Nehalem Uncovered (Sandra Prior)
Mar 27, 2009
Core 2, Conroe was an amazing processor, and with the Penryn update the highly successful architecture was taken from 65nm to 45nm. Smaller paths for electrons means higher speeds for data and at 45nm the Core 2 processor flourished. That said, the time has come for the highly successful Penryn and Conroe architectures to be retired.
The effect that Conroe had on the desktop market was amazing, the NetBurst architecture was a disaster that saw the good name of Intel become mud, so when Conroe was released, Intel fan boys rejoiced. They could once again say with pride that they had an Intel chip in their rig.
The performance difference between Conroe and Prescott was substantial but, as in the nature of processors, its nothing compared to the performance difference between Conroe/Penryn and Nehalem. Prior to dissecting the finer points of the new chip, it is necessary to take a brief look at the chips, that have brought us to this point.
Part and parcel of Intel's roadmap is the distinction of a processor as either a tick or a tock. Intel's Core architecture formerly known as Merom for example was a tock, meaning it was a processor with a new micro-architecture. Penryn was a tick, which means it was merely a new process implementation, and in this case it was the 45nm fabrication process. Intel then took all its experience and began design on the Nehalem, a tock.
Nehalem was in fact, being designed before Penryn. A tock starts off its design cycle before a tick for the previous tock has even been started. The team in charge of the tick then looks at all the features being implemented in the future tock and then simply decides which of the new features they can incorporate into the tick processor. So now, after five years in development Nehalem is just a few months away from being released to the public.
The name is i7, Core i7
Intel has, over recent months, released a steady stream of information regarding the new Nehalem architecture. One of the latest pieces to come out of the Santa Clara giant is the name for the new chip. Nehalem was the code name for the architecture after all and while these names usually stick, the market nomenclature will be Core i7. It seems that Intel is tight lipped about the new naming scheme but has promised clarity once the rest of the new line-up is revealed.
So what features can one expect from the new Core i7? Most notable is of course speed. Intel has finally designed a chip with an on-die memory controller. This will support triple channel memory, and motherboards with six DIMM slots have already been spotted. If you do your numbers right and think of placing 4GB memory modules on a board like that you could get a staggering 24GB of memory, Vista fans rejoice. There is however a downside; Intel in its infinite wisdom has only included official support for DDR3-1066, although bizarrely performance is said to increase with faster memory despite the controllers apparent lack of support for it.
Additionally, the use of an on-die memory controller has meant a change in the supporting architecture; primarily to the motherboard and the traditional communication pathway called the Front Side Bus. Proponents of much faster and efficient systems can now rejoice as Intel has equipped the Core i7 with an internal communication system know as Quick Path Interconnect, or QPI for short. QPI is very similar to AMD's HyperTransport which is something that AMD has had for a while now as well as the on-die memory controller (this is a fact AMD likes to point out). The use of the IMC has caused the memory bandwidth to reach an astounding 12GB/S, which is huge when considering that the maximum an Intel QX9770 system could achieve was 6.9GB/S, yet these figures pale in comparison when looking at the bandwidth afforded by the QPI over its predecessor, the FSB. The QPI bandwidth is 25.6GB/S which is a staggering two times that of the theoretical bandwidth of the X48 chipset. Intel really has out done itself with this one.
HyperThreading that works
A feature that we see returning to Intel's processors is HyperThreading or as Intel are now calling it Simultaneous Multithreading. You may recall that HyperThreading was implemented in the Pentium 4 with rather unappealing results. A performance increase of around 5% was an average for this supposedly amazing technology. The use of SMT on the new Core i7 chips and we can say without a doubt that it now works.
While performance results are wrapped up in legal red tape, what we can say, is that if what Intel was showing wasn't tweaked, SMT finally works and oh boy does it work well. Intel regards SMT as the most power efficient performance feature on Core i7 because it's able to keep the execution engine fed with multiple threads and therefore the processor never gets a single nanosecond to idle. Intel says that the biggest problem with HyperThreading in Pentium 4 was that the cache was not big enough and the memory bandwidth was insufficient, so because there is more cache in the Core i7 and the memory bandwidth is huge SMT/HT finally works.
The real benefit of SMT is that it costs very little in terms of silicon and it is far easier to implement rather than squeeze another core onto the die.
About the AuthorSubscribe to Sandra Prior's Online Newsletter and get up to date Computer Technology News delivered right to your email box for free. See website for details http://usacomputers.rr.nu and http://sacomputers.rr.nu..