linux though.

I am running a P3-600 with 256 Mb of RAM and Seagate Barracuda VI 80Gb. So considering those apps, above WinXP lives up to the expectations.

WinXP may suck though at "guessing" which it has to run. I tweaked mine to be running 16 processes at the startup, because I disabled many background services i didnt need. Also i removed the Active Desktop. I am running themes though. ITs all about understanding how the OS works, because good as it is, WinXP is dumb, and is distributed for an average user, thats why it usually installs with all kinds of unneeded tasks running, like the CTFMON.EXE which is text recognition service for FUTURE Office products. I am running it as well though - Who knows, maybe i ll buy one of those "Shutdown boy!" devices in near future ;) I think the CTFMON is used with that Microsoft thingie, which looks like a headset with mic, that enables you to talk to ppl when playing games - forgot the name.

[John E]

The answer to CBasicNet is yes, I had been using XP with the old board. In fact, nothing else changed. I simply installed the new board with its new memory, processor and all its new drivers etc.

Interesting point about XP though. Up until recently, I very much preferred Windows 98 to any other Microsoft OS. In my view, NT (on the workstation) is sheer crap and 2000, although better, isn't much better. I know that 98 gets plenty of stick - and probably justified - but it always worked a treat with my old board. Now the situation is reversed. With my new board, XP is very stable, whereas I haven't really got 98 to work well at all.

I think Amn's earlier points are the real explanation here. Manufacturers are ramping up clock speeds to please the marketing people - but the chips are being put through all kinds of "fancy dances" in order to deliver the increased clock rate. This reduces the efficiency of the chip and therefore, any speed increases are being largely wasted. Hence, the perceived reduction in expected performance. The manufacturers do put impressive new features into each new chip but software isn't keeping up and/or taking advantage of the new features at the same pace.

[Yves M]

I don't really agree with your conclusion. The thing is that your processor is labelled as 2000+, so you expect it to be 1000Mhz faster than the old one. But it's running at 1666Mhz internally, so it's "only" 666 Mhz faster. And 666 Mhz is only 66% faster. That is really not that much. It's not a matter of the new chip not being faster, but your expectations are too high.

Personally, I tend to only upgrade my system if I can get a 4x faster one. Then I always notice the speed improvement :p So I went from 200Mhz to 2x500 and then to 2x2000+ (ok admittedly 2x1666Mhz ;))

And hum, I think one of the biggest and most overlooked performance improvement in the PC sector in the last few years has been IDE harddrives. When you get a good drive today, it transfers data at about 35-40MB/s. That is huge compared to the 7MB/s one of my old HDs has (I'm still using it ;)).

As for the OS, I prefer 2000 over XP. I'm having a bad experience with XP at work when it randomly locks up the computer from time to time when the system is under high load. It stays choppy for a bit and then goes back to normal. I've not had this problem with 2K. But apart from that the differences between XP and 2K are not that big.

[John E]

Originally posted by Yves M
The thing is that your processor is ... running at 1666Mhz internally, so it's "only" 666 Mhz faster. And 666 Mhz is only 66% faster.I take your point Yves, but don't forget that I also upgraded to DDR RAM which, apart from being able to deliver data to the chip on both clock edges, is also - if I understand this correctly - running at a higher front-side bus rate of 266MHz. This compares with 133MHz for my old, Pentium III. Surely this points to bottlenecks in the new chip that weren't present in the old one?Personally, I tend to only upgrade my system if I can get a 4x faster one.Well, that's a fair point. I've tended to go in 'threes' instead of 'fours'. My first PC ran at 33MHz, the next was 100MHz, then 333MHz, then 1GHz. I accept your point about the hard drive though - although I do have a fairly nippy Seagate Barracuda (40GB) - it's about a year old.As for the OS, I prefer 2000 over XP. I'm having a bad experience with XP at work when it randomly locks up the computer from time to time when the system is under high load. It stays choppy for a bit and then goes back to normal. I've not had this problem with 2K. Tell me something, does this happen when you're debugging? I get a weird problem with XP where my PC slows to a crawl occasionally when I'm trying to debug in VC++. It happens especially after pressing 'F5'. It doesn't lock the PC up but I usually have to exit from VC++ and restart it. That's the only oddity I have with XP though and it never happens with 98 or NT.

[Amn]

Originally posted by Yves M
And hum, I think one of the biggest and most overlooked performance improvement in the PC sector in the last few years has been IDE harddrives. When you get a good drive today, it transfers data at about 35-40MB/s. That is huge compared to the 7MB/s one of my old HDs has (I'm still using it ;)).


Yeah, and they are also the slowest part of the system as a whole. Nothing is as slow as a harddrive. Ever seen all the virtual memory being blitted on the screen when Windows redraws the backgrounds and repaints windows after some app hogged on resources ? Or how the harddisk itches and scratches as it searches for files. Hard disks are mechanics still, not electronics...
And 30-40 Mbs is well....1/50th of the nominal system data transfer speed ;) Not to mention access time of the harddisks is 10ms ;) Thats like what, one million times slower than the average access time for DDR 2100 RAM ;) AAAAAaaaaa!!

I am just *****ing Yves, dont get me serious ;) ...I had a WD harddisk, that was louder than the civilian warning alarm in Oslo, now i replaced it with the quiet Seagate disk and this one starts to spin off louder and louder too... I dont know whats with them all .... :rolleyes:

I WANNA MY FLASH EPROM HARDDISK :D

[Amn]

Sorry to add an extra post, but for the sake of clarity.

Originally posted by John E
Tell me something, does this happen when you're debugging? I get a weird problem with XP where my PC slows to a crawl occasionally when I'm trying to debug in VC++. It happens especially after pressing 'F5'. It doesn't lock the PC up but I usually have to exit from VC++ and restart it. That's the only oddity I have with XP though and it never happens with 98 or NT.

It happens to me occasionally. I am on XP Pro. And i have VC .NET which indicates the problem might be in XP. VC just stops responding completely, the system doesnt respond very well either, windows lock up, without getting or handling repaint messages, start menu is dead etc. The only solution is to patiently wait each time you do something, as the system tends to wake itself up only to lie dead again for another 10 seconds. ANd this continues until i abnormally terminate VC. This problem is even more frequent when i hit F5 to debug a secondary thread of a process. Also when i disassemble (Show Disassembly instead of Show Source) my code, and hit Shift-F5 (end debug) it takes about 5 seconds to kill debug mode it seems.

[John E]

That's exactly the problem!! I thought I was the only person who'd encountered this...

[giladasaf]

I have played quiet a bit with hardware combinations.
So call it more practical expirience rather then knowledge

but if I would have to advice a friend how to buy a cumputer I'd say


ram ram ram and some more ram start with 512 and the sky is not the limit!!!

I dont know what microsoft are putting there but windows are to memory as women are to jewelry :p

[Yves M]

Yes, Amn and John this is pretty much the sme problem I have with XP at work. It's made worse by the fact that I debug my applications with VB (so hitting F5 opens up VB ;)). I'm not sure whether it's the fact that I've got a dual at home, but this never happens to me with 2K.

I take your point Yves, but don't forget that I also upgraded to DDR RAM which, apart from being able to deliver data to the chip on both clock edges, is also - if I understand this correctly - running at a higher front-side bus rate of 266MHz. This compares with 133MHz for my old, Pentium III. Surely this points to bottlenecks in the new chip that weren't present in the old one?
I don't know what it in your case, but it would suprise me if it were the chip. Or maybe your timings in the BIOS are not set optimally.

accept your point about the hard drive though - although I do have a fairly nippy Seagate Barracuda (40GB) - it's about a year old
Well that's just an idea ;) A lot of the percieved responsioveness comes from the Harddisk's speed, so well...

And 30-40 Mbs is well...
Add two fast harddisks with a large cache and put them into RAID0 and you'll hit 100MB/s ;)

[John E]

Originally posted by Yves M
I'm not sure whether it's the fact that I've got a dual at home, but this never happens to me with 2K.Interestingly, it was a dual system that I was using when I had this problem (presumably, you mean a dual processor?) - although it only had one socket populated. The problem hasn't occurred yet with my new (single processor) board, but I've only had that one for a week.Add two fast harddisks with a large cache and put them into RAID0 and you'll hit 100MB/s ;) Well, in fact, I've already bought the drives but I won't be putting them in until I get the rest of the system stable....