As expected, firm details of Intel’s new Nehalem processor and a sneak peak of its upcoming Larrabee processor dominated the three days of announcements from Intel’s Developer Forum (IDF) held in the US recently. net.work went to San Francisco for the event.
Ever since chipmakers like Intel and AMD began edging up on the physical limits of what clock speeds their processors could run at, the semiconductor space became more interesting.
Instead of being judged on the clock speed of their silicon, vendors are getting judged by the number of features they can cram into their new offerings, to what degree architectural changes can eke extra performance out of a chip and lastly, how many cores (distinct processing units) they can cram onto a single die.
Although AMD contends its architecture will allow it to take the performance crown back in the coming months, Intel remains the leader for the time being.
With the availability of its new processor line, Nehalem, and the upcoming Larrabee platform that it says is set to revolutionise the way we think about conventional computers, AMD might have a tougher time than it anticipates.
Just plain faster
Nehalem is the code name for the new family of processors Intel showed off at IDF.
Officially dubbed Core i7, the processor family is based on the 45-nanometre manufacturing process that its Penryn processors were built on, but features an evolved Core architecture.
While clever design tricks have been the way Intel and AMD have delivered most of their increases in performance, sometimes you cannot argue with the benefits derived from making things just plain faster.
One example of this shines through with the on-die triple-channel DDR-3 memory controller that Intel has built for Nehalem. Coupled with its new ‘QuickPath Interconnect’ – a new high-speed interconnect that resides between the processor, chipset and memory – Nehalem’s memory bandwidth climbs to three times that of previous generation Core micro-architecture solutions.
The new micro-architecture also features evolved Hyper-Threading technology capable of running up to 8-threads on four cores; and a couple of new instructions, causing it to rename its instruction set for Nehalem SSE 4.2.
For the more technically minded, these new instructions deliver accelerated string and text processing, accelerated searching and pattern recognition of large datasets and new communications capabilities.
The really interesting parts of Nehalem are not about performance however – we have seen faster chips before.
Intel designed Nehalem to be more power efficient and more importantly to be faster. And that is something that is remarkable.
Focusing on power consumption
As the energy crunch continues to affect the cost of IT, green computing is becoming very en-vogue.
And fitting in with that trend, Nehalem was designed with a non-negotiable rule in mind – every feature proposed for Nehalem that would increase the processor’s power consumption by 1% had to deliver a corresponding 2% or greater increase in performance. If the feature could not equal or beat this ratio, it was not added, regardless of how desirable.
So Intel’s engineers used 1-million transistors (more than the number of transistors used for the 486 processor) and built a new on-die power control unit (PCU) for Nehalem.
This new PCU allows for individual Nehalem cores to be near-completely shut off when they are in deep sleep states – thereby allowing them to consume a truly negligible amount of power.
The other benefit of handling power management on the processor die is that the cores can be woken up extremely quickly. Nehalem thus goes from zero to hero in the performance stakes, very quickly.
Turning power saving on its head
But saving power is not everything. With Nehalem Intel has changed power management on its head and actually allowed headroom in thermal envelope created by deep sleeping cores, to its boost performance.
Called Turbo mode, the feature works as follows. When only one or two of the cores in the Nehalem processor are active (and the others are shut-off by the PCU), the chip runs far cooler than what it was intended to.
Taking advantage of this, Turbo mode automatically allows the active cores to go up a single clock step of 133 MHz. If only a single core is active, the PCU allows for one additional clock step, pushing the clock speed on that one core up by 266 MHz.
These new features make Nehalem a really impressive addition to the processor line-up in Intel’s stable, and one that will no doubt do wonders in the server market and desktops that run heavily threaded applications.
The processor family is destined for the server space first (a six core Xeon is already available to the market) and will soon be available for high-performance desktops.
Intel has said that the long-term view is for this technology to migrate into the mobile space too, where the power savings and performance boosts will definitely come in handy.
Looking ahead to Larrabee
A subject of much speculation in the market is Intel’s upcoming Larrabee processor – a processor that is rumoured to have the graphics vendors in the market quite worried.
That is because Larrabee is designed to cater for much of the functionality currently dealt with by graphics cards, right on the processor. It is a bold set of goals for Intel however, since previous attempts to get into the graphics market have not been all that successful.
Unfortunately, with the product due to begin shipping sometime next year, the details are still sketchy.
So here is what we know so far:
* Larrabee will be the industry’s first many-core x86 Intel architecture.
* Initial product implementations of the Larrabee architecture will target discrete graphics applications, support DirectX and OpenGL, and run existing games and programs. Additionally, a broad potential range of highly parallel applications including scientific and engineering software will benefit from the Larrabee native C/C++ programming model.
* While Nehalem has been interesting to take a look at, the market is waiting in eager anticipation of Larrabee. If Intel gets Larrabee right, it has the potential to change the way PCs are architected today.
New chips and their accompanying performance boosts are always a good thing. That said, the software market still has a great deal of catching up to do in order to consume all of the performance chip vendors are providing. Right now there simply is not enough software available capable of taking true advantage of the multiple cores and multithreadability of modern processors.
The Nehalem family of processors does however provide a native speed improvement for applications that have not yet been written with multithreading in mind, albeit about 10-15%.
On the solid-state drive front, even though prices have come down, we are still alarmed at how much extra SSDs cost by contrast to conventional mechanical drives. On the upside though, prices are coming down and the more SSDs vendors sell, the quicker prices come down.
It is clear that SSDs will become commonplace in every computer over the coming years. But it is going to take longer than two or three years.