Because it wasn't even released till early 2006 many months after Apple had made the last PowerPC systems and already had Intel systems on the market. They couldn't have used a chip that didn't exist yet.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Best g4 CPU upgrade?
- Thread starter AtmChm
- Start date
- Sort by reaction score
Because it wasn't even released till early 2006 many months after Apple had made the last PowerPC systems and already had Intel systems on the market. They couldn't have used a chip that didn't exist yet.
A Sonnet 1,8GHz single, 7457 (also never shipped by Apple) came to my knowledge recently (on Ebay). I thought, that Sonnet had only made a 1,8GHz 7447A. So I googled a bit. Not adding anything to the discussion so far, I would like to add a link, though. Maybe someone has interest for it.
http://beta.ivc.no/wiki/index.php/PowerMac_G4_Upgrades
It lists a comparison of the 7447 vs 7455 and says the slower BUS G4s profit from the L3 of the 7455, but it states, that it is running hotter, though. I guess a 7457 would behave the same against a 7448(?)
If one looks at the benchmark, the difference (7447 vs 7455) is actually minimal...
A complete(?) list of previous available CPU Upgrades can be found here http://lowendmac.com/ppc/g4upgrades.html
Here is a company that sells G4/G3 Upgrades, they say one has to contact them for prices, but they do not offer an email address or phonenumber http://www.mac4u.at/30.htm
http://beta.ivc.no/wiki/index.php/PowerMac_G4_Upgrades
It lists a comparison of the 7447 vs 7455 and says the slower BUS G4s profit from the L3 of the 7455, but it states, that it is running hotter, though. I guess a 7457 would behave the same against a 7448(?)
If one looks at the benchmark, the difference (7447 vs 7455) is actually minimal...
A complete(?) list of previous available CPU Upgrades can be found here http://lowendmac.com/ppc/g4upgrades.html
Here is a company that sells G4/G3 Upgrades, they say one has to contact them for prices, but they do not offer an email address or phonenumber
http://www.mac4u.at/30.htm
Given what I saw from Barefeats' tests on DP upgrades, no.
The memory and bus bandwidth are just too anemic to deliver what is required of two 1.6 GHz CPUs.
Things are already totally saturated on my Beige G3 w/1 GHz G4 upgrade... (66 MHz bus); and that's a SINGLE CPU. You're basically trying to do this with a 1.6 GHz CPU. :/
I was actually a little surprised at how little performanc improvements were shown.... you can check out Barefeats archives if you're interested.
This above quote is loaded with diluted truths and even complete untruths. Any G4 tower with AGP like the OP has will have at least 4x at the low end and up to 7-8x the memory throughput in a QS. Even the Sawtooth and GE models with just a 100MHz bus have 3x the memory bandwidth of a B&W G3 because the memory controller is far superior. Do I really need to tell you that your computer is slower than a B&W?
Sorry but comparing a beige G3 to any AGP G4 tower is delusional.
A dual WILL be faster.
*Rolls eyes*
And as usual, in your haste to naysay everything I've just said, you get most of it wrong.
Not even close. Not even possible, actually. I have about 300 MB/s memory bandwidth on the Beige according to XBench and other sources (I get a bit over 200 MB/s in GeekBench), and 1.1 GB/s for copies in XB, as that's largely cached.
Given that a 100 MHz bus is capable of a peak theoretical 800 MBps, and in the real world is closer to 600 MB/s sustained, that's 2-3x faster, tops; 3-4x faster for a 133 MHz DA.
But we're talking a bout TWO CPUs, not one, and they're running at 1.6 GHz.. so the bandwidth is effectively cut in half, and that's before considering a 60% higher clockspeed.
As you can see for yourself here, the improvements were definitely not worth the extra money
http://www.barefeats.com/g4up2.html
Yes, the MaxBus was superior to the MPC106/107 (Grackle) found in the Beige and B&W. But as I've already pointed out above, it's not as great a leap as Apple made it out to be in their advertisements at the time, and obviously you need twice as much of it for twice as many CPUs.
Well given that with any G4 CPU upgrade of 700 MHz or higher (as is in mine) it defaults the bus to 66 MHz.. yeah, I guess you'd need to explain how it's faster.
Not really, in this situation. Also please note that the jump from 166 MHz to 1 GHz bus speeds didn't seem to help the G5 much in most situations. Although granted, the G5's memory controller had a horrific latency problem...
Yeah, but only by small amounts that would hardly be noticed, and not worth the extra money.
Again... same link. http://www.barefeats.com/g4up2.html
If you actually look at those benchmarks you posted the barefeats link for the dual CPU gets great scores in things that are actually dual CPU optimized like the cinebench and the PS tests. You argue a dual won't help and then produce benchmarks that show they completely do with the right software. You also didn't notice that the only dual G4 upgrade in those tests is a 7448.. not a 7447 like the OP is interested in.
Those numbers you rely on from Xbench are sketchy at best. If you're going to throw out benchmark numbers then at least choose a reputable app. You are really so foolish as to believe a benchmark that tells you a G3 logic board can move anything at that speed.
Shape truths to your delusions all you want but your arguments are virtually all without merit.
Those numbers you rely on from Xbench are sketchy at best. If you're going to throw out benchmark numbers then at least choose a reputable app. You are really so foolish as to believe a benchmark that tells you a G3 logic board can move anything at that speed.
Shape truths to your delusions all you want but your arguments are virtually all without merit.
Last edited:
Right, because a synthetic benchmark that relies heavily on cache and isolates the processor from the I/O subsystem is a good indication of real-world performance
I didn't say it wouldn't be faster in any cases, I said it wouldn't be fast enough to justify the price difference. Or at least that was my implication.Aside from Cinebench, the improvements were lackluster to say the least. Obviously throwing on another CPU will show SOME improvement, but it was rarely more than 10-20%, which is a bit pathetic.
And it being a 7447 is even worse... they're slower, as they feature half the cache of the 7448, which is rather important for bandwidth-starved systems. One of the chips was a 7447, by the way, check again.
Like I said, Geekbench reports ~200 MB/s. Other applications have reported the same.
At the very least, back up your arguments with evidence. All you're doing is bandying about conjecture, hurling petty invective, and coming off as someone who talks a lot but doesn't know very much.
You rely on a few tests from one url to be an example for a dual CPU's overall worth. The only 10-20% boost you refer to is software not optimized for dual. Do you not understand that? Those tests also don't display the performance gain from the most dual CPU optimized software on the Mac.. OS X. A dual will be faster even when there is not optimized sw and far faster when there is. I have owned 3 dual Macs in my days with the dual 1.42 MDD being the fastest. I base my every word on actual experience and the "proof" is all around you but I guess you're too limited in your thinking to see it.
In your first post of our exchange you list xbench gives you 300MB/sec then you go on to say in your 2nd post that it's 200MB/sec. How did you lose this 100MB/sec overnight? And about that copy/fill test in xbench.. my G4 gets 5.7GB/sec in that same test you get 1.1GB/sec. Thats over 5x faster memory on a 100mhz bus Sawtooth. The differences you claim to not be so different are in fact extremely so. Your thinking is so limited that you base far too much weight on the bus speed. It's about the technology onboard and how it works together. Any AGP G4 board would belittle any G3 board no matter what CPU it had. You can't even put dual upgrades in any G3's because the boards simply can't cope with it. A G4 can easily.
Say whatever you want but I base everything I say on things I have actually seen and experienced and not just a flood of jargon like you throw out. Any PowerPC user should understand all this perfectly. You get lost in bus mhz etc. like a wintel guy and seem to completely forget how RISC architectures get things done.
In your first post of our exchange you list xbench gives you 300MB/sec then you go on to say in your 2nd post that it's 200MB/sec. How did you lose this 100MB/sec overnight? And about that copy/fill test in xbench.. my G4 gets 5.7GB/sec in that same test you get 1.1GB/sec. Thats over 5x faster memory on a 100mhz bus Sawtooth. The differences you claim to not be so different are in fact extremely so. Your thinking is so limited that you base far too much weight on the bus speed. It's about the technology onboard and how it works together. Any AGP G4 board would belittle any G3 board no matter what CPU it had. You can't even put dual upgrades in any G3's because the boards simply can't cope with it. A G4 can easily.
Say whatever you want but I base everything I say on things I have actually seen and experienced and not just a flood of jargon like you throw out. Any PowerPC user should understand all this perfectly. You get lost in bus mhz etc. like a wintel guy and seem to completely forget how RISC architectures get things done.
Photoshop is threaded reasonably well. Obviously the others are at least threaded, or there would be no performance gain.
Unless the machine's so bandwidth starved that it doesn't help much at all. Or the application is written in C, or another language that doesn't support threading.
That's nice..
Great. So you've run a machine with two 1.42 GHz processors each outfitted with a full 1 MB of cache and sharing 1.33 GB/s bandwidth, vs. putting two, 2 GHz (or 1.8 GHz) chips sporting only 512 KB of cache in a machine sharing 1 GB/s bandwidth. Fantastic comparison. You do realize the reason Apple started putting 2 MB of SRAM L2 cache in their G4s was to make up for the horrific mem/FSB bandwidth bottlenecks, right? Shame there was never really a PPC7470.
Mkay. I could make claims to the opposite and say the proof was "all around you," but that wouldn't make what I said any more true. Unless you're going to provide actual evidence, people aren't going to buy what you're saying when you're arguing with someone who is in fact supporting what they say with demonstrative evidence.
Um. Actually if you'd read my first post a bit more carefully you'd notice I said XBench gives me ~300MB/s, and GeekBench (and a couple others) give me in the 200+ MB/s range.
Computer got tired. But don't worry, I gave it coffee.
Maybe your best is 5.7... mine's ~1.6 GB/s. The lowest I score is around 800 MB/s. The Fill test, as I mentioned, is somewhat meaningless.. or at least has no real relationship with memory speed. Obviously the code is small enough to run in cache, which is what it does.
Given that the 1 GHz 7455 in my G3 has 1 MB of backside L2 cache in the form of two 512 KB SRAM chips running at 250 MHz (256-bits wide), it's going to have a much higher latency and a far slower clock than what I can only assume must be in your G4.. a 7447 or 7448 (likely a 48), of on-die L2 cache running at 1:1 with the CPU.
Obviously neither my machine nor yours is capable of the scores it posts-- the Beige G3 has a maximum theoretical memory bandwidth of 533 MB/s, and the Sawtooth 800 MB/s. This is measuring cache speed, not memory bandwidth. If they'd made a 7448 for the G3, I'd be able to hit 5 GB/s+ too.
As I've explained, no. Fill measures cache speed. If you want accurate memory bandwidth tests, check the Fill in Geekbench.
>_> Throw a 500 MHz chip back in your Sawtooth and see how it stacks up. Hell, why don't you just post your XBench subscores here? Especially the memory tests.
Erm... there was actually a Dual 500 MHz G4 upgrade for a short time from DayStar for Beige G3s. What exactly is it about the G3 logic board that makes you think it... "can't cope with it"?
I can plainly see that you haven't.
PowerPC ceased to be RISC a long time ago... and the CISC vs. RISC argument was a bit silly to begin with. You do realize extra stages were added to that G4+ you've got in there in order to ramp up the clockspeed, right? RISC was never really inherently "superior" to CISC in the first place. *That's* jargon.
If everything could be run out of cache, then sure, memory bandwidth wouldn't really matter. But that's not how things work.
You see.. the issue here is that you believe your delusions as fact. I can't argue against deeply embedded misconception. Then you take the numbers from said delusions and slip them into a bunch of over killed long winded reply's and think thats argument.
BTW.. a 7455 has 256KB L2 and 1 or 2MB L3. A 7448 has 1MB L2 and it's the 7447 that has 512KB L2. You don't even understand the specs of your own CPU.
I will never agree with you so this is all just pointless. It's also against the rules to dominate other peoples threads with your own interests.
BTW.. a 7455 has 256KB L2 and 1 or 2MB L3. A 7448 has 1MB L2 and it's the 7447 that has 512KB L2. You don't even understand the specs of your own CPU.
I will never agree with you so this is all just pointless. It's also against the rules to dominate other peoples threads with your own interests.
*rolls eyes* You're right, I misspoke. How awful of me, how could I make such a mistake afer getting home from work after getting up at 4 in the morning?
Yes it's got 256k L2, and the SRAM is the 1 MB off-chip cache. That isn't really the point, as the Fill code in XBench fits in the SRAM, not the 256KB on-chip.
As for the 7447 and 48.. that's exactly what I said. If you'll read through my posts, you'll basically see that everything I said pretty much shot down what you've been saying pretty conclusively.
You tried looking at your cache bandwidth and calling it memory bandwidth... good job :/ You also said a number of things (i.e. G3s can't support dual chips) that simply weren't true. But you don't admit any of your mistakes, you just go back on the offensive with more vague statements and counterclaims without evidence.
Everything I've said has been factual.. you've just made a few vague statements and pointed out a few times where I've misspoke... but that had really little relevance to the argument at hand.
I'm fairly certain you realize I'm right, or you wouldn't have spent all this time arguing without posting anything to support your claims.
Yes it's got 256k L2, and the SRAM is the 1 MB off-chip cache. That isn't really the point, as the Fill code in XBench fits in the SRAM, not the 256KB on-chip.
As for the 7447 and 48.. that's exactly what I said. If you'll read through my posts, you'll basically see that everything I said pretty much shot down what you've been saying pretty conclusively.
You tried looking at your cache bandwidth and calling it memory bandwidth... good job :/ You also said a number of things (i.e. G3s can't support dual chips) that simply weren't true. But you don't admit any of your mistakes, you just go back on the offensive with more vague statements and counterclaims without evidence.
Everything I've said has been factual.. you've just made a few vague statements and pointed out a few times where I've misspoke... but that had really little relevance to the argument at hand.
I'm fairly certain you realize I'm right, or you wouldn't have spent all this time arguing without posting anything to support your claims.
Sorry but you constantly keep proving to be delusional. You have convinced yourself that you're right and even that I think you're right when I clearly said I will never agree with you.
I never saw a single thing to actually support one thing you claim. Just a couple sad attempts at using narrow situations as broad results. Combine that with the fact that everything I know as fact from hands on experience (not delusion) is disputed by you.
I am always happy to admit when I'm wrong and I know I'm not. I doubt that will make any difference to you as you seem to block out any sense of reality.
I officially exit from this sad discussion. Please don't give people here computer advice. You will sway them in the wrong way no matter what the request I imagine.
I never saw a single thing to actually support one thing you claim. Just a couple sad attempts at using narrow situations as broad results. Combine that with the fact that everything I know as fact from hands on experience (not delusion) is disputed by you.
I am always happy to admit when I'm wrong and I know I'm not. I doubt that will make any difference to you as you seem to block out any sense of reality.
I officially exit from this sad discussion. Please don't give people here computer advice. You will sway them in the wrong way no matter what the request I imagine.