Sandy Bridge New Micro-architecture by Intel

[JCAddy]

Compared to a current notebook, equipped with Nvidia Geforce 310M that was - that a small-GPU - was Sandy Bridge at the same speed and completely liquid. To recall the future processors Sandy Bridge will be engraved in 32nm and will use a new socket: LGA 1155. These CPUs should not be compatible with existing cards using the LGA 1156. Benefit Intel to launch a new series of chipsets with a first step, according to the latest rumors, P67, H67 and Q67. For now, they still do not have a USB 3.0 controller. The motherboard manufacturers will therefore again obliged to go through additional chips. The S-ATA 6 Gbps will by cons of the game, but only two ports while the chipset supports a total of six (the others will therefore only S-ATA 3 Gbps). To accompany the launch, Asus has seen the big picture in terms of functionality.

Possible by the Sandy Bridge architecture to communicate with the CPU and GPU cores together on the L3 cache. The cores communicate via a sort of ring bus together. The requests from CPU and GPU units should not get in your way. In addition, all calculators use Intel's Turbo Boost. Previously it was given for the GPU only for some mobile models.

Indeed, all maps will have the manufacturer, serial, BIOS UEFI (Asus calls always oddly EFI). In addition, some of them will receive a digital power stage Digi +, torque TPU / EPU and a Bluetooth module. Attention, it was reported initially that all models involved were sold by these three points, but we have since learned that this would of course not the case for entry-level models that hide among the fifteen references to be proposed by Asus. We therefore decided to analyze in detail the features that are offered, along with four models that have been presented more thoroughly: the P8P67 Pro Deluxe, the Sabertooth, and finally P67 Maximus Extreme IV. The entry will consist of Core chips i3 3.1 and 3.3 GHz, 3 MB L3 cache, and four logical cores (two physical cores and hyperthreading). I5 line will also have four hearts, four physical cores, without Hyperthreading. The frequencies are similar, but rises to 6 MB cache finally, there will be a model for Core i7, peaking at 3.2 GHz, but it will have four physical cores and Hyperthreading, with 8 MB of cache.

We know that Sandy Bridge provides a set of instructions called Advanced Vector Extension whose objective is to expedite calculations audio / video, 3D and scientific (read: Sandy Bridge will reach the end of the year ), and it integrates a management system of its consumption, but it is not known yet what will be the thermal envelope of these processors. Intel promised it will be much more powerful than the current one: This could perhaps allow Apple to drop the Core 2 Duo on the Mac mini and MacBook, which it was obliged to keep to use an NVIDIA chipset and does not degrade the graphics performance of its entry-level models.

Sandy Bridge on Intel

Already present in the first Core i7 i5 and the graphics chip (12 calculation units, DirectX 10) is now fully integrated into the die of the processor. This proximity allows us to improve the efficient exchange of information between CPU and GPU, through a unified cache and organized in a ring ("ring"). The performances are thereby logically increased and now enjoy the famous TurboBoost introduces the architecture of the first Nehalem Core i. In practice, we found the fluidity of a game like Starcraft 2, while the processor is instructed to encode a video of the current action.

Under current plans are 13 processors for the desktop segment on the program expanded to include six new notebook CPUs for the high-end segment. They are:

• Core i3-2120
• Core i3-2100
• Core i3-2100T
• Core i5-2500
• Core i5-2500S
• Core i5-2500T
• Core i5-2400
• Core i5-2400S
• Core i5-2390T
• Core i7-2600K
• Core i7-2600
• Core i7-2600S
• Core i5-2500K

[JCAddy]

Overclocking

Sandy Bridge opens therefore a new version of turboboost. It is now a matter of overclocking (or "overclocking") of the total processor cores, even when they are used. Thus, provided that the processor is not too warm, the incidence of heart may temporarily increase when a software application of power, even if the TDP (thermal dissipation maximum value of the processor) is exceeded. Turboboost this may also apply on the graphics processor if the benefits do not exist. In other words, it becomes more effective in maximizing the chip.

USB 3.0 Support

Like previous versions, the new Intel-based Sandy Bridge incorporate DDR3 memory controllers and PCI Express. There are also plans to support DisplayPort and HDMI 1.4 (we return to this issue in later news on the 3D video). Alas, the USB 3.0 is the biggest loser in history, since there is still no way to integrate it natively in the Sandy Bridge architecture. Finally, a new instruction set "AVX" just accelerate the processing operations of the image, which seems indeed very useful if one believes that the demonstration has been made.

Intel has everything it takes to stay number one in terms of performance and efficiency. Do not look forward too quickly, however, since the prices should be matched with the level of performance achieved by these new processors. Arrival early 2011 for an immediate move (hopefully) in the hands of Regis, our specialist processors.

Socket

With the Clarkdale-models, also known as Core i3 i5 5xx and 6xx occurred Core, Intel earlier this year added the first step of a complete integration of a graphics accelerator in the processor. In the Clarkdale offshoots while the iGPU sits together with the memory controller or separate. Communication with the, which contains the processor cores and the cache is, by means of the Quick Path Interconnect (QPI). These indeed both sit on the same base material but because the 45-nm manufacturing of the GPU this is that much larger than the "CPU-die".

With the "Sandy Bridge" models for Socket 1155 Intel, the graphics unit is now fully integrated into the processor, i.e. there is only one Die, which both the processor cores, memory controllers, caches, and the graphics accelerator includes. This caused there is some restructuring, we will present now and finally make the change from the current Socket 1156 new socket 1155 is necessary. Unlike for the base 1156, which is the Lynnfield processors models without graphics accelerators are also, it is the base in 1155 probably only branch with iGPU provide for, that even the fastest quad-core processors, then a graphics unit have to be built that.

As with the current Socket 1156 processors also have the latest models for the base 1155 a built-in PCI-Express 2.0 interface with 16 lines for the connection of discrete graphics cards. The 16 lines can be optionally completely as well as in a subdivision of 8 / 8 are used. The use of graphics card is thus drawn carriages and with the new processors are no obstacles. Also back of the party is the Direct Media Interface (DMI) for the connection of the chipset. However, Intel did not reveal whether it will continue to exist in a version of four PCI-Express 2.0 lines, or whether this increased this to the benefit of the bandwidth slightly. The memory controller that provides two channels of DDR3 memory is more or less from the current processors. Some internal optimizations will have it but certainly given here, after all, intended to DDR3 memory clock speeds to be supported in 1600 by the house.

[JCAddy]

Cache

The UOP cache is a new component of the decoding of the already decoded UOP commands (1.5 K) stores. Fetches the instruction fetch (Instruction Fetch) a new command, it first checks the cache to UOP any matches. If this is the case is passed directly to the instruction code corresponding to the UOP other pipeline stages and the complexity of decoding can be disabled. The UOP cache is therefore an L0 cache, if you will. Only when the command is not found in the UOP-cache, it is of the actual - still 32 KB large - L1 instruction cache (includes only non-decoded x86 instructions) will be removed. According to Intel, the latter situation occurs, but only in 20 percent of cases. Thus helping the UOP cache not only the command processing in the front end to speed up significantly, but he also contributes also to reduce energy consumption.

Unlike already launched with the Core 2 imported Loop Stream Detector (LSD) for loop detection, the UOP cache can in the case of commands to be used outside of loops and is thus much more effective than the 28 UOP comprehensive LSD, which only in the case of multiple-loop decoding can handle a command. A little of the new UOP-cache is also reminiscent of the trace cache from the Pentium 4 However, this could not save UOPs independent, so that the UOP cache compared to the trace cache also has a clear advantage.

Intel has also revised the L3 cache. This is now called the last-level cache because the graphics on this unit - can access - via the ring bus. Unlike the Nehalem architecture, but the LLC is no longer in Uncore area but in the core area. Thus, the LLC will operate at the same speed as the cores, which just leads to the fastest spin-offs to a much higher bandwidth between LLC and nuclei.

In addition, Intel has in the LLC blocks - so-called Slices - divided. Each core in this case has exactly one assigned LLC block. This block and the core have the same starting point in the ring bus and thus can replace up to 32 bytes per clock cycle. Each core, however, can also use the other LLC-blocks, but needs to go via the ring bus in the purchase, resulting in higher latencies. Initial tests have shown that Intel has reduced over the Nehalem offshoots the latencies again. Instead they are now apparently 36 cycles at 26 ° to 31 cycles depending on the access point.

Sandy Bridge for Mac

If the adoption of the future platform Sandy Bridge Intel already did and no doubt was also not surprising in terms of CPU, the question still hung about the choice of GPU. The successors of the white MacBook, MacBook Pro 13-inch MacBook Air and would thus benefit from a leap forward in terms of CPU, skipping an architecture, at the cost of stagnation or a decline in terms of multimedia. Improved support for OpenCL however the case that a software update.

The entry of Apple was sticking up for this older Core 2 Duo processors because of a dispute between Intel and NVIDIA, preventing the latter to design an IGP (non-dedicated graphics chip) for the current processor i. Core Both companies have recently found an agreement but it has not been revealed and NVIDIA has fallen behind.

Conclusion :

Concrete benchmarks laid Intel not yet available - except by pointing out that the 25 times faster graphics on the synthetic 3DMark06 benchmark relates. There were also more demos to other parts of the new architecture. Steps should help together with the CPU and GPU transcoding of videos. The Media Show, Intel ago with espresso. This program is already well adapted to CUDA and Stream, and now apparently also on Intel's GPU cores. Intel introduced it only with a clip of what espresso is not fair: It can convert several videos in almost the same speed on GPUs. Unlike AMD's new with the Bulldozer architecture changes in Intel's "Sandy Bridge" over its predecessors are to be seen rather as a gentle evolution rather than revolution. A revolutionary design - such as the module concept of AMD's Bulldozer - is not there. However, there are many new features that may constitute "Sandy Bridge" after its launch in early 2011 a successful basis for a variety of processors for servers and desktop and mobile computers.

Sandy Bridge unlike most of the code name for Intel - does not name a place or a geographical event but a metaphor. "Sandy Bridge" in bridging the gap between the "Pentium Pro" - and "Pentium 4" design, and Intel graphics accelerator architecture. Sandy Bridge is to have more revolutionary than it might at first glance looks like yet less revolutionary than many may have hoped. This applies less to the pure architecture of the processor cores but rather for the establishment of the entire processor with ring bus, CPU and GPU shared cache and an integrated graphics unit.

Last but not least is the verdict of Sandy Bridge but also largely on performance in everyday applications and not on the number of changes over the previous generation. First benchmarks certify "Sandy Bridge" while consistently better performance over the current generation, the results still vary greatly one. It appears, however, such as above, that give the LLC, which operates at core clock, and just the top models can a decent performance boost. The range of Sandy Bridge CPUs currently comprises 13 models for the desktop, including k-CPUs, but also Xeons with only 20 watts. An allegedly appearing Core i7-990x (Westmere on current basis) to remain the top model Q3/2011 until he e-Bridge-with up to eight cores replace Sandy.