- Joined
- Jun 4, 2012
- Messages
- 247
- Motherboard
- GA-Z77X-UD5H
- CPU
- i7-3770K
- Graphics
- HD4000
- Mac
- Mobile Phone
I took the opportunity to collect about 200 of the geekbench scores from the recent 'Post your GeekBench scores' thread. I've also cherry-picked a few Ivy Bridge and nice Sandy Bridge posts from the Golden- and User-builds categories. Along with these scores, I also collected the CPU, speed, motherboard and RAM. I've put together some charts below.
This first one simply plots the GB scores vs. clock speed. I've color-coded the points by processor.
Note, I've only included 64-bit GB scores. Also, I believe a number of people have reported GB scores for machines that have been OC'ed, but they only reported the base-clock speed. Where this was absolutely obvious, I omitted them from these charts. However, to avoid throwing the baby out with the bath-water, I was very lenient with the data that was there. You can see the resulting pile-up at the 3.4 GHz mark. I believe that more than a few of those systems are probably OC'ed, but their owners didn't state that when they posted their scores.
Note also that I lumped together the locked and unlocked versions of each chip. So, the i7-2600 and i7-2600K, for example, are considered together.
Interesting to see how linearly the GB scores are affected by increases in clock-speeds. in fact, from this data, I would say it is clear that for a given CPU, one can expect a GB score roughly equivalent to their clock speed time a CPU-specific multiplier:
i5-2500 : 3,000 GB/GHz
i7-2600 : 3,700 GB/GHz
i7-3770 : 4,100 GB/GHz (there are very few data points on this one)
i7-870 : 3,700 GB/GHz
i7-980 : 4,700 GB/GHz
So for example, a i7-2600K running at 4.5GHz should expect a GeekBench score of about 4.5 GHz x 3,700 GB/GHz or 16,650.
If your machine's GB/GHz rating is significantly above this please share your build recipe!. If your machine is significantly beneath the GB/GHz score shown above, I would have a look to see where you're leaving performance on the table. In particular, there is a i7-2600-based machine with a GB score of only about 6,000. Whoever owns this rig is essentially leaving half their investment unused. (though, this is probably the most extreme example).
Here's another chart showing the range of GB/GHz values for the more popular CPUs. This also includes only 64-bit GB scores. For each chip there is a max, min and average GB/GHz rating from the data collected. Some of these chips only have a single data-point, but the results are interesting anyway.
Note that the vertical axis is GeekBench score divided by clock speed (GB/GHz), not just the raw GB score.
Note, I've also included another metric on a secondary axis (on the right) which is the GB score per Watt. This is based on each chip's TDP (Thermal Design Power), and more accurately measures the amount of cooling required for the chip's operation, but serves as a nice (but not perfect) proxy of a chip's power requirements.
A couple of other interesting observations from the data collected:
I would like to say that building/tuning specifically for a GB score is probably not as important as I am suggesting in this post. I think it is always important to look at the common use of each machine and tune for that. In many cases, this may require investment/tuning for disk speeds or GPU capability, which are completely ignored by GB.
However, that being said, I think that by examining a machine performance, as measured by the GB score vs. GHz of clock speed and comparing to the figures above, one can identify issues in the build process. For example, not having correct kernel, kexts, etc. for the CPU/mobo. There is no obvious reason to not tune for these things when the investment is already made!
If there is interest, I may offer to setup a web-application to allow people to enter their own GB scores and perhaps other metrics (with more metrics about their systems, in fact), and update this dataset and also see how their system compares to others' and perhaps provide clues to why there are differences. I wish there was more data about RAM speeds, for example.
Finally, if Tony is interested and can offer up a OpenID service, then I could password protect the whole thing with the same passwords everyone already has here (without revealing to anyone–not even me–what that is). This would allow people to manage their metrics on an ongoing basis.
EDIT: 2012-07-17 Updated with a few more data points (both charts)
This first one simply plots the GB scores vs. clock speed. I've color-coded the points by processor.
Note, I've only included 64-bit GB scores. Also, I believe a number of people have reported GB scores for machines that have been OC'ed, but they only reported the base-clock speed. Where this was absolutely obvious, I omitted them from these charts. However, to avoid throwing the baby out with the bath-water, I was very lenient with the data that was there. You can see the resulting pile-up at the 3.4 GHz mark. I believe that more than a few of those systems are probably OC'ed, but their owners didn't state that when they posted their scores.
Note also that I lumped together the locked and unlocked versions of each chip. So, the i7-2600 and i7-2600K, for example, are considered together.
Interesting to see how linearly the GB scores are affected by increases in clock-speeds. in fact, from this data, I would say it is clear that for a given CPU, one can expect a GB score roughly equivalent to their clock speed time a CPU-specific multiplier:
i5-2500 : 3,000 GB/GHz
i7-2600 : 3,700 GB/GHz
i7-3770 : 4,100 GB/GHz (there are very few data points on this one)
i7-870 : 3,700 GB/GHz
i7-980 : 4,700 GB/GHz
So for example, a i7-2600K running at 4.5GHz should expect a GeekBench score of about 4.5 GHz x 3,700 GB/GHz or 16,650.
If your machine's GB/GHz rating is significantly above this please share your build recipe!. If your machine is significantly beneath the GB/GHz score shown above, I would have a look to see where you're leaving performance on the table. In particular, there is a i7-2600-based machine with a GB score of only about 6,000. Whoever owns this rig is essentially leaving half their investment unused. (though, this is probably the most extreme example).
Here's another chart showing the range of GB/GHz values for the more popular CPUs. This also includes only 64-bit GB scores. For each chip there is a max, min and average GB/GHz rating from the data collected. Some of these chips only have a single data-point, but the results are interesting anyway.
Note that the vertical axis is GeekBench score divided by clock speed (GB/GHz), not just the raw GB score.
Note, I've also included another metric on a secondary axis (on the right) which is the GB score per Watt. This is based on each chip's TDP (Thermal Design Power), and more accurately measures the amount of cooling required for the chip's operation, but serves as a nice (but not perfect) proxy of a chip's power requirements.
A couple of other interesting observations from the data collected:
- Most of you are rocking the i7-2600K (by a large margin), next most popular CPU would be the i5-2500K[/*:m:3v07ifgf]
- A LOT of people have spent $$$ for a unlocked cpu that they never OC
[/*:m:3v07ifgf] - The motherboard selection appears to make less of a difference than I would have thought in terms of GeekBench scores. In fact, I cannot see any obvious correlation of performance vs. motherboard at all.[/*:m:3v07ifgf]
- There are way too few data-points for the Ivy Bridge chips yet, but they look very promising and I would expect their numbers to jump noticeably once ML is upon us.[/*:m:3v07ifgf]
I would like to say that building/tuning specifically for a GB score is probably not as important as I am suggesting in this post. I think it is always important to look at the common use of each machine and tune for that. In many cases, this may require investment/tuning for disk speeds or GPU capability, which are completely ignored by GB.
However, that being said, I think that by examining a machine performance, as measured by the GB score vs. GHz of clock speed and comparing to the figures above, one can identify issues in the build process. For example, not having correct kernel, kexts, etc. for the CPU/mobo. There is no obvious reason to not tune for these things when the investment is already made!
If there is interest, I may offer to setup a web-application to allow people to enter their own GB scores and perhaps other metrics (with more metrics about their systems, in fact), and update this dataset and also see how their system compares to others' and perhaps provide clues to why there are differences. I wish there was more data about RAM speeds, for example.
Finally, if Tony is interested and can offer up a OpenID service, then I could password protect the whole thing with the same passwords everyone already has here (without revealing to anyone–not even me–what that is). This would allow people to manage their metrics on an ongoing basis.
EDIT: 2012-07-17 Updated with a few more data points (both charts)
