The "SPEED" of progress

[Harrison]

PC Pro this month have dedicated much of the issue to Retro computing. They are looking back at the past of computing and have covered many aspects of the past.

Some of the interesting articles cover how hardware has advanced over the years, and they even cover retro gaming and show how to get emulators such as MAME and a SNES emulator working.

A couple of the most interesting articles relating directly to the PC are comparing old issues of PC Pro with current benchmarking standards they use.

In one of the articles the writer decided to swap his current computers for an early Windows 95 PC running the first release of the OS. His point was to show that much had changed over the years since its original release to try and quash the people who always comment that not much was really changed with Windows since the release of 95 other than its visual look. This really does show how so much really has changed since then.

The other interesting article is comparing benchmarks from the start to the present.

Sadly the first PCs they ever reviewed couldn't run Windows XP so they instead moved forward 50 issues and used a Pentium 2 333MHz PC for the benchmark, against their current benchmarking standard they use, a Pentium 4 3GHz PC which is itself getting quite out of date. Even by this standard current systems they are benchmarking are often scoring over 2 which means they are twice as fast as the P4 3GHz benchmark system.

And the results? Well on average the old P2 333 was over "50 times" slower than the P4 3GHz benchmark!!! When they did manage to get their benchmark tests to run. The problem was that the benchmarking suite often had trouble trying to record a benchmark result of less than 0.1 compared to the P4 standard test results. :lol:

To put this into real world examples, on the P2 it took over 4 hours to complete the Photoshop benchmarking tests. This compared to five and a half minutes for the P4! And for a 3DSMax test the P2 took 28 minutes to render a scene, compared to just 50 seconds on the P4. Quite a large variation you have to agree and such real world speed comparisons would mean a lot if you were trying to do anything that was time critical. It would also mean the difference between wasting days of year life instead of hours on the same tasks!

So Steve, does that "vintage" P3 look so modern now?

[Submeg]

:blink: that is quite a big difference

[Demon Cleaner]

Well on average the old P2 333 was over "50 times" slower than the P4 3GHz benchmark!!!I'd never imagined that, but that is also because software nowadays is much hungrier than in the past. You can only see it running new software on older PCs, and not the other way around. Office f.ex. does not start 50 times faster now.

[Stephen Coates]

Harrison, I'm sure my Pentium III will be much faster than a 333MHz Pentium II.

[Harrison]

Harrison, I'm sure my Pentium III will be much faster than a 333MHz Pentium II.

Your PIII was 500MHz wasn't it? Therefore you are correct Steve, your P3 is a whole 167MHz faster than that P2 333MHz machine! ;)

In reality, yes the P3 chip was slightly more advanced than the P2, but is based on the same basic design but with better memory management. So the speed increase of a P3 500 compared to a P2 333 would probably be about twice the speed in real world tests.

But that would not half the benchmark times of the tests I mentioned to make it 25 times slower than the P4 due to the bigger differences in other factors such as complete architecture changes between early P3 500MHz machines and even the later P3's, and a complete redesign for the P4 and beyond. I would expect in a similar test the P3 500 would around 45 times slower than the P4, or there about.

And for the 3DSMax and Photoshop tests I expect the resulting times would be pretty close to the P2 tests unless a lot more ram was given to the P3 during its testing.

[Harrison]

Well on average the old P2 333 was over "50 times" slower than the P4 3GHz benchmark!!!I'd never imagined that, but that is also because software nowadays is much hungrier than in the past. You can only see it running new software on older PCs, and not the other way around. Office f.ex. does not start 50 times faster now.

That is true to a certain extent, although Office has never really taken that long to load on any systems I can remember, even P2 and even Pentium 1 systems.

And yes, they are benchmarking newer more advanced software to compare, but you do need to run identical tests on both systems to get the realistic results. And if you think about it, if they ran less resource hungry tests on the older system, those same tests on the newer system would be even faster and may even show the older system as being even more than 50 times slower!

[Submeg]

But really, you think about it, back in the day, you KNEW it was going to take a while to load and so you went and did something else and came back...Nowdays everyone wants everything now, so people become less patient when it comes to loading times....maybe some people should be forced back to using 286's...

[Harrison]

That was indeed very true that when doing certain tasks such as processing a batch of images, compiling a script, rendering a video process or rendering a 3D scene we knew it would take some time, and often the program would show an estimate of time left to finish the process, so you would then go away and do other things whilst you waited.

But the thing is that we always wished all of these processes were quicker so we could reach the end result and then get on with the next part. Needed to wait for any of these tasks to complete was never something we were really all that happy to wait for. These days, being able to render a relatively complex 3D scene in under a minute, compared to over half an hour (and sometimes much longer) in the past, makes a huge difference to the possibilities for experimentation within the work being done these days.

Going back to Steve's comment for a second. The Pentium 3 was slightly ahead of the Pentium 2 in terms of architecture, but it wasn't much faster overall due to many other factors. His Pentium 3 shares exactly the same speed of north bridge, ram, IDE interface and all other elements of the PC design at the time, therefore although the CPU speed is faster, the system cannot move the processed data around the system any faster than the Pentium 2 in the test could and therefore the test results using the Pentium 3 would not be all that much faster than the Pentium 2 used.

In contrast, systems since then have moved on a lot in terms of all areas of PC architecture, not just the CPU architecture and speed. Lets compare Steve's Pentium 3 architecture with current architecture to show this fact.

The P3's 32bit PCI bus runs at 33MHz and can push a maximum of about 125MB/s to say a PCI graphics card. In contrast a current PCI-E interface of today can push a maximum of 8GB/s. A huge increase you have to agree. Next the IDE interface. The old PATA interface the P3 is using would probably be the original ATA33 version with a maximum parallel bandwidth of 33MB/s, equally around 10MB/s actual data transfer. In contrast today an SATA interface has a maximum serial bandwidth of 300MB/s, with the current SATA2 drives easily managing 100MB/s sustained transfer.

I could go on to include the huge bandwidth and speed increases in Northbridge, motherboard data transfer, ram bandwidth, ram timings, and the huge increase in ram speed compared to current systems, but I hope you get the idea from the examples given above.

One final thing to cement the speed fact and actually an easy way to show exactly how fast a Pentium 3 500MHz CPU is compared to a Pentium 2 333MHz CPU and the benchmark Pentium 4 3GHz CPU is the FLOPS (FLoating-point Operations Per Second) speed rating.

The Pentium 2 333MHz CPU has a rating of 333MFLOPS, which is actually how earlier CPUs got their speed rating because the speed a CPU was shown as was the actual number of theoretical FLOPS it could manage. Equally the Pentium 3 range also processed data in a very similar way, so the 500MHz CPU has rating of 500MFLOPS, so is theoretically 167MFLOPS faster than the Pentium 2 333 CPU.

Now moving on to the Pentium 3 3GHz PC things are not quite as simple. More modern CPUs no longer just process one floating point operation per clock cycle. The Pentium 4 for example processes 4 per clock cycle, so the theoretical FLOPS rating for the Pentium 4 3GHz CPU is 12,000MFLOPS or 12GFLOPS, which if directly compared with the P2 and P3 CPUs would give you the P4 as being 36 times faster than the P2 and 24 times faster than the P3. But as I already mentioned above, the final real world tests also factor in much more than just raw CPU theoretical processing limits. The rest of the PC's architecture is a huge factor that goes towards the complete processing speed of the whole system. It is not good having a CPU that could process X amount of data per second if the rest of the system such as the HD, ram and motherboard datapaths cannot shift it at a decent speed.

Also even that rough estimate of the P4 3GHz CPU being 12GFLOPS isn't really accurate if you factor in other elements such as Hyperthreading and a lot more cache memory as all these things increase the true FLOPS processing speed a lot more that 12GFLOPS.

So in reality the old P2 and P3 CPU's are very very slow compared to current systems.

Also remember that in current tests PCPro are doing on new systems they are often scoring 2 or more against the Pentium 4 3GHz reference system. This means they are twice as fast in real world tests and therefore this makes the old Pentium 2 333MHz PC 100 times slower than a current Athlon 64 X2 or Intel Core 2 Duo system.

Would anyone seriously want to wait 100 times longer if they had the choice?

That would be like someone travelling at 70 MPH and another going at 0.7 MPH. The faster would take just under 1 hour 26 minutes to travel 100 miles, whereas the slower would take just under 6 days. I know which I would prefer!

[Harrison]

Something else quite interesting I just discovered. An AMD Athlon 500MHz CPU actually had a theoretical peak of 1GFLOP as it's instruction set allowed two operations per cycle, making it twice as fast as a Pentium 3 500MHz CPU! I bet you didn't know that at the time!

[AlexJ]

I always knew there was a reason I bought AMD instead of Intel! Still did anything actually take advantage of it, or was the speed difference noticable in tests?

[Harrison]

Yes, it was noticeable as this was a feature built into the main instruction set of the Athlon range so was taken advantage of during all CPU operations. Obviously all the other system architecture did effect the overall speed, so the fact that both Pentium 3's and Athlon's were using slow 100MHz and 133MHz DIMMS at the time wouldn't have helped.

But interestingly before the Athlon, AMD CPUs were actually slower than 1 floating point operation per clock cycle, compared to Intel which were always 1 measures as 1 per cycle, so before the Athlon range the AMD CPU's were slower than Intel CPU's on a clock for clock basis.

The introduction of the first Athlon with double the operations per clock cycle was the turning point for AMD and the point when AMD CPU's started to gain popularity. My first AMD CPU was a 400MHz K3 CPU in a laptop, so that would actually of had a MFLOP rating under 400MFLOPS, whereas my next was an Athlon 900MHz PC I had at work which would of had 1800MFLOPS. :)

[AlexJ]

The introduction of the first Athlon with double the operations per clock cycle was the turning point for AMD and the point when AMD CPU's started to gain popularity. My first AMD CPU was a 400MHz K3 CPU in a laptop, so that would actually of had a MFLOP rating under 400MFLOPS, whereas my next was an Athlon 900MHz PC I had at work which would of had 1800MFLOPS. :)

My first AMD chip was a K6-166MHz which had 197MFLOPS. :huh:

[Harrison]

Are you sure that is MFLOPS and not MIPS?

MIPS = Millions of instructions per second.

Those are different and are derived from benchmark results and cannot be compared between CPU families.

[Stephen Coates]

The P3's 32bit PCI bus runs at 33MHz and can push a maximum of about 125MB/s to say a PCI graphics card. In contrast a current PCI-E interface of today can push a maximum of 8GB/s. A huge increase you have to agree. Next the IDE interface. The old PATA interface the P3 is using would probably be the original ATA33 version with a maximum parallel bandwidth of 33MB/s, equally around 10MB/s actual data transfer. In contrast today an SATA interface has a maximum serial bandwidth of 300MB/s, with the current SATA2 drives easily managing 100MB/s sustained transfer.

And my AGP graphics card...?

[Harrison]

Your AGP graphics card slot speed will depend on the speed of the actual AGP slot in your system and the supporting AGP speed of the graphics card installed in it.

All AGP slots run on a 66MHz system bus which is twice the speed of the PCI slots. The original AGP 1x slot specification allowed a maximum data rate of 266MB/s, and as the AGP slot developed, faster 2x, then 4x and finally 8x AGP slots appeared. But the graphics card needed to still support the maximum speed of the slot, or it would clock down to the speed of the card, so a 4x AGP card in a 8x slot would still run at 4x speed.

All AGP slots still ran on a 66MHz bus speed, but for the faster AGP slot types the channel is double pumped for 2x, quad pumped for 4x and strobed eight times per clock for 8x AGP speeds, meaning that in theory the actual theoretical speed is actually 2x, 4x or 8x faster than the original 66MHz bus speed. This makes the 8x AGP speed a maximum of 2GB/s, which is roughly 4 times slower than PCI-E. Although also take into account that most motherboards, their north bridges, and the system ram at the time could not shift data this fast until quite late in the AGP slot's life.

For the age of your P3 I would guess it is using an AGP 2x slot and the graphics card is also a 2x speed specification. This means your graphics card is running on a double pumped 66MHz bus, effectively making it 133MHz, with a maximum datarate of 533MB/s.

Comparing this obsolete AGP 2x slot with the current PCI-E graphics slot, your P3's AGP slot is roughly 16 times slower at data transfer compared to the current standard.

BTW, did you know that early 1x and 2x AGP graphics cards are not compatible with later 4x and 8x slots? Putting such early cards in the later slots could completely fry the motherboards beyond repair due to a different core voltage being used in the two different specs.