Our obsession with benchmarks and Geekbench continues! Earlier this year we did a couple of Geekbench Comparisons using a number of different machines. While the comparisons were popular, a number of people thought we were being, well, less than fair since we didn’t have enough high-end CPUs from AMD and Intel.
So, for this Geekbench comparison, I’m including most of the recent AMD and Intel desktop processors (along with some not-so-recent desktop processors) along with a wide variety of recent Mac models, ranging from the PowerBook G4 to the Mac Pro.
This time ’round, instead of using results from a single computer, I’m using average results from all the computers in the Geekbench Result Browser database (provided the results meet certain criteria outlined below). To make sure I’ve got a good sample size for each computer, I’m only including computers (or processors) that have at least three distinct results in the result browser. This should help compensate for differences in components (such as RAM) between different computers.
Setup
For Intel and AMD processors, results were collected from Geekbench for Windows (32-bit) running on systems with processors running at stock speed (i.e., no over-clocked processors) and with at least 512MB of RAM.
For Mac systems, results were collected from Geekbench for Mac OS X (32-bit) running on systems with standard processors (i.e., no over-clocked processors or processor upgrades) and with at least 512MB of RAM.
I’m reporting the average overall score for each processor, where 100 is the score a Power Mac G5 @ 1.6GHz would receive. As always, higher scores are better.
AMD Processors
AMD Athlon 64
AMD Athlon 64 2800+ (Socket 754)
AMD Athlon 64 3000+ (Socket 939)
AMD Athlon 64 3000+ (Socket AM2)
AMD Athlon 64 3200+ (Socket 754)
AMD Athlon 64 3200+ (Socket 939)
AMD Athlon 64 3200+ (Socket AM2)
AMD Athlon 64 3300+ (Socket 754)
AMD Athlon 64 3400+ (Socket 754)
AMD Athlon 64 3400+ (Socket 939)
AMD Athlon 64 3500+ (Socket 939)
AMD Athlon 64 3500+ (Socket AM2)
AMD Athlon 64 3700+ (Socket 754)
AMD Athlon 64 3700+ (Socket 939)
AMD Athlon 64 3800+ (Socket 939)
AMD Athlon 64 3800+ (Socket AM2)
AMD Athlon 64 X2
AMD Athlon 64 X2 3800+ (Socket 939)
AMD Athlon 64 X2 3800+ (Socket AM2)
AMD Athlon 64 X2 4200+ (Socket 939)
AMD Athlon 64 X2 4200+ (Socket AM2)
AMD Athlon 64 X2 4600+ (Socket 939)
AMD Athlon 64 X2 4600+ (Socket AM2)
AMD Athlon 64 X2 4800+ (Socket 939)
AMD Athlon 64 X2 5000+ (Socket AM2)
You can see differences in performance for each processor depending on its socket type (thanks to the type of RAM each socket uses); Socket 754 (which uses single-channel DDR RAM ) is slower than Socket 939 (which uses dual-channel DDR RAM), which in turn is slower than Socket AM2 (which uses dual-channel DDR2 RAM)
You can also see that AMD model numbers are fairly consistent; an AMD Athlon 64 3800+ offers roughly the same performance as an AMD Athlon 64 X2 3800+ (180.2 vs 172.5 for Socket AM2), Unfortunately both are slower than a Pentium 4 3.8GHz (211.0), which suggests the model numbers are somewhat optimistic when compared to Intel CPUs.
Intel Processors
Intel Pentium 4
Intel Pentium 4 2.00GHz (Socket 423)
Intel Pentium 4 2.20GHz (Socket 478)
Intel Pentium 4 2.26GHz (Socket 478)
Intel Pentium 4 2.40GHz (Socket 478)
Intel Pentium 4 2.53GHz (Socket 478)
Intel Pentium 4 2.60GHz (Socket 478)
Intel Pentium 4 2.66GHz (Socket 478)
Intel Pentium 4 2.66GHz (LGA 775)
Intel Pentium 4 2.80GHz (Socket 478)
Intel Pentium 4 2.80GHz (LGA 775)
Intel Pentium 4 2.93GHz (Socket 478)
Intel Pentium 4 2.93GHz (LGA 775)
Intel Pentium 4 3.00GHz (Socket 478)
Intel Pentium 4 3.00GHz (LGA 775)
Intel Pentium 4 3.06GHz (Socket 478)
Intel Pentium 4 3.06GHz (LGA 775)
Intel Pentium 4 3.20GHz (Socket 478)
Intel Pentium 4 3.20GHz (LGA 775)
Intel Pentium 4 3.40GHz (Socket 478)
Intel Pentium 4 3.40GHz (LGA 775)
Intel Pentium 4 3.60GHz (LGA 775)
Intel Pentium 4 3.80GHz (LGA 775)
Intel Pentium D
Intel Pentium D 805 (2.66GHz)
Intel Pentium D 820 (2.80GHz)
Intel Pentium D 830 (3.00GHz)
Intel Pentium D 840 (3.20GHz)
Intel Pentium D 915 (2.80GHz)
Intel Pentium D 920 (2.80GHz)
Intel Pentium D 925 (3.00GHz)
Intel Pentium D 930 (3.00GHz)
Intel Pentium D 940 (3.20GHz)
Intel Pentium D 945 (3.40GHz)
Intel Pentium D 950 (3.40GHz)
Intel Pentium D 955 Extreme Edition (3.46GHz)
Intel Pentium D 965 Extreme Edition (3.73GHz)
Intel Core 2
Intel Core 2 Duo E6300 (1.86GHz)
Intel Core 2 Duo E6400 (2.13GHz)
Intel Core 2 Duo E6600 (2.40GHz)
Intel Core 2 Duo E6700 (2.66GHz)
Intel Core 2 Duo X6800 (2.93GHz)
Looking at these numbers, Pentium D processors don’t seem particularly attractive; on one end, high-end single-core Pentium 4 processors offer better performance, while on the other end dual-core Core 2 chips offer better performance and better power-consumption. I’m not surprised that it’s getting harder and harder to buy Pentium D processors these days (save for the low-end Pentium Ds).
Another interesting thing to note is the jump in performance between the Core 2 Duo E6400 and the Core 2 Duo E6600; moving from 2×1MB to 2×2MB of L2 cache makes a big difference for the Core 2 Duo processor. Speaking of Core 2 Duo processors, the Core 2 Duo X6800 is the fastest dual-core processor in this comparison (with only the dual-processor dual-core Mac Pro @ 3.0GHz being faster overall).
Mac Models
iBook and MacBook
iBook G4 (PowerPC G4 (7450) @ 799MHz x 1)
iBook G4 (PowerPC G4 (7450) @ 933MHz x 1)
iBook G4 (PowerPC G4 (7450) @ 1.0GHz x 1)
iBook G4 (Early 2004) (PowerPC G4 (7450) @ 1066MHz x 1)
iBook G4 (Early 2004) (PowerPC G4 (7450) @ 1199MHz x 1)
iBook G4 (Early 2004) (PowerPC G4 (7450) @ 1.3GHz x 1)
iBook G4 (Mid 2005) (PowerPC G4 (7450) @ 1.3GHz x 1)
iBook G4 (Mid 2005) (PowerPC G4 (7450) @ 1420MHz x 1)
MacBook (Intel Core Duo @ 1.83GHz x 2)
MacBook (Intel Core Duo @ 2.00GHz x 2)
MacBook (Late 2006) (Intel Core 2 Duo @ 2.00GHz x 2)
Mac mini
Mac mini (PowerPC G4 (7450) @ 1.5GHz x 1)
Mac mini (Late 2005) (PowerPC G4 (7450) @ 1.3GHz x 1)
Mac mini (Late 2005) (PowerPC G4 (7450) @ 1.5GHz x 1)
Mac mini (Early 2006) (Intel Core Solo @ 1.50GHz x 1)
Mac mini (Early 2006) (Intel Core Duo @ 1.66GHz x 2)
iMac
iMac (Flat-Panel) (PowerPC G4 (7450) @ 700MHz x 1)
iMac (Flat-Panel) (PowerPC G4 (7450) @ 800MHz x 1)
iMac G4 (17-inch Flat-Panel) (PowerPC G4 (7450) @ 800MHz x 1)
iMac G4 (20-inch Flat-Panel) (PowerPC G4 (7450) @ 1.25GHz x 1)
iMac G5 (PowerPC G5 (970) @ 1.6GHz x 1)
iMac G5 (PowerPC G5 (970) @ 1.8GHz x 1)
iMac G5 (Ambient Light Sensor) (PowerPC G5 (970) @ 1.8GHz x 1)
iMac G5 (Ambient Light Sensor) (PowerPC G5 (970) @ 2.0GHz x 1)
iMac G5 (iSight) (PowerPC G5 (970) @ 1899MHz x 1)
iMac G5 (iSight) (PowerPC G5 (970) @ 2100MHz x 1)
iMac (Early 2006) (Intel Core Duo @ 1.83GHz x 2)
iMac (Early 2006) (Intel Core Duo @ 2.00GHz x 2)
iMac (Late 2006) (Intel Core 2 Duo @ 2.00GHz x 2)
iMac (Late 2006) (Intel Core 2 Duo @ 2.16GHz x 2)
iMac (24-inch) (Intel Core 2 Duo @ 2.16GHz x 2)
iMac (24-inch) (Intel Core 2 Duo @ 2.33GHz x 2)
PowerBook and MacBook Pro
PowerBook G4 (DVI) (PowerPC G4 (7450) @ 800MHz x 1)
PowerBook G4 (1 GHz/867 MHz) (PowerPC G4 (7450) @ 1000MHz x 1)
PowerBook G4 (12-inch) (PowerPC G4 (7450) @ 867MHz x 1)
PowerBook G4 (17-inch) (PowerPC G4 (7450) @ 1.0GHz x 1)
PowerBook G4 (15-inch FW800) (PowerPC G4 (7450) @ 1.25GHz x 1)
PowerBook G4 (12-inch 1.5 GHz) (PowerPC G4 (7450) @ 1.5GHz x 1)
PowerBook G4 (Double-Layer SD, 15-inch) (PowerPC G4 (7450) @ 1.67GHz x 1)
PowerBook G4 (Double-Layer SD, 17-inch) (PowerPC G4 (7450) @ 1.67GHz x 1)
MacBook Pro (Intel Core Duo @ 1.83GHz x 2)
MacBook Pro (Intel Core Duo @ 2.00GHz x 2)
MacBook Pro (Intel Core Duo @ 2.16GHz x 2)
MacBook Pro (17-inch) (Intel Core Duo @ 2.16GHz x 2)
MacBook Pro (Core 2 Duo) (Intel Core 2 Duo @ 2.16GHz x 2)
MacBook Pro (Core 2 Duo) (Intel Core 2 Duo @ 2.33GHz x 2)
MacBook Pro (Core 2 Duo 17-inch) (Intel Core 2 Duo @ 2.33GHz x 2)
Power Mac and Mac Pro
Power Mac G4 (Mirrored Drive Doors) (PowerPC G4 (7450) @ 1.0GHz x 1)
Power Mac G4 (Mirrored Drive Doors) (PowerPC G4 (7450) @ 1.25GHz x 1)
Power Mac G4 (Mirrored Drive Doors) (PowerPC G4 (7450) @ 866MHz x 2)
Power Mac G4 (Mirrored Drive Doors) (PowerPC G4 (7450) @ 1.0GHz x 2)
Power Mac G4 (Mirrored Drive Doors) (PowerPC G4 (7450) @ 1.25GHz x 2)
Power Mac G4 (Mirrored Drive Doors) (PowerPC G4 (7450) @ 1.3GHz x 2)
Power Mac G4 (Mirrored Drive Doors) (PowerPC G4 (7450) @ 1.42GHz x 2)
Power Mac G5 (PowerPC G5 (970) @ 1.6GHz x 1)
Power Mac G5 (PowerPC G5 (970) @ 1.8GHz x 1)
Power Mac G5 (PowerPC G5 (970) @ 1.8GHz x 2)
Power Mac G5 (PowerPC G5 (970) @ 2.0GHz x 2)
Power Mac G5 (June 2004) (PowerPC G5 (970) @ 1.8GHz x 2)
Power Mac G5 (June 2004) (PowerPC G5 (970) @ 2.0GHz x 2)
Power Mac G5 (June 2004) (PowerPC G5 (970) @ 2.3GHz x 2)
Power Mac G5 (June 2004) (PowerPC G5 (970) @ 2.5GHz x 2)
Power Mac G5 (June 2004) (PowerPC G5 (970) @ 2.7GHz x 2)
Power Mac G5 (Late 2005) (PowerPC G5 (970) @ 2.5GHz x 4)
Mac Pro (Intel Xeon @ 2.00GHz x 4)
Mac Pro (Intel Xeon @ 2.66GHz x 4)
Mac Pro (Intel Xeon @ 3.00GHz x 4)
Mac mini performance almost doubled with the switch from a PowerPC G4 @ 1.5GHz to an Intel Core Solo at 1.5GHz with no increase in clock speed. Zomg! iBook and PowerBook (er, MacBook and MacBook Pro) performance also increased dramatically with the switch to the Intel Core Duo (and later to the Intel Core 2 Duo), especially since both went from single-core to dual-core processors.
iMac and PowerMac (er, Mac Pro) performance didn’t benefit as much with the switch to Intel processors since the PowerPC G5 isn’t nearly as old and busted as the PowerPC G4, but the high-end Mac Pro does have a considerable performance advantage over the top-end Power Mac G5. Even the low-end Mac Pro is competitive against the high-end Power Mac G5 (and is substantially cheaper to boot).
Final Thoughts
It’s interesting to see where the bottlenecks in processor and memory performance have occurred over the years (hello, PowerPC G4!) and how processor and computer manufacturers worked around these bottlenecks (hello, Intel!). Really, though, it’s graphs like these that make me want to “accidentally” spill a gallon of water on my three-year-old PowerBook G4 so I can go out and get a shiny new MacBook Pro.
What about Intel Xeon ?
I can’t find, what is the whole geekbench about :/
Whot set of tests it makes etc…
fork,
I didn’t include the Xeon (or the Opteron, for that matter) mostly due to a lack of results for both the Xeon and the Opteron.
Satai,
http://www.geekpatrol.ca/geekbench/benchmarks/ should tell you anything you need to know about the tests Geekbench runs.
ok now that you seen the false results please post some real scientific results too. The only valid and optimized client for the test is distributed.net’s dnetc. Re-do all the tests and you will be stunned how slower the new machines are.
Xeon in Mac Pro
I think this sentence is wrong:
Even the high-end Mac Pro is competitive against the high-end Power Mac G5 (and is substantially cheaper to boot).
Shouldn’t that be “low-end Mac Pro” ?
This does not accurately depict anything about the results. different motherboards are notorious for having lower benchmarks than other boards even with both set to the same settings. was the same exact memory used in every board possible, and if not did they have the exact timing? (CAS latency and the like).
also since you say you included more high end processors in the list, why do I not see a single FX chip anywhere on the list? those ARE high end procs from AMD. and for fairness include Intels EE chips in there too
Aku,
I didn’t include any Athlon FX processors because I didn’t have more than three results for any one Athlon FX processor. As I mentioned at the beginning, I only included processors with more than three results because I wanted to compensate for things like different motherboards and memory configurations.
Besides, the whole point of this article isn’t to figure out what the maximum performance is for each processor; rather, it’s to figure out what the expected (or average, if you prefer) performance is for each processor, even in non-optimal situations.
Mike,
Thanks for pointing that out. It’s been fixed.
Apostolos Karakoussis said:
The only reason dnetc runs so fast on the PPC is because its one of a handful of apps that really get a big boost from Altivec’s permute instruction. In most everything else, the new machines burn. My dainty white Macbook is faster than my big honking PowerMac (dual-core 2.3 GHz) for almost everything that doesn’t hit the HDD hard.
So geekbench is only running on two cores? That’s what I’m gathering from the scores on the C26800 vs the XEON 3.0. By all rights if GB was running on four cores, the performance should be on par with the difference that I see between my Homemade Media Center, and my MacPro running Boinc, the latter being about 87% faster.
Clay,
Geekbench runs tests both in single-threaded mode (where it runs on just one core) and multi-threaded more (where it runs on all cores), so doubling the number of cores won’t double the overall score.
Plus, Geekbench also tests memory performance, which is an area where performance is about the same on the Core 2 Duo and the Mac Pro (i.e., it’s not affected by the number of available cores).
Rayiner Hashem said:
Totally not correct. Sorry ’bout that. dnetc runs THAT (4-5x faster) fast because its one of the rare optimized applications that run at full speed on all platforms. Since it’s not using any ram and purely only the L1/2/3 caches it’s a pure sign of cpu power. This is the only true test on how much optimization you can get from the cpu.
Your machine if faster but your processor is NOT. Sorry dude.
Notice also that the intel compiler does 90%-ish speed of the best code someone could write by hand optimizing. On the G3/G4/G5 platforms this is not the case. All code has to be written by hand (gcc is crap as hell on ppc). This is the reason most applications run slow. Optimizing for Altivec is one option but accelerating and optimizing for the platform is something deeper.
dnetc is still the no1 choice for cpu benchmarks by far. Anything else out there is mostly synthetic or dependent from the board (and not the cpu purely).
Anyone thinking RC5 performance can be extrapolated to anything other than RC5 is an idiot.
The canonical reason for why RC5 performs so well on PPC machines w. altivec is because Altivec has a packed rotate instruction.
I have one concern about your chart- it doesn’t differentiate between Pentium 4 processors with and without Hyperthreading, which I’ve noticed makes a significant difference to the test results.
For example, you’ve got “Intel Pentium 4 2.80GHz (Socket 478)” at 123.4, while my P4 2.8GHz machine (with HT enabled) consistently scores in the 160-170 range. I just ran the test again and got a 165.6. At least, I can only assume that Hyperthreading is making the difference, because my machine is not overclocked in any way, its only real performance optimization is dual-channel RAM.
I realize that this only affects a small number of outdated processors, but forty points seems to me to be a big difference (it’s an extra iMac G4! Heh). The latest version of Geekbench does distinguish between logical and physical processors, so maybe you can make a distinction next time you publish one of these comparisons.
hunter do your homework. altivec rules in video and audio processing and will for some time more. your comments are not helping. results DO scale. even without rotation ops. Even the G5 suffered from having a weak altivec. A G4 was even more well suited for audio/video than a G5.
No education. No power.
I fail to see how that post had any relevance to what I said at all.
When AMD or Intel design/release a new chip, the designers must say, this chip “X” will fit in between Chip “V” and chip “Y”, i.e. it will be faster than “V”, but slower than “Y”, so they obviously have a chart which depicts a scale of processor to relative speed/horespower wihout having to take into account all the external factors of mother boards/memory configs etc?
Ian
While this is a good comparison it doesn’t take into account the PPC proccesors which are used in some of the top 10 supercomputers of the world are RISC proccesors and Intel/Amd are CISC proccesors.
It also doesn’t take into account the latest Power6 CPU from IBM. The G5 came out when the P4 was in it’s hay day of LONG pipelines to multiply proccesing power (the GHZ factor). The G5 PPC hasn’t been updated much until the Power6 CPU… Needless to say it would spank ANY CPU hands down in ANY comparison. But then again it’s like fighting a tank with a wicker jeep. You can’t compare to stable overclocking to 300ghz (0* F) with 32mb of level3 cahce running on a 4ghz bus…
Also while this sheet may show some good comparisions of Board/Cpu combo’s it lacks heavily in showing real world work. Comparing a Mac to a Windows box is hardly a fair fight. Especially without telling us the price factor involved. Apple builds there machines for straight perfomance of the OS. While the Intel/Amd’s can be combined with any multitude of memory and motherboard combo’s in a LARGE variety of manufacture and speeds.
What should be used to compare straight proccesing power is what was used back in the day. The time taken to compute Pie to the 1000000 power. At least back in the early 90’s-late 90’s this was a good compairson as it took multimedia/board configs/memory out of the picture almost entirely. Not sure what they use today but I know it’s a straight calculation… And even if you used the Pie method they’d all probably perform within nanoseconds of each other.
Most younger people don’t quite get the whole “System” as a whole speed, they just look at the CPU specs.
Otherwise Good Job.
I am going to preface this by admitting I am a big AMD fan. Anyhow, I can’t help but feel AMD is being sidelined here. You reported a score of 179.5 for an Athlon 64 X2 4800+. I have this processor and under Linux I get 214 and not much less under windows. Whats going on here? 34.5 points is rather significant.
Other scores seem off to me, for example you report the LGA775 P4 2.8 GHz at 123.4; however, I just ran it one one my low end dell servers with that CPU and got 166. Is it Linux?
I recommend running your Intel and AMD benchmarks on Linux, not just from my experience, but because you would be using a more similar complier.
Thanks for all your hard work. I’m a fan.
I get 228.6 with my OCed Athlon64 3200+ (Venice). I’m running it at ~2.6 GHz on Linux. It would go far beyond this, but i need a new southbridge cooler cause this one is shit (Asus A8N-E, the cooler for the nForce4 Ultra runs at ~6000 RPM and is quite loud).