Another year, another processor release from Intel. Since 2006, Intel has introduced a yearly refresh cycle they refer to as “tick-tock.” One year they release a “tock”, a major micro-architecture enhancement that brings significant changes. The next year they release a “tick”, which shrinks the design to a smaller process — bringing cooler operation, faster speeds and minor changes to perfect the architecture. Ivy Bridge, the 2012 release, was a “tick” of Sandy Bridge, brought down to 22nm. Now Haswell, the next tock, launches this week.
Haswell redefines the nature of Intel’s market from here on out. For years Intel has been marching towards better performance per watt and graphics without making those specs the most important features. Now everything changes. If you are looking for a laptop this year, make sure it has a Haswell processor. Here are the things Haswell brings to the table (I’ve already gone over the architecture in-depth, so this is more about the end-consumer features).
Impressive Power Management
Since the dark days of the Pentium 4′s terrible power consumption, Intel has been working to bring it under control. It’s been a steady march since then, but with Haswell, Intel’s processors make the most significant jump yet. Power, and by association heat output, is measured on a chip by its “thermal power design” or TDP, which is measured in watts. Processors can’t go over their TDP or they risk becoming unstable – the silicon shorts or literally melts. New technology allowed for dynamic clock speed, allowing chips to quickly slowdown to cool off or shut down cores while speeding up others (what we called Turbo Boost). Modern CPUs all have forms of power gating and clock gating to save power and be more dynamic.
Haswell brings the logic that controls those changes in power directly on to the chip. An integrated voltage regulator allows the chip to self manage all those independent parts that can be change their power state. Bringing any piece of logic on to the chip makes it faster, so the chip can be quicker to ramp up and down. Faster chips complete tasks faster and can then go to sleep quicker.
Intel also totes Haswell’s tremendous improvement in idle power consumption – up to 20x better. New power states allow the processor to effectively shut more of itself down during idle times. Because the voltage regulator is on chip, bringing those parts back online takes a trivial amount of time.
So what does it really mean? Much better battery life. A Haswell part can last up to 50% longer in situations like playing movies or light browsing – tasks where the data is sequential and not intensive. Even heavy duty tasks will see an appreciable boost in battery as the processor can more dynamically allocate power. Standby mode, Intel says, could keep a computer on for almost a month as opposed to a week today.
Drastically Better Graphics
Intel’s integrated graphics were a joke for so long. With Sandy Bridge, Intel delivered graphics that most consumers could depend on for video acceleration and keeping in-step with complex UIs. Ivy Bridge gave an appreciated 60% jump and made Intel’s graphics well suited for the majority of tasks (even a game in a pinch and low settings). With Haswell, Intel has taken aim at its competitors and wants to ship the only graphics you’ll ever need.
Intel’s integrated graphics on Haswell come in a variety of configurations, but at the most base level have three configurations: GT1, GT2, and GT3. The difference between the three are in the amount of “execution units” (EUs) each graphics unit has. Much like “compute units” from AMD or “streaming multiprocessors” from Nvidia, these are the basic processing groups that graphics processors use to compute data. GT1 has 6 EUs, GT2 has 20 and GT3 has 40. This is up from a maximum of 16 EUs for Ivy Bridge’s graphics.
Across the board, you will find significant boosts in performance from Ivy Bridge, but it will be important for you to know what you are picking. Intel HD Graphics refers to GT1, the most basic graphics unit. You’ll find this in ultra-budget solutions like Celeron and Pentiums. Ignore them. Intel HD Graphics 4×00 refers to any number of parts with GT2. These range from the lower-end 4200 to the more common 4600. Chances are you’ll most likely see the HD Graphics 4600 part, which is a solid jump in performance from Ivy Bridge, around 30%. That’s more than enough to handle HD video, high resolution displays and the occasional game.
The fun comes with GT3 and is the pièce de résistance of Haswell. The HD Graphics 5000 is what you will see in every ultra-low voltage model, the chips that fit inside ultrabooks and Macbook Airs. With a fully enabled part, the graphics unit is larger than the general purpose cores and goes to show how much attention Intel gave to graphics this generation. You can expect at least 50% better performance than the last generation if not more, and with a lower TDP. Intel wants to remove any lament an ultrabook user would have over not buying a discrete graphics card, so they’ve placed the best where it’s needed.
Unless you go bigger. Because Intel did. Like AMD’s Radeon and Nvidia’s GeForce, Intel has named their best graphics “Iris” and “Iris Pro.” Iris is a GT3 part, but supports faster clockspeeds and better memory bandwidth. You can expect twice the performance from Iris as from Ivy Bridge, and very close to that of a mainstream discrete graphics card. Iris will be found in Intel’s mid-voltage models with a TDP of 28W. It’s designed for laptops that are not ultrabooks, but still have a premium on space and power. Think the 13-inch Retina Macbook Pro.
Iris Pro comes on a special sub-set of quad-core mobile parts. Their overall clockspeed is slower than quad core chips with HD 4600, but their graphics are tremendously improved. Iris Pro uses GT3e, the “e” standing for the 128MB embedded DRAM die located on the same package as the chip. This memory pool acts as an “interposer” – a type of cache that is very fast and very low power. It allows the Iris Pro to have memory bandwidth approaching that of a discrete graphics card (memory bandwidth is crucial for graphics). The eDRAM is also a general L4 cache for the processor, meaning the CPU itself will see better performance. Intel says that Iris Pro is as powerful as Nvidia’s high-end mainstream part: the GT 650M, which powers the 15-inch Retina Macbook Pro. While benchmarks have found that Iris still trails in real-world results by around 15%, being able to talk about Intel’s graphics and Nvidia’s high end mobile part in the same sentence is pretty astonishing. Iris Pro is more than enough for video acceleration, moderate video editing, and even mainstream gaming. Haswell chips equipped with Iris Pro make excellent “in-between” options for OEMs that don’t want to rely on a discrete GPU but want to drive high resolution displays (it’d make a terrific entry part for a 15-inch Macbook Pro or high-end 13-inch).
Most importantly, it demolishes any competitive advantage AMD had with its Trinity parts. While their Fusion APUs incorporated a Radeon GPU, Intel’s graphics solution combined with their general computing advantage makes Haswell better than AMD from every perspective…. except price.
Better General Performance?
With all this talk about graphics and power management, what about the processor’s general computing performance? Here, things are less evolutionary. Intel isn’t able to reinvent the wheel here and after the major technological jumps of Nehalem and Sandy Bridge, they’re grasping at straws (was that enough idioms in one sentence?).
Haswell widens the execution engine (think the business end of the processor) to 8 ports from 6. These are like lanes in a highway, but where each lane is for specialized traffic. Depending on the operation commanded by the instruction (or in this case chunks called micro-ops but we’re getting technical), they’ll enter one of the execution ports where the proper hardware will execute on the data. Haswell’s wider execution engine means that there is more hardware to keep more instructions “in-flight” at once and thus complete more at once. However, these are very low level and specialized parts of the chip, so it does not translate directly to 25% more performance.
Haswell also supports something called Fused Multiply-Add, a complex mathematical instruction that allows a processor to, as you might imagine, multiple and add floating point values at once (floating points are basically really big numbers that are approximated by a computer using a “precision level” of either 32 or 64-bits). FMA is a big feature in other processors that increase floating point performance significantly. There’s also a new form of Intel’s vector technology called AVX2 (vector processing does a single action on large amounts of small data – we call this single instruction, multiple data or SIMD).
In all, Haswell is good for around 15-20% more performance per clock than Ivy Bridge. It’s not a major jump but taken with the impressive power improvements and graphics jump, Haswell amounts to a significant change.
Haswell marks a major change for Intel. The company that was once known for pure CPUs is now a fully integrated platform manufacturer. Now more and more of that platform is located on one tiny silicon die. Haswell brings awesome graphics and great performance per watt to the masses. With Haswell, Intel can build a chip that fits into every need. From the thinnest and lightest ultrabooks to the most powerful desktop computers, Intel has it locked up.
If you are buying a computer in 2013, make sure it has Haswell. You won’t be disappointed.