PDA

View Full Version : Any idea how much faster a Power4 or 970 would be(%-wise) in comparison w/ a G4 chip?


GeneR
Jan 21, 2003, 10:59 PM
I'm thinking: "Yeah, baby, I really want a faster chip."
I was at the Apple Store today and I liked running FCP3 on the 1GHz 15" TiBook, BUT of course, I wish it was much, much faster (I'm a pretty impatient fellow when it came to rendering transitions and the footage selected in FCP.)

Of course, (in an ideal world: Apple-centric, no such thing as M$ j/k) to be able to have the rendering done lickity-split is the end goal, I'm just trying to figure out what to expect for 2003. If anyone here has any info about this, I sure would appreciate it. Thanks a bunch!

:D

Chaszmyr
Jan 21, 2003, 11:04 PM
For normal consumer use, a 970 will be mhz per mhz about the same speed as a current G4, most likely. The RAM access abilities and 900mhz bus of this 64 bit chip give it a lot of potential, and make it for high-level processing. Also, this chip would be absolutely killer if Apple took it and put it into a dual-processor/dual-system bus computer.

Chaszmyr
Jan 21, 2003, 11:05 PM
(Also, if you arent aware, the 970 is supposed to launch at 1.8ghz)

janey
Jan 21, 2003, 11:15 PM
970 rules.

Foucault
Jan 21, 2003, 11:19 PM
speed is everything in a computer. this new IBM chip could propel Apple to the next generation of power computing... I just hope it won't be as slow in progressing through its speed as the G4 chip has been...

janey
Jan 21, 2003, 11:25 PM
the powerpc 970 rules. apple should have like a quad processor machine when it comes out.

arn
Jan 21, 2003, 11:28 PM
Originally posted by Chaszmyr
For normal consumer use, a 970 will be mhz per mhz about the same speed as a current G4, most likely.

I think you are mistaken.

If you go by SPECint numbers... and I know you can't necessarily correlate this with real world performance... DaveGee posted these numbers (http://forums.macrumors.com/showthread.php?threadid=12960#post171140) in October

SPECint_base2000
=================
242 G4 800Mhz
259 G4 867Mhz
306 G4 1000Mhz
937 970 1800Mhz
=================

SPECfp_base
=================
147 G4 800Mhz
153 G4 867Mhz
187 G4 1000Mhz
1051 970 1800Mhz
=================

Now don't go crazy over it... as they probably don't correlate to real-world performance... but I do think the 970 will have a higher perforance per Mhz than the current G4s.

arn

Catfish_Man
Jan 22, 2003, 12:01 AM
OK, the 970 can dispatch 5 instructions (1 of which has to be a branch) each cycle. It can load 8 (the extra three are, afaict, in case one of the others can't be run immediately. The G4 doesn't do it because it can't rearrange instructions). The G4+ can load and dispatch 4 (1 branch), iirc. The 970 also has 2 floating point units compared to the G4s one. Also, the 970 has twice as much L2 cache and L1 data cache as the G4, but no L3 cache. So:

On purely memory bandwidth limited code, the 970 can go about 4.8 times as fast as a G4.

On purely floating point code, the 970 can theoretically go twice as fast per clock as a G4

On purely integer code, the 970 can theoretically go the same speed per clock as a G4
-----------------------------------------

On Altivec code, the G4+ can go slightly faster than the 970 (the 970's Altivec unit is like a G4s, rather than a G4+s).

On normal, poorly optimized code, the 970 can rearrange instructions and use its extra floating point power and cache to go significantly faster.

Overall, I would guess the 970 will average about 3 instructions per cycle (+/- .5 by my guess), while the G4+ averages 2.31. That would put a 1.8GHz 970 at somewhere between 1.56 and 2.19 times as fast as a 1.25GHz G4.

<disclaimer> everything past the dotted line is a guess, everything above it isn't, but may or may not be true. </disclaimer>

ddtlm
Jan 22, 2003, 12:24 AM
Catfish_Man:

The G4 doesn't do it because it can't rearrange instructions.
Actually the G4 does have very limited out of order execution abilities. I believe it can only look at 12 to 16 (I forget) at once when reordering them.

On Altivec code, the G4+ can go slightly faster than the 970 (the 970's Altivec unit is like a G4s, rather than a G4+s).
As far as I know, the G4+ really can't issue instructions fast enough to fill all 4 AltiVec functional units anyway. This in addition to whatever memory starvation problems it has. I wouldn't worry much about the reduced number of AtliVec functional units.

On normal, poorly optimized code, the 970 can rearrange instructions and use its extra floating point power and cache to go significantly faster.
This is where it's at. The PPC-970 can go faster per clock in ideal cases, but it really pulls away in less-than-ideal cases.

My main concern about the PPC-970 is the poor cache setup. The L2 is OK right now, but this fall 512k is going to be behind the times.

will
Jan 22, 2003, 07:40 AM
I would guess IBM will offer a range of cache sizes in the long term. Remember it isn't just Apple who will be using the 970, IBM will be using it in a range of workstations and servers.

What I want to know is how quickly IBM will be able to ramp up the clock speed of the 970. As the 970 performance per MHz is roughly 70% higher than the P4, if IBM can ramp up the speed the 970 could be significantly faster.

Also it should be noted the 970 is designed for multiprocessor systems, so I hope to see 4x970 workstations from Apple and IBM (imagine rendering on that).

Then there's the 64 bit issue. How quickly can Apple get 64 bit support into OS X, hopefully they're working on it now. That would allow Apple to deliver servers with more than 4GB of RAM and support large applications.

If Apple don't get a modern fast CPU into their machines within a year I think they'll be in serious trouble.

wormy
Jan 22, 2003, 08:00 AM
can someone explain the G4+ to me? is that the next generation of G4 chips before the 970 comes out? and will they be in the soon-to-be-released power macs?

will
Jan 22, 2003, 08:07 AM
The G4+ is the PowerPC 7457 see http://www.macobserver.com/article/2002/10/22.4.shtml

Telomar
Jan 22, 2003, 08:27 AM
Originally posted by will
The G4+ is the PowerPC 7457 see http://www.macobserver.com/article/2002/10/22.4.shtml Actually it's any G4 since they redesigned it (ie. 7450, 7455 and both the cache free counterparts).

will
Jan 22, 2003, 08:48 AM
I think it has been used in a variety of confusing ways. Perhaps we should stick to the full Power PC designations?