Intel's New Core i7 and Xeon 5500 CPUs

Original Article Date: 2009-03-30

This month, Intel bring the radically new "Nehalem" architecture to the key dual-processor space for servers and workstations.

There is much to talk about here as it's not just a new CPU release, but a major upgrade of an entire computing platform. I'm sure you'll want to know the details straight away, so to get you oriented, here's a quick overview of what's new with Nehalem.

• The memory controller moves from the mainboard to the CPU itself using new Quick Path Interconnect
• System memory jumps from DDR2 667/800MHz Dual-Channel to DDR3 1066/1333/1600MHz Triple Channel
• Hyper-Threading is back - 8 threads of execution now available for each CPU
• Turbo-Boost Technology - dynamically "overclocks" CPU during peak loading
• A shared L3 cache of 8MB (some models have 4MB) instead of individual L2 caching

Needless to say, beneficial features of the previous generation of CPUs are also present, such as Virtualization Technology, 64-bit computing, Speed Step power saving, Execute Disable Bit virus-protection etc. And all CPUs are manufactured on 45nm process technology.

With the release comes the following new hardware packages:

• 3 new single socket CPUs
• 12 new dual socket CPUs
• 3 new chipsets
• At least twenty new motherboards from Intel, Asus, Supermicro and Tyan

The QuickPath to Success

Intel's new QuickPath Interconnect - the Front Side Bus is finally retired

Let's start with one of the most significant changes in Intel CPU design since the beginning of the x86 processor - the shift of the memory controller from the mainboard to the CPU itself.

Some of you more savvy folks may know that AMD did this trick back in 2003 with the launch of the Opteron CPU, and it was for this reason that this CPU outperformed the Intel Xeons of its day so well. One could wonder why it's taken so long for Intel to follow suit.

Why is this move significant? Well, it basically means that the CPU can now talk directly to the RAM instead of having to take a longer path through the front side bus to the memory controller on the motherboard. This results in much lower latencies and higher bandwidths. And since the interaction between CPU and RAM is the key governor of overall system performance, this move spells for big improvements in computing power.

But it's not just a copy of AMD's design - Intel have gone one further, and added a third channel to their on-chip memory controller. This means that now the CPU can communicate with RAM across a 192-bit link, instead of 128-bit in dual-channel, which in theory means up to 50% extra data rates.

This revision of architecture has led to an increased socket pin-out - 1366 "pins" to be exact, but like the previous 775/771 sockets, the chip itself is pinless, with the pins on the mainboard in a Land Grid Array ("LGA") - so you'll see the term "LGA1366" in descriptions of these processors.

Lastly, and as if the above wasn't enough, Intel have switched to DDR3 RAM from DDR2. DDR2 was limited realistically in its architecture to 800MHz. DDR3 at least doubles that ceiling to 1600MHz, although initially, RAM speeds of 1066MHz and 1333MHz will be the limit supported.

Intel have also conceded the unpopularity of Fully Buffered DIMMs, which transferred some of the work of the memory controller to the DIMM. There was a legitimate reason for this design, but the additional workload at the DIMM resulted in significant extra heat being generated, making cooling difficult, especially with fully populated memory banks. The new CPUs will support regular ECC Registered DIMMs in DDR3, and unbuffered versions for single core CPUs.

The Return of Hyper-Threading, plus Turbo Boost

Remember Hyper-Threading from the days of single-core CPUs? It fooled everyone into thinking they had more than one processor under the hood. When true dual-core processors came along, Intel ditched the feature, which presented two virtual CPUs to the operating system, and allowed the OS to use those "CPUs" to improve multi-threaded performance. Hyper-Threading did indeed help in smoothing out traffic on multi-threaded or multi-process demands placed on the CPU, although gains were modest, at best only 20%. In any case, HT is back and so now, your quad-core Nehalem CPU will appear as an eight-core processor in your OS, and dual processor systems will display sixteen!

Turbo Boost sounds like something out of a 1980s TV-show which starred a would-be lifeguard with big hair, and a car with Cylon front-lights and an 8086 processor under the hood. But twenty-five years later, Turbo Boost makes a real-life incarnation in the new Nehalem CPUs. But what exactly is it? Well, basically, it's on-the-fly, dynamic overclocking. When the CPU is stressed to the max, it increases the clock multiplier on the RAM by a number of increments, resulting in the CPU clock going over it's advertised clock speed. It's like an extension to the extant Speed Step technology that slows down the CPU clock during idle moments to preserve power.

What's particularly clever about Turbo Boost, however, is it can be core-selective, i.e. the amount of overclock depends on how many cores are being stressed. If all four cores are being maxxed-out, the clock-speed boost to all of them may be modest - only 7-8%. However, if just one core is being stressed, then Turbo Boost will increase the clock on that core by a much greater margin (if the other cores are idle, then their clock speeds will actually be reduced with Speed Step). This is a boon for legacy software that might only run on a single thread.

Incidentally, the Nehalem single socket CPUs I have personally tested overclock very easily, so when you get one for your workstation, you can experiment with higher clocks if you're desiring to squeeze more power out of it with minimal risk. The ease of setting up overclocking will depend upon the mainboard.

New CPU and Mainboard Models Revealed

So enough of all the technology. What can you actually buy today in terms of a new Nehalem CPU? There are two main subdivisions, CPUs for single-socket systems, and those for dual-socket configurations.

The single-socket CPUs come under the Core i7 brand name, which you've probably already heard of, since they have been available since late last year. There's a slight wrinkle, however, as Intel have further rebranded these CPUs as workstation processors under the newly released Xeon 3500 series. They are the same CPU underneath, however. These are shown in the table below.

The dual-socket CPUs are branded as the Xeon 5500 series, and these are new, released just today.

As far as mainboards go, there are a host of new models launching today that will work with dual Xeon 5500 series CPUs - 6 from Intel, 10 from Supermicro and 6 from Tyan. You can find out which of these boards we are integrating in the section below on our new systems.

The Core i7 - Hints of the Upcoming Power of Xeon 5500

The Core i7 Nehalem tears up the competition. Note E5440 score is 2 CPUs.

Ok, so I've given an overview of what's new in the Nehalem processors, and informed you of what choices of specification we can expect in the new range of boxed processors released today. But beyond all the hype, will you really see a difference in the performance of your workstation or server?

The answer, I believe, is a very confident "Yes!" Why? Because I've already seen a trailer of things to come with the release of the Core i7 chip back in December of 2008. I've built a number of workstations using this chip and have been amazed at how fast it performs in comparison to the previous generation Core 2 chip. And the customers that have received the systems have testified that system power and responsiveness exceeded their expectations.

I know benchmarks aren't everything, and can often be a distortion of real-world performance. But they do hint at overall trends in power, especially if a number of different tests are used and then brought together under a combined score. The nearby chart shows such a method of scoring on Memory, Integer and Floating Point performance when comparing a Core i7 CPU, a Core2Quad CPU and two Xeon 5400 CPUs, all at the same clock speed of 2.83GHz. 

The first thing to notice about the graph is that the Core i7's overall memory reading and writing performance is practically off the chart compared with existing Intel CPUs - at least doubling performance from its predecessor Core 2. This is demonstrating the integrated memory controller, QuickPath, and the triple-channel high-speed DDR3 memory capability in action. Expect the Xeon 5500s to have a similar jump in performance for memory-intensive applications that so many workstation and HPC users need. This type of application is the one which will likely respond most favorably to the switch to Nehalem and may experience a 100% speed increase overall.

Additionally, the Core i7 performs well on general CPU integer and benchmarks being almost 50% faster than it's predecessor Core 2 chip. Note that the single Core i7 is compared with two E5440s, and puts up a good fight in such a seemingly unfair test! Bearing in mind that QuickPath Interconnect is expected to scale linearly from one CPU to two, since it is NUMA aware (that is Non Uniform Memory Architecture - a method of improving dramatically access to RAM from the local bank or the other CPU's bank). It is expected then, that two Nehalem Xeon 5500 CPUs should outperform their clock-speed equivalent 5400 series predecessors by up to 50% in general computing tasks. That's definitely worth taking notice of.

NEW Nehalem Servers and Workstations Available NOW!

Ok, this is where the rubber hits the road. So you want a new Nehalem-based server or workstation? Well the good news is you can have one today! Electronics Nexus is ready with the launch of Nehalem to integrate your new server or workstation immediately! Our new line-up comprises three workstations, three tower servers, a rack server and a high-performance computing cluster node. The following table lists out the key aspects of these new systems.

Details on each of the new systems can be found by accessing each system directly on the links within the table, or by browsing our website. Which one do you want gracing your desktop or server room?

Summary

It's been a while since I've been this excited about a new product launch. But the arrival of Nehalem signals a quantum leap in the availability of very high performance computer systems to the wider market at prices that would have been unthinkable just 18 months ago.

Nehalem represents the final break in a decade-old reliance of Intel on the front-side bus technology that for so long capped the performance of their CPUs. With the bottleneck between CPU and RAM gone, and the embracing of high-speed DDR3 at the same time, it's no wonder that these processors, and their accompanying platforms, will deliver on a big promise.

Whatever you are doing in the world of computing, whether it is digital content creation, engineering design and analysis, administering database, web or virtualized servers, or a high-performance scientific-computing cluster, this launch applies to you, and could change the way you work from here on out.

Best regards,

Ben Ranson
Chief Systems Engineer
Electronics Nexus
http://elnexus.com
ben@elnexus.com
1-877-773-5366