Our testing followed the usual line of Synthetic and Real-World testing to cover both sides of the performance curve. For overclocking* we ran both a single “suicide shot" [the highest clock we could get a validation from CPU-Z] and a working overclock. The working overclock is the highest stable clock speed we were able to run all of our tests in series [no reboots] and get valid scores. The working clock is a better indication of OC headroom than a simple suicide screen.
* Our results may differ from yours and are indicative of potential only. Since we are using an engineering sample CPU you may have better or worse results from an OEM or retail product. As with all overclocking the combination of hardware used and operating system can affect your maximum and stable OC.
With all that out of the way let’s dive into performance.
Our synthetic testing is the most easily reproducible testing as it is simply running a fixedcode on the hardware. This does not make it invalid but bear in mind that this is only a half of the picture. At BSN*, we strongly believe that using synthetic benchmarks is a standard way of testing equipment but it cannot be the only one. Just like automotive publications test the cars on a racetrack and real roads, we believe and we will insist on using both worlds in order to obtain complete picture of a product, regardless of what market that product is targeting.
Of course no evaluation would be complete without throwing in the Futuremark numbers. We use both 3DMark Vantage and PCMark Vantage [in both x86 and x64 versions, where available] to cover all the possible bases there.
In 3DMark Vantage we were looking to see how well the CPU does with handling AI and Physics. We ran the test with GPU enabled Physics on and off to show the difference between the two CPUs with the system and the CPUs on their own.
The i7 shows off some impressive CPU power in 3DMark Vantage
Looking at the CPU scores here, they are simply staggering even without the nVidia PhysX the i7 is impressive. The 975 grants you an extra 1,000 CPU points over the 965 at stock speeds. Given the increase in default clock from 3.2 GHz to 3.33 GHz, we did not expect higher or smaller result. But it is more than obvious that the Turbo mode, present in all Core i7 processors is working hard on making those extra CPU cycles work.
PCMark Vantage is a complete suite of tests that is designed to cover all of the uses a system would normally be put to. There are content creation tests, gaming, etc. We use it as a base line for comparison but it is not meant to be definitive.
PCMark Vantage shows us a few interesting scores
Again we see a small performance lead for the 975 which is about what you would expect. What is interesting is the performance loss when running the 64-bit version. Here the 965 actually performs better than the 975 for some reason. Once a healthy overclock is running on the 975, the tables do certainly turn in favor of the 975 once again.
Sisoft Sandra 2009 SP3
Sandra is another synthetic that yields an approximation of performance at the component level. Here we can break down the CPU by specific tasks (arithmetic, multimedia, and multi-core efficiency) and give you an idea of how each should perform. We like SiSoft Sandra due to the fact that it also features GPGPU e.g. GPU computing tests, and the benchmark itself is very light on the system in terms of installation and overall size.
As you would expect the 975 performs better than the 965 in almost every test, no questions asked. The only discrepency that was noted was the memory bandwidth test where the overclocked 975 did worse than the stock one.
Everest Ultimate 5.01
Everest was used in this case to show differences in available bandwidth to each CPU and again when the 975 XE was overclocked. But yeah, brace for impact - since we encountered issues when it comes to the performance part. See the picture below:
Again we see the 975 performs better, that is until we overclock the system. For some reason [and despite multiple runs] Everest would keep showing the 975 as significantly slower than the stock 975 and the 965. It read the memory bus at 3.8MHz and the BClock at 0.9, the CPU itself read a whopping 25.4MHz. Yes all of those read MHz. The performance scores of overclocked CPU certainly made us think of "Back to the Future" trilogy, since these numbers were last seen around 20 years ago, with 486 processors. Yes, even the allmighty Pentium scored much higher than these clocks... we think that Tamas and his Lavalys team have serious work to do in order to put the support for i7 975.
HyperPi allows you to run an instance of SuperPi on each CPU core, this is a good test of system stability as well as the speed of the Memory to CPU to HDD speed of a system.
HyperPi is another one where the System seemed to have a problem with the memory bus when overclocked. I am not certain why as of this writing but it looks like there is a bottleneck [or oversaturation] of the memory bus. It is possible that this is a mainboard issue and not a CPU issue but I have not been able to pin this one down yet. After Everest, this was the second application to experience some issues with the 975 processor. We cannot attribute this to the "engineering sample" marking on the CPU itself, but it looks like software vendors will require some work to get things up to speed. Literally.
Cinebench R10 x64:
Cinebench is from Maxon, the makers of Cinema 4D and is a synthetic rendering test. It also allows you to see the difference between rendering on one CPU core Vs Multiple CPU cores.
While we saw memory performance issues with Sandra, Everest and HyperPi there was no problems with Cinebench. The i7 975 tacks on extra 1,000+ points to the multiple CPU score and a very nice 5,000 when overclocked. The single CPU scores were closer but still showed a performance improvement.
Continued on the next page: Real-world testing, Overclocking, Value.
© 2009 - 2013 Bright Side Of News*, All rights reserved.