Versione 2017 del popolare tool diagnostico e di benchmarking del proprio sistema. Qui di seguito le principali novità introdotte:
* Support for Windows 10* Tech Preview
- Updated to the latest tools & libraries for forthcoming Windows 10, with various minor compatibility changes and support for new APIs, technologies as well as hardware.
- Will support DirectX 12.
- Support for very-wide SIMD AVX-512 family.
- Updated resources for high-DPI (retina+) up to 200% 192ppi or higher - e.g. huge 256x256 icons.
* Updated System benchmark: Overall Score 2015
- Removed .Net/Java Multi-Media (vectorised) benchmark as current JVM/CLR do not vectorise even today.
- Added CPU Scientific (GEMM, FFT, N-Body), CPU Financial (Binomial, Monte Carlo, Black-Scholes) Analysis - while GPGPUs do accelerate these complex algorithms, wide-SIMD (e.g. AVX-512F) can significantly accelerate them in large CPUs with many cores and large caches (e.g. 12C+ 24MB L3/128MB L4)
- Added GP Scientific Analysis (GEMM, FFT, N-Body) (using CUDA, OpenCL, DirectX Compute) - to match CPU Scientific
- Updated weights to match modern devices (mobile, laptop, desktop and server)
* Updated Crytography CPU Benchmark
- Optimised AES HWA code-path for improved performance across hardware constrained by memory bandwidth.
* Optimised GP Benchmarks
- GP Cryptography - optimised AES code-path.
- GP Scientific - optimised N-Body path.
- GP Financial - optimised Binomial Options path.
* Updated CPU Multi-Media (SIMD) Benchmark
- Combined score is now made of "combined integer (int32 and int64)" and "combined floating-point (single and double)" SIMD performance. Sandra 2014 used "int32 single-float and double-float" only.
- Combined integer averages int32 (32-bit integer) and int64 (64-bit integer) SIMD performance (SSE, SSE2, AVX, AVX2) - and thus measure also 64-bit integer performance that is extensively used today.
- int128 (quad-integer emulated) and quad-precision float-point (emulated, mantissa precision extended) is also measured but not included in the index as both are emulated at this time. Once native SIMD support is available in hardware, they will be added to the index.
* Updated Memory Bandwidth Benchmark
- Now reports average (rms - root mean sqare) not maximum performance. Some scores *will* be lower depending on the hardware, usually 3-5%. You can use the "maximum value" as index if you want to match Sandra 2014 and earlier.
* Updated Cache & Memory Bandwidth Benchmark
- Now reports average (rms - root mean sqare) not maximum performance. Some scores *will* be lower depending on the hardware, usually 5-20%. You can use the "maximum value" as index if you want to match Sandra 2014 and earlier.