Graphics And Reduced Power Highlight Intel's 4th Generation Architecture

Graphics And Reduced Power Highlight Intel's 4th Generation Architecture

Increased Graphics Performance And Reduced Power Highlight Intel's 4th Generation Haswell Architecture

As usual, Intel's Haswell has been one of the worst kept secrets around although Intel has done a pretty good job of limiting the details until now. This is Intel's fourth generation Core processor and comes on the tail of Intel's Ivy Bridge (see Understanding The Ivy Bridge Architecture) based on the 22nm 3D FINFET tri-gate transistor architecture (see Moore's Law Continues With 22nm 3D Transistors). Haswell also utilizes this 22 nm technology (Fig. 1). The tock in Intel's tick-tock strategy will address the next design shrink.

Figure 1. Intel's Haswell processors utilize the same 22 nm 3D FINFET technology as the earlier Ivy Bridge.

Haswell will probably overshadow AMD's latest APU announcements (see X-Series APU Targets Microservers) that also delivered combined graphic and power improvements. It will take a little while to shake out the improvements but both are currently available on third party products. It is clear that the new chips have spawned a plethora of new hardware platforms that have an emphasis on low power and high performance especially where GPUs are concerned. Enhancements across the board are part of these latest announcements but GPU number crunching is the most major change this time around.

The basic chip design is 1.4 billion transistors. The 177 mm2 die (Fig. 2) is divided into four hyper-threaded processor cores with a 14-stage pipeline and a GPU connected via a high speed loop with a shared L3 cache. The cores each have a 32/32 Kbyte L1 cache and a 256 Kbyte L2 cache. The chips have dual DDR3 memory controllers and PCI Express links on-chip. The GPU drives Display Port links.

Figure 2. The 1.4 billion transistor, 177 mm2 die hosts four hyperthreaded cores and a GPU with Display Port outputs along with dual DDR3 memory controllers and PCI Express Gen 3 ports.

Improved Security and Power Reduction

This new Intel Core processor family features a U-series processor with 15 W TDP (thermal design power). The family encompasses Xeon server products that have as little as a 45 W TDP. They also incorporate the GPU support found in the Core i5 and i7 chips. Likewise, the top end Core products include ECC support. ECC support can be found in the Intel C226, Q87 Express and Mobile HM86 Express chipsets. The first two also support Intel's vPro technology.

All the new platforms take a two chip approach but with all the CPU, GPU and memory controllers in the main processor (Fig. 3). The dual DDR3 memory controllers have ECC support on select chips but all incorporate a fully integrated voltage regulator that reduces the system bill-of-materials (BOM) while reducing board complexity.

Figure 3. Haswell utilizes a two chip solution with PCI Express and Display Port support in the host processor along with the GPU. Dual DDR3 memory controllers and an on-chip voltage regulator round out the processor chip.

It also provides improved power management that complements the other power management enhancements in Haswell. It runs at 125 MHz which is faster than most off-chip regulators. It can also deliver more current that improves wake-up speed and efficiency.

The on-chip power optimizer has control over all the peripherals as well as the processor. This allows finer control especially during power down and power up sequences. It knows how long these operations take so slower devices can be started before faster ones such that the entire system is fully started at the same time. Of course, these requires operating system coordination too hence new software support will be waiting in the win.

There are a number of chipsets that support these new processor chips. All provide USB 3.0 and 6 Gbit/s SATA support along with x1 Gen 2.0 PCI Express. These interfaces have flexibility in terms of pin and interface allocation to minimize the number of chip configurations while allowing designers to chose their own combination of PCIe, SATA and USB.

There will be a one-chip BGA support as well that combines the processor and chipset into a single package.

The processor includes new AES encryption instructions that provide better performance and Intel's McAfee is now showing up more often. The McAfee Enpoint Encryption takes advantage of these instructions. Intel's Boot Guard and BIOS Guard are designed to prevent rootkits and other malware and are the base for Intel's Active Management Technology that in turn ties into the McAfee ePO Deep Command support. This provides secure remote management capability. Intel also uses TPM (Trusted Platform Module) support for secure boot support.

The chipsets that support Intel's vPro take remote management further with remote KVM support. The ability to view the video display remotely is further enhanced by management overlay support. This allows the management interface to be viewed at the same time as the GPU's video display. Intel's vPro Powershell Module for Microsoft Windows Powershell provides additional management customization capabilities.

Haswell provides significantly finer grain power management control over all aspects of the system and its peripherals. The computing performance of the processor has seen only a minor bump up with a major bump up on the GPU side but Intel has been able to significantly cut the power requirements. Mobile platforms should see a 50% battery life improvement with a double or triple graphics performance improvement.

CPU/GPU Improvements

The new AES instructions are just the tip of the iceberg. One major change is the move to AVX 2.0. AVX was a major improvement (see Intel's AVX Scales To 1024 bit Vector Math) that added vector instructions that were significantly more than the SSE support found in earlier Intel architectures. It scales to 1024 bits but the current incarnation is 256-bits. AVX 2.0 rounds out this support bumping up cache sizes and provide 2 load/1 store operations per clock. It adds new instructions including gather support and improved bit manipulation. This iteration also supports FMA3 (fused multiply/add). FMA3 was available earlier in AMD's Piledriver and Bulldozer architectures.

AVX 2.0 is having a significant impact in a number of application areas including mili/areo. FFT speed improvements of 1.5x to 2x are common and other algorithms are showing DSP speed improvements on the order of 30%.

Haswell also adds support for the Transactional Synchronization Extensions (TSX). These add support for hardware locks and restricted transactional memory.

The GPU actually comes in different versions depending upon system requirements. These have been designated GT1, GT2, GT3 and GT3e. The top end has 40 execution units (EU). The Ivy Bridge family had a GPU with a maximum of 16 EUs.

The GPU supports Direct3D 11.1 and OpenGL 4.0. It supports OpenCL 1.1 for computational chores.

Long Term Support

The Haswell family encompasses a large number of SKUs but there are eight that have been released under Intel's Intelligent Systems Group long term support program (Fig. 4). These chips will be available for at least 7 years.

Figure 4. Intel has released eight Haswell products with long term support necessary for many embedded applications.

This announcement includes the mobile 47-W, 2.4 GHz Core i7-4700EQ with ECC support. It can be combined with a number of chipsets that include ECC support and vPro technology. There are also three Xeon chips that all include GPU support and 8 Mbyte L3 caches.

Long term support is critical for many embedded applications such as mil/aero where design times and system lifetimes are longer. Not all of the Haswell chips are part of this program so designers will have to choose carefully. Still the list covers a wide range of solutions from mobile to servers.

New Hardware And Software

The real key to Intel's Haswell release date is the plethora of third party companies with boards and modules using Haswell as well as software vendors that have tuned their products to take advantage of Haswell.

Not surprisingly, Intel's Wind River has expanded is VxWorks and Wind River Linux and Android offerings to take advantage of Haswell. The same is true for LynuxWorks that will have versions of its LynxOS, LynxOS-178 and LynxSecure (see Embedded Hypervisor Delivers Separation Kernel) available to the new chips. Kozio will have a version of their Verification and Test OS (VTOS) and Real-Time Systems will have their hypervisor running on Haswell.

An example of some of the boards available with Haswell chips includes Curtiss-Wright Controls 6U OpenVPX CHAMP-AV9 (Fig. 5). It is their flagship HPEC board running a pair of 2.6 GHz Intel Core i7-4700EQ Haswell chips. It also has four 40 Gbit/s Mellonax ConnectX3 interfaces with Ethernet and InfiniBand support.

Figure 5. The 6U OpenVPX CHAMP-AV9 board runs a pair of 2.6 GHz Intel Core i7-4700EQ Haswell processors. It also has four 40 Gbit/s Mellonax ConnectX3 interfaces with Ethernet and InfiniBand support.

The board is available in rugged 71°C air and 85°C conduction cooled versions. Because of Haswell they require no or minimum CPU throttling to meet these cooling requirements.

Two of the four Display Port interfaces are brought out via the backplane. The others are on the front panel. PCI Express support includes four 8x and two 16x ports.

The boards have 8 Gbytes of flash memory and up to 16 Gbytes of DDR3 DRAM. There are two XMC/ PMC mezzanine module slots as well. A VxWorks BSP and Linux LSP are shipping now. Also available is Curtiss-Wright’s Inter-processor communications (IPC), OFED/MPI, Continuum Vector Intel AVX 2.0-optimized signal processing library and Continuum Insights Graphical Management and Debug tool.

Haswell will have a major impact on the systems being deployed over the next couple years. It addresses everything from mobile devices to rugged OpenVPX systems.

Hide comments


  • Allowed HTML tags: <em> <strong> <blockquote> <br> <p>

Plain text

  • No HTML tags allowed.
  • Web page addresses and e-mail addresses turn into links automatically.
  • Lines and paragraphs break automatically.