Synthetic Benching Suites

SiSoft Sandra 2014

For all of our testing including gaming, we used our 240GB Kingston Hyper-X SSD.  Although there are no average framerate differences running games on a SSD or HDD, we wanted consistency and identical conditions for all of our tests.  We also limited our game benchmark suite to ten games; our eight very latest games and two older games that we also benchmarked on Core i7-920.  Before we get to gaming, we want to see exactly where the Haswell and Ivy Bridge architectures differ, and there is no better tool than SiSoft’s Sandra 2014.

There are several versions of Sandra 2014, including a free version of Sandra Lite that anyone can download and use.  It is highly recommended!

SiSoft’s Sandra 2014 is the very latest and it was released November 25, 2013.  It has quite a few improvements over Sandra 2013:

* New CPU benchmark: Scientific Analysis

– 3 common algorithms: GEMM (matrix multiply), FFT (fast Fourier transform), N-Body simulation

– 2 precision: single/float, double/FP64

– SSE2, SSE3, AVX, FMA support where possible

– same workload as GP scientific analysis benchmark – to directly compare CPU vs. GPGPU performance.

* New GP benchmark: Scientific Analysis

– 3 common algorithms: GEMM (matrix multiply), FFT (fast Fourier transform), N-Body simulation

– 2 precision: single/float, double/FP64

– 2 interfaces: CUDA, OpenCL (with DirectX ComputeShader to come)

– not using any vendor specific library like cuBLAS, cuFFT, MKL, etc.

– same workload as CPU scientific analysis benchmark – to directly compare CPU vs. GPGPU performance.

* Updated System benchmark: Overall Score

– Replaced GP crypto with financial analysis (now that most CPUs have crypto HWA support there is no point doing crypto on GPGPU)

– Updated weights to match modern devices (mobile, laptop, desktop and server)

* Updated CPU benchmark: Arithmetic

– New libraries for Whetstone SSE3 / SSE4.1 / AVX

– 2 precision: single/float, double/FP64

– float score is geometric mean of both float and double (was just double before)

* Updated cache/memory benchmark: latency

– measure L0 cache latency

– data and code caches

* Updated cache/memory benchmark: bandwidth

– measure L0 cache bandwidth

First we run the very latest Sandra benchmark suite with the EVGA FTW Z77 motherboard and Core i7-3770K. Our charts comparing the individual benchmarks are in the performance summary as well as here for easy cross comparison. Here are the Whetstone and Dhrystone results: 

Here are addt cryptographic results:

Now it’s i7-4770K/Haswell’s turn at the same settings:    Here are the Dhrystone and whetstone benchmarks

 here is the cryptographic benchmarl

Here is a comparison chart including some of the Sandra highlights comparing Ivy and Haswell using the same settings, and also including Bloomfield’s i7-920 results from 2012:

In every case, the new architecture shines over the older ones.

PC Mark Vantage

PC Mark Vantage was developed primarily for Vista and it has been superseded by PC Mark 7.  We are using the Advanced Edition of PCMark Vantage.  As with all of our tests, we use 2x4GB Kingston HyperX Predator RAM.  The only differences are that we run at 2800MHz on Haswell, but 2400MHz on Ivy Bridge.

Here are the 4770K’s results.

There is a noticeable difference between the scores with Core i7-4770K winning in every metric over Core i7-3770K although settings are identical and both processors and the system memory are overclocked similarly. The newer CPU architecture and platform wins.

Final Scores: 23033 Ivy Bridge to 27342 with Haswell. Bloomfield’s i7-920 scored 18867 as a comparison.

We will keep our settings and hardware the same and now run the detailed tests of the basic version of PCMark 7

PCMark 7

We are testing at the same default settings that we tested with PC Mark Vantage. Even the free basic version has some very detailed test results.

First up is the i7-3770K.

And now the Haswell i7-4770K

We notice that the IPC of Haswell is higher than for Ivy Bridge. The synthetic tests clearly show that the newer architecture is faster.

The results: We scored  6300 with Ivy compared with 6886 with Haswell,but only 5117 with Bloomfield. Will this hold true for gaming?

Before we look at gaming differences, we need to look at more benchmarks including real world applications that many of us use daily, including calculation, file copying, SSD testing, encoding video, multi-tasking, image editing and gaming.