Core i7 4790 processor review

Processors 199 Page 5 of 22 Published by

teaser

The Haswell Architecture

An integrated GPU (continued)

So the graphics engine is DirectX 11.1 compatible. In total there are actually three variants of Haswell graphics segmented as GT1, GT2, and GT3 with the high end part. GT3 has 40 of the enhanced Ivy shaders. The desktop GPU core (GT2) is armed (depending on mode) with twenty EUs -- Execution Units (another synonym for unified processors like CUDA cores / Execution Units / Shader Cores / Stream processors). Clarkdale had eight of these, Sandy Bridge has twelve, Ivy Bridge has sixteen EUs. The EUs have been revamped and improved though, you'll see a performance increase of up-to 35% with the new IGP. Each Execution Unit can manage more work per clock opposed to the older ones as well. Now there are two different GPU configurations as Intel applied a dual-graphics core to segment their products, available with 1 core (Intel HD Graphics 2500) and 2 cores (Intel HD Graphics 4000). A single GPU core holds 10 EUs. So one last time, as you can understand from the specifications, don't expect heaps of gaming and fragging fun (though very simple games should be playable) but see this more as a desktop integration and implementation for very good Windows usage and, importantly, high-definition 1080P transcoding, decoding and acceleration.


Added Instruction Sets called HNI

Haswell's architecture supports four new extensions. These are tagged and bagged as 'Haswell New Instructions' (HNI) and include Advanced Vector Extensions 2 (AVX2), gather, bit manipulation, and FMA3 support). AVX2 widens integer SIMD (Single Instruction Multiple Data, a form of vectors) to 256-bit vectors, and that adds a gather instruction for irregular memory access. AVX stands for Advanced Vector Extensions. AVX is very well suited for floating-point-intensive applications and has become really popular. AVX provides features and instructions. AVX is a sophisticated form of SSE. Intel opened up the data path which has been expanded from 128 bits to 256 bits, the two-operand instruction limit is increased to three operands, and advanced data re-arrangement functions are included. Now, AVX doesn't just add new instructions, Intel states it allows the CPU to execute older instructions more quickly over the AVX extensions as well.

HNI includes Advanced Vector Extensions (AVX) 2 with support for SIMD (Single Instruction, Multiple Data) operations on 256-bit integer data types, bit manipulation instructions for improved handling of variable bitstreams, cryptography, compression and large number arithmetic, gather instructions for vectorising codes with non-adjacent data elements, any-to-any permutes with support for DWORD and QWORD granularity permutes across an entire 256-bit register, vector-vector shifts and floating-point multiply accumulate functionality for boosted floating point performance. What does that mean, well software that supports these instruction sets can take advantage of them and thus speeding them up, this should uphold significant enough performance benefits. Better vector handling for example could help out with much more enhanced parallel processing capabilities and that on its end would increase the generic compute performance by good amounts, especially in multi-threaded applications. So from image or video processing to face detection to database manipulation to generating hashes the new instruction can help out. There's one caveat though, you do need the software applications that are optimized for this.


Img_9384

Share this content
Twitter Facebook Reddit WhatsApp Email Print