I don't have results on Leopard because I always ran dual boot OS 9.2.2 & Tiger on it, but here is the 2.2.0 score on Tiger:
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.
G4 7448 and 7457 swap
- Thread starter JoyBed
- Start date
- Sort by reaction score
The 256 KB of L2 is probably the main limiting factor of the setup. I have tried out dual 7457s to make up for this since they have 512 KB of L2, but I have never witnessed a 7457 that will actually run stable above ~1.58 GHz, so I usually don't bother with them. Too bad Motorola didn't make a "prototype" 1400 MHz rated 7457 as that would have been incredible!!
Regardless of Geekbench scores though, "feel" matters alot too with these machines. With a 64-bit SeriTek card and an SSD, this is probably one of the snappiest PPC G4 Macs I've seen yet.
I am working on taking it even further by trying to get even faster rated L3 cache chips to work. I am targeting somewhere in the ballpark of 370 - 400 MHz. So far no luck, but if I can get them to work I will post about it.
Last edited:
one of the things to keep in mind is GB tests both the CPU and Memory bandwidth of the system and uses both results to calculate the final score
and memory bandwidth of G4's aint fantastic due to MAX bus which was showing its age somewhat, and esp with 2 fast CPU's like this can get somewhat congested
note that if you look at the actual processor sub scores, they are much higher then the memory sub scores
and these lower memory scores can skew a Geekbench result somewhat
so where your doing work which entirely resides in L2/L3 cache and does not require too much communication between the CPU and or back to main system memory, this system will be very performant still I imagine
Do you know howto change L3_CLK bits to change Core-to-L3 frequency?
And do somebody know, if it is possible to use 4MB L3 with 7457 CPU ( i.e. if L3_ADDR 18 is wired on CPU board, ideally somewhere near the memory chips )?
And if it is possible, what is purpose of additional +2 MB of private memory? CPU Technical Data Sheet is very silent about use of private memory.
I was afraid about such high voltage ( nominal is 1.6V ). Your results are very nice, I will test it again with 1.8V, if I solve better cooling.
The core-to-L3 ratio is set in hardware via a serial EEPROM soldered onto the daughtercard. I am looking into how that ratio is encoded, but currently I don't know. I will post if/when I figure it out.
The absolute maximum Vcore voltage for a 7455 is 1.95V per the datasheet. Yes, until you are able to effectively control your temps, I would absolutely not increase the voltage higher than stock.
That's an impressive overclock, thanks for sharing your results. You must be pushing 150w like that; how are you cooling it? Is this one of those water cooling setups like aquamac used to do for MDD computers?
Haha, quite honestly I have never measured the power consumption but yeah it is quite a beast. One of my favorite machines
Believe it or not, the cooling is 100% stock. This is a real 1.42 GHz MDD machine so it has the copper heatsink and original loud Delta SHE fan. I never do any modifications that would hack up the case or destroy an otherwise really cool looking machine.
I use a few simple tricks inside the case to increase efficiency of the airflow across the heatsink, but other than that, cooling is achieved using a very aggressive fan curve. I have tested the quieter PAPST fan that came with some MDDs and it does not push enough air. I will actually soon be experimenting with a new, modern production Delta fan (same dimensions) which pushes almost double the amount of air as the original. This should be able to cool even more effectively with (hopefully) less noise.
Before CPU swap I look into power consumption of overclocked CPUs to check cooling conditions.
All datas are from NXP/Freescale technical data sheets and exactly match CPU partnumber.
1. XC7455B RX1400PF = real CPU used in Powermac G4 2x 1.42 GHz:
![Power-7455B-1400.jpg]( "Power-7455B-1400.jpg")
Unfortunately XC7455B datasheet has only two Power consumption values - for 1333 and 1400 MHz. The third one ( 1000 MHz ) is from CPU type MC7455A RX1000LG and used for calculation of polynomial trendline.
Graph form two original data points is only linear aproximation and looks:
2. XC7455B RX1400PF - linear
![Power-7455B-1400-linear.jpg]( "Power-7455B-1400-linear.jpg")
If we replace CPUs for more modern ones ( but with lower nominal frequency ) it looks like:
3. MC7457 RX1267LC - new CPU candidate:
![Power-7457-1267.jpg]( "Power-7457-1267.jpg")
Of course, these results are theoretical calculation, but not far from reality.
So if I have frequency somewhere around 1600 MHz
7455B: 63 W max
7457: 33 W max
And if @socom_22 said that he used stock copper cooling for 1667 MHz. i.e. 2x 70W, it is safe to use even the aluminium cooler for 7457 CPUs.
Fortunatelly I have 1.42 version with copper cooler, so at first I try overclock 7455B, and then I use second CPU daughterboard to 2x7457 CPUs.
P.S. Original 7455B CPU from second board are hard working in my AmigaOne XE ;-)
All datas are from NXP/Freescale technical data sheets and exactly match CPU partnumber.
1. XC7455B RX1400PF = real CPU used in Powermac G4 2x 1.42 GHz:
Unfortunately XC7455B datasheet has only two Power consumption values - for 1333 and 1400 MHz. The third one ( 1000 MHz ) is from CPU type MC7455A RX1000LG and used for calculation of polynomial trendline.
Graph form two original data points is only linear aproximation and looks:
2. XC7455B RX1400PF - linear
If we replace CPUs for more modern ones ( but with lower nominal frequency ) it looks like:
3. MC7457 RX1267LC - new CPU candidate:
Of course, these results are theoretical calculation, but not far from reality.
So if I have frequency somewhere around 1600 MHz
7455B: 63 W max
7457: 33 W max
And if @socom_22 said that he used stock copper cooling for 1667 MHz. i.e. 2x 70W, it is safe to use even the aluminium cooler for 7457 CPUs.
Fortunatelly I have 1.42 version with copper cooler, so at first I try overclock 7455B, and then I use second CPU daughterboard to 2x7457 CPUs.
P.S. Original 7455B CPU from second board are hard working in my AmigaOne XE ;-)
Were the chips you used MC7457s or SC7457s? Do you have any photos of the chips that you can share?
1.3V seems quite low for stable running at 1.6 GHz. Are you running with L3 cache or cacheless? Maybe you simply have incredibly high quality chips, as every SC7457 I've tried so far hit a clock speed wall at or below ~1.58 regardless of voltage or cooling supplied.
If possible, can you please post a link to the original datasheet you found for the RX1400PF? Every MPC7455 datasheet I have is either for the 7455A models, or is a later B revision with the 1400 MHz part information removed.
Yeah, I think the aluminum heatsink should be OK for 7457s, regardless of clock speed.
I want to emphasize though. While I do use the copper heatsink, under load I have the fan running really fast, basically at full tilt, in order to keep temps below 50 C. It required a highly customized fan curve -- I am quite certain that without it, the temp sensor would hit 62C+ and the chips would overheat within a minute or two using the default fan curve. Be very careful if trying this out...
The chips are the "MC" version and look like any others. I guess they are new enough that the substrate has the freescale logo instead of motorola. Yes the L3 is enabled. I think this is the whole point of the 7457, otherwise they are the same as 7447 chips which can often clock higher. They did need more voltage at 1.73GHz but after a year I eventually got tired of the fan noise. I think I also tried them at 1.8 but decided it would run too hot without a more involved cooling setup.
![dual7457.png]( "dual7457.png")
Wow, what voltage are you running these at? I could never get mine stable at above 1.5 GHz. Also, check and make sure the chips are actually making good contact with the heatsink. The normal 0.76MM solder balls used will cause the chip to sit considerably lower than stock, and I found that originally my MDD heatsink had a pretty significant gap between itself and the chips. You can check the gap by just looking at how the thermal paste is after removing the heatsink; if it has a very thin layer and looks to be pushed out around the edges of the chip, you're good, but if it looks like a uniform thicker glob, there is too much space.
I am not found original link, so look on attachment.
Excellent result!
And do you try also 167 MHz bus frequency? Or you just leave default value?
Oh yes, it is Quicksilver. I also don't how change bus frequency here. But for sure it is fastest Quicksilver ever !
I don't believe that the Quicksilver bus can be overclocked. Even if someone found a way, it would almost certainly require PC166 SDRAM, which is pretty difficult to find these days and may not even work in a Power Mac. The Sawtooth/Gigabit/Cube machines can apparently overclock their bus up to 133MHz, but no one seems to be able to get it working above 120MHz. Even at 120, I've only seen benchmark results and nothing in system profiler that suggests it is actually running at that speed. The only Power Macs I'm aware of that can easily tolerate a bus overclock is moving a 133MHz MDD to 167MHz.
Here’s a hack / mod that supposedly attains 150 MHz:
![Final.png]( "Final.png")
I soldered a DIP switch into the S3 position on my Gigabit logic board today to try overclocking the bus. I started with a 7400 set to a 4x multiplier. Bumping up to 120MHz does actually boot. Mac OS does report the CPU speed at only 400MHz and the bus at 100MHz regardless. The benchmark results really surprised me though; I saw about a 20% improvement over stock in CPU and floating point scores in MacBench 5. I also tried 133MHz and the machine actually chimed, but only resulted in a black screen and no further sounds. I'm going to try again with a 7410 when I get another working one in a few days, but I doubt I'll be able to get 133MHz working. The clock multiplexer for the memory (PI6C180) is only rated for 100MHz and probably can't time the RAM properly at 133MHz even if the CPU can support it. There are the pin compatible PI6C180A and PI6C180B that support faster clocks, but they haven't been manufactured in years and are probably impossible to track down. If anyone ever gets a hold of one though, I'm inclined to believe that one of these early G4 Power Macs could actually run a 133MHz bus with a better PI6C180 installed.
I also dug up a schematic of an iMac G4 that uses the same C5003 clock chip as the Quicksilver. It does actually look like it supports up to 150MHz. Pins 18, 19, and 20 are the controls (likely wired to the S3 holes, but I don't have a quicksilver to confirm). Pin 18 pulled low is the default for 133MHz, but removing a resistor from it and pulling pin 19 low instead should set the bus speed to 150MHz. None of the stock CPUs and most memory available aren't rated for that speed, but it isn't much of a jump and will probbaly work.
I also dug up a schematic of an iMac G4 that uses the same C5003 clock chip as the Quicksilver. It does actually look like it supports up to 150MHz. Pins 18, 19, and 20 are the controls (likely wired to the S3 holes, but I don't have a quicksilver to confirm). Pin 18 pulled low is the default for 133MHz, but removing a resistor from it and pulling pin 19 low instead should set the bus speed to 150MHz. None of the stock CPUs and most memory available aren't rated for that speed, but it isn't much of a jump and will probbaly work.
I was playing around with my Power Macintosh 7600 the other day and noticed that the XLR8 MACHSpeed Control shows the bus and CPU speed in Mac OS 8.6 and 9.2.2. I have a CPU card that allows bus overclocking in that machine and the XLR8 software always reports it correctly. I figured I'd give MACHSpeed a shot on my Gigabit G4 and it actually does work. It shows that I really do have a 120MHz bus speed in that machine and the CPU is running at 480MHz. This is the first time I've ever seen something actually displaying bus overclocks on these older machines.
Last edited:
I have confirmed that my Gigabit Power Mac will not boot at 133MHz even with a CPU installed that supports that bus speed. Still chimes, but never get video out and it will not respond to an PRAM reset either. That fact also makes me wonder if the Apple north/southbridge chips can't handle 133MHz and just don't respond properly. The Cube can also have its bus overclocked to 120MHz by setting the appropriate resistors on the SG500 chip similar to the Sawtooth/Gigabit machines.