ThinkSecret.com has posted a screenshot that shows results from Skidmarks GT, a benchmarking utility that’s part of Apple’s CHUD performance tools. Based on the test ThinkSecret ran, this provides a “rough comparison of G5 and G4 performance.”
“The Skidmarks scale has ‘100’ equal a Power Mac G4 at 1GHz,” ThinkSecret reports. “The Dual 2GHz received scores of 172 for integer performance, 270 for floating point performance, and 208 for vector performance.”
ThinkSecret also notes that the Power Mac G5 is running Mac OS X Version 10.2.7 known internally as “Smeagol,” and features screenshots of “About This Mac,” “Apple System Profiler,” and the “CPU Preference Pane.”
Full story here.
Interesting to note that a 2Ghz G5 is only 1.7 times faster than a 1.0 Ghz G4 for integer performance. Twice the speed, faster bus, and 64bits and can’t even do twice the performance?
I don’t think the Skidmarks’ app can fully test a dual CPU machine. If it can, then these benchmarks are even more disappointing.
But benchmarks don’t really apply in the real world. We need to get our hands on the machines and see what they can do.
Hopefully the new OS X with true 64bit optimizations and code written for a 64bit machine will show what the G5 can really do.
Al’s right.
Whats up with the meager performance increase of the dual G5 when compared to the single G4? Not very impressive at all. And this is on a system that has a 1GHz bus, over 5X faster than the G5’s, uses DDR like it supposed to be used (not like the hacked DDR in the G4). This is so far off what I expect that I’m wondering if its a mistake.
Lets not forget that the spec integer and floating point scores are much higher on the G5 vs the G4. So, I’m wondering if something is wrong.
– Mark
wait till apple installs the same amount of l3 cache in the g5, than the tests will be more enjoyable…
Yea, just like when the PIV came out, it was actually SLOWER than the PIII, until Windows XP was released.
Does anyone know if the G5 is a RISC or a CISC? I ask because the PIV was slower mostly because it had a more complex instruction set. Perhaps that’s the case with the G5. I have little doubt that Panther won’t drastically improve performance.
What about that fast user switching feature? I gizzed all over myself when I saw that! COOL!!
The G5, like all the PowerPC family, is RISC, & uses the instruction set originally developed for the IBM RS/6000. According to documentation I found elsewhere, Skidmarks is purely a CPU test & does not test overall system performance. And of course it’s 32-bit code, which means that it can’t do a fair test of a 64-bit system without recompiling.
As for L3 cache, don’t expect it in the G5. According to maconline.com, the L3 cache in the G4 runs at speeds ‘up to 500 MHz’. That’s barely faster than the DDR400 RAM in the G5. If you include latency time, a 500 MHz cache could actually slow the G5 down. (If the G5 goes to the cache for data & doesn’t find it, it then has to poll system RAM — two read operations instead of one. This is only worth doing if system RAM is significantly slower than cache.)
I’ve just downloaded Skidmarks myself & plan to report on the results I get.
I think what the skidmark test is honestly showing is just how poor the G5 is as an improvement over the g4!!
Lets face it when the revised g4 is put into laptops and other Macs, they are going to come awfully close to g5 integer performance. That is something to keep in mind.
It true seems like Apple has delievered to us, in the form of the G5, a machine that has the singular advantage of addressing more memory. And then they crippled the memory expansion capability. I will also give them the floating point advantage, but a faster bus on the g4 may very well deal with that.
I’m really hoping that Apple doesn’t go about hobbling the new G4’s when they come out.
Thanks
Dave
Remember that the article states that the test machine is running Smeagol, not Panther. We’ve been told all along that the processors will be crippled until 10.3 is released, so this shouldn’t be that much of a surprise.
If it’s twice as fast as the fastest G4s or Pentium IV’s while crippled, imagine how fast it will be when it’s un-crippled, when Panther comes out.
It will be a smokin’ machine but I still think Apple should go to AMD.
Am I missing something? If a single 1Ghz G4 = 100, and the Dual 2Ghz G5 = 172, 270, 208, isn’t that pretty poor performance? Even if the the baseline is a Dual 1Ghz G4, one might consider the performance about equal, considering the clock speed of the G5 is twice that of the G4. And all this in spite of the faster RAM and 64-bit data path in the G5…
PEOPLE! Hello!! The tests were performed on an operating system that was written for a different computer. Dispite that, the tests were still twice as fast.
That’s like running the same tests within VirtualPC on a G4/1GHz, having the same results (that it’s twice as fast in VPC) and saying it sucks.
Think about it people! The Pentium 4 was actually SLOWER than the Pentium 3 before Windows was optimized for the Pentium IV. PIV has different instruction sets, so the present version of Windows had to emulate the instructions.
The same damn thing is happening here. Guess why 10.2.7 (Smeagol) was made. It was made with software emulation that could run the G5. EMULATION. 10.3 (Panther) wasn’t ready by the time the tests were to be run.
Now, even with these things considered, IT’S roughly TWICE AS FAST! In emulation!! Clock speeds are supposed to reach up to 5Ghz.
So, stop saying it’s not very impressive. It is quite impressive!
By the way “skidmark” test? What? Did someone shit their drawers or something?
As promised, I ran Skidmarks on my dual G4/500. I can absolutely confirm that Skidmarks only measures the performance of a SINGLE cpu. (My results: Integer 45, FP 59, Vector 50. A single 500-MHz G4 should theoretically be half the speed of a 1-GHz G4, or 50 on all tests.)
The fact that I did not get 50 on all three tests suggests that the Skidmarks suite is not perfectly accurate. I wouldn’t be surprised if even a 1-GHz G4 showed small variations from 100, depending on exact system configuration & OS version.
‘Dave’ gripes about ‘just how poor the G5 is as an improvement over the g4’. Well, think about it, Dave! Skidmarks rates the speed of a SINGLE cpu, it’s written in 32-bit code, & it is running under a 32-bit version of the Mac OS. On one of the sites I checked, it was quite frankly stated that it scales linearly with clock speed. I haven’t looked at the source code (nor, I suspect, would I make much sense of it if I could), but from everything I have read, I’d guess Skidmarks is a pretty linear test & does not even make the best use of the multiple execution units on the G4. If so, it can hardly be expected to do better with the G5.
However, ‘MacDaddy’ is also mistaken on one point: The G5 doesn’t run G4 software in emulation. This is not anything like Virtual PC. It’s just that half the chip is sitting idle when running 32-bit code. So in effect, a G5 running a 32-bit OS is . . . well . . . a G4 with a higher clock speed & a really, really fast system bus. (Somebody tell Dave that a 1-GHz processor bus & 800-MHz point-to-point RAM bus beat the crap out of his G4’s 167 MHz bus shared among multiple devices.)
Meanwhile, Mathematica on a dual G5 runs rings round the same program on a Dell dual 3.06-GHz Xeon, which I guarantee you couldn’t do with any G4.
This is NOT a crippled chip. This is a HUGE upgrade over any version of the G4, including the vapourware 2-GHz G4 promised by Motorola.
This is getting interesting.
Alright, what I meant by the G5 being run on a foreign computer is exactly what you were talking about Jay. I guess I just chose a bad comparison. I see your point though.
Oh, just so you didn’t get me wrong, I never meant to say that it’s a crippled chip. I just meant that with Smeagol being the OS it was tested on, it was at a disadvantage.
I want to make another point. I don’t know much about microprocessor technology but I do know a bit about the physics of RF radiation (frequencies – clock speeds). I don’t know how they can bring a single CPU up to clock speeds of 5 Ghz or even 3 Ghz. I’m not sure at what frequency you can no longer transmit a signal via a metallic conductor, or silicone for that matter. I know that even at around 8 Ghz, you need a pretty small waveguide. The waveguide dimensions to pass a frequency are inversely proportional to the frequency. That is, as the frequency increases, you need a smaller waveguide. I know that around 30Ghz, it’s less than 1/4″ waveguide.
To get to my point, theoretically we won’t be seeing as much of an improvement in processor speeds as we have seen in the trends of the past 10 years. Not with the current silicone IC microprocessor technology.
Theoretically, we can go from using LASERS to RASERS (Radiation, rather than Light). Instead of light, Alpha, Gamma, Beta, X-rays can be used to etch the chips, but can we put any higher frequencies on a motherboard?
The G5 will evolve. There may be a significant improvement with the G6 (if we haven’t wiped ourselvees off the face of the Earth by then). But I think we’re coming close to a ceiling in the capabilities of computers as they are today. Until we come up with another way to make microchips, I think we’ll see things taper off soon.
If this G5 can reach 5Ghz within the next couple of years, that by itself may mean that it will be the fastest processor that we’ll see for some time. I’m convinced that it’s much faster than any Wintel chipset. Apple and IBM have something here. You’ll see. Oh, I’m glad I didn’t invest in Motorola.
Comments?
MacDaddy — I think we’re on the same wavelength. It was ‘Dave’ several comments back who called the G5 ‘crippled’.
You’re right, we’re approaching the limits of clock speed on silicon. I’ve heard suggestions that 30 GHz may be theoretically attainable, but at what cost in power consumption & reliability problems, I shudder to think. I have to laugh at Intel’s predicament, though. Having got half the world to believe that GHz = performance, they are now trying to explain why a 1-GHz Itanium is faster than a 3-GHz P4. I wish them joy of it.