[selowitch]

It seems as though some earlier PowerBooks had 1MB of L3 cache, whereas the newer ones have only 256k of L2. What's better? Are the newer ones actually not as fast, at least in terms of the cache only?

[belldandy]

It is very hard to describe the performance trade off based on cache. I can tell you this is entirely upto the progams you used.

L2 cache are on-die, which means the interconnect (think of it as a private channel) between the arithmatic unit and the L2 cache are very very fast. However, L3 cache are located off the die. That means the arithmatic unit needs to go off to another bus (think of it as a channel) to talk to this L3 Cache. This bus is much much slower than that of L2 Cache. The penality is about 50x slower.

The new Al-Book series has 512Kb of L2 cache vs. the older series TiBook/17" 1Ghz only have 256Kb of L2 cache. You can expect the often used instruction and data will be able to fit more into the 512Kb of L2 Cache.

What about the L3 cache?
It is true that the newer Al-Book series does not have L3 cache; But since it already has the higher capacity L2 cache, the probability of a L2 hit is going to be higher than a L3 hit.

Ok... So... is it faster or not?
The new al-book is faster because:
1) L2 cache is always on, L3 cache is turned off when the powerbook is using off the battery. Since newer al-book only has L2 cache, don't have to worry about being taken away precious low latency memory access when u are are on battery.
2) @ 1.25Ghz or 1.33Ghz, your G4 already runs faster.
3) @ PC2700 >>>>>>>> PC133

Disclaimer: But it really depends on how smart the programmers are. If they don't take advantage of spatial temporal locality (meaning optimizing their codes to use the cache effectively), doesn't matter if i give u 1Mb of L2 cache as the program will just fill it uselessly anyway.

[anaphora68]

From what I have heard L2 is much faster than L3, so you go for a higher l2 count. There is good discussion on this over at tomshardware.com.

[power142]

Originally posted by selowitch:
It seems as though some earlier PowerBooks had 1MB of L3 cache, whereas the newer ones have only 256k of L2. What's better? Are the newer ones actually not as fast, at least in terms of the cache only?

I have a particular program (pymol) that I run a lot that I've noticed a hit from going from Rev A 17-inch to 15-inch Albook (1.25GHz) - the 1GHz 17-inch renders almost twice as fast as the "faster" 15-inch. This is a specialized program, however, and I suspect is just one of those rare cases where the dataset likes to sit in processor cache (regardless of the L3 being much slower than the L2) rather than main memory.

In general usage, the 15-inch is noticeably quicker to respond and do everything else than the 17-inch (1GHz).

[power142]

Just thought I'd mention that the same program totally smokes anything else I've seen on a Power Mac G4 with dual 1.42GHz chips - 2MB L3 cache per chip. Unfortunately, I haven't had the chance to try a G5 yet.....

If I were you, I'd go for the new Powerbooks and be happy

[Riemann Zeta]

PC2700 >>>>>>>> PC133

In the PC world (and for G5 computers) this is true. However, the G4 does not support a DDR bus. Hence, Apple's implementation of DDR is best described as "pseudo-DDR," in that the RAM talks to the chipset at 333 MHz, but the CPU is still stuck at 167 MHz. Thus, Apple's DDR on G4 machines is epiphenomal--it's strictly for buzzword compliance and does little (if anything) to improve real-world performance.

[Davidarm]

Virtual PC, at least in the Connectix incarnations, was optimized for 512k cache, so any machine with only 256k, whether L2 or L3, is noticeably slower running VPC than a machine with 512k or 1M. Microsoft could conceivably change that equation in VPC 7...

[ae86_16v]

Originally posted by anaphora68:
From what I have heard L2 is much faster than L3, so you go for a higher l2 count. There is good discussion on this over at tomshardware.com.

BellDandy---> Hitted it right on the head.

But I thought speeds were more like this. . . L2 on die runs at full speed of the Processor (i.e. 1.25GHz or 1.33GHz) whereas L3 runs at 1/4 the speed of the processor.

So in the Titaniums 1GHz model they were running at 250MHz.

[vancenase]

Originally posted by power142:
I have a particular program (pymol) that I run a lot that I've noticed a hit from going from Rev A 17-inch to 15-inch Albook (1.25GHz) - the 1GHz 17-inch renders almost twice as fast as the "faster" 15-inch. This is a specialized program, however, and I suspect is just one of those rare cases where the dataset likes to sit in processor cache (regardless of the L3 being much slower than the L2) rather than main memory.

In general usage, the 15-inch is noticeably quicker to respond and do everything else than the 17-inch (1GHz).

power142 -- doyou use the fink version or 'native' version of pymol? i'll have to do a (time) test of a revB 17" and a 15 Ti-book ...

[power142]

Originally posted by vancenase:
power142 -- doyou use the fink version or 'native' version of pymol? i'll have to do a (time) test of a revB 17" and a 15 Ti-book ...

vancenase - I use the native version (since most things I do use the typed commands or scripts anyhow), but I also have the fink version installed. This isn't just for reasons of speed, but features too... the fink version is stuck at 0.86 (for Panther) whereas the native version is 0.93 I think. There is a big speed difference between 0.88 and 0.90 not just on the Mac but on all machines I've tested (Mac OSX, PC-Linux, 2 and 4 processor SGI). I'd be very interested in your timings... if you let me know what you are timing, I'll try to run it on a couple of other machines I have access to for comparison (but as I said before, my Mac has smoked everything else!)

ae86_16v - I'm not 100% certain what the L3 speed is, but even if it's only 250MHz, it's still faster than the main memory speed of 133 and likely has much lower latencies compared to main memory.

[ae86_16v]

Originally posted by power142:


ae86_16v - I'm not 100% certain what the L3 speed is, but even if it's only 250MHz, it's still faster than the main memory speed of 133 and likely has much lower latencies compared to main memory.

Yeah, if I remember correctly from my CS class it is 1/4 of the processor speed. So 1.25GHz would be 312.5MHz; 1.33GHz would be 332.5MHz; and out course 1GHz would be 250MHz.

Yeah, it definitely will be faster than the normal RAM.

[urrl78]

The 17" version A had L3 while the 17" Version B has L2.
Here is what Barefeats has to say:

http://www.barefeats.com/al15.html

Notice also that the Ghz Ti surpasses the Ver A 17" in certain tests.
Also the 15" Aluminum really excells over the 15" Ti in certain tests.