Perhaps the biggest factor that had previously prevented me from building a new server was the upcoming release of Intel’s new micro-architecture: Nehalem. Built to address the shortcomings of Intel’s processors in the server market (which is still ruled by AMD), Nehalem promised to be “the next big thing” for Intel. So what better processor to adopt in MY new server than one optimized for servers?
Well, I can now comment on the wisdom/foolishness of my decision to forego the wait for Nehalem and build a P45/Penryn Core2 Quad server. Anandtech recently reviewed the Nehalem release (officially called “Core i7”), so I have a much better idea for whether my decision to buy existent hardware vs. cutting-edge technology was a good choice.
The processor I purchased is a Penryn Core 2 Quad Q9400 processor. The stock clock speed of this processor is 2.66 GHz, and according to what research I’ve done browsing online forums, they are easily overclockable to 3.5 GHz on air cooling and stock voltages. In other words, you don’t have to do much to get a lot of extra processing power out of this chip. This Penryn processor is built on Intel’s 45nm manufacturing process, so it has the same manufacturing process advantages as Nehalem (also 45nm).
Nehalem ships in three versions, the lowest of which is a 2.66 GHz processor – same clock speed as the Q9400. My question was, given that a similar part for Nehalem was available, was there any reason to stick with Penryn and not go to Nehalem? As it turns out, there are several reasons. For the time being, I’m going to be extremely glib in my discussion of the comparison between Penryn and Nehalem to get across the basic justification for my choices.
The 2.66 GHz Nehalem processor has a 130W TDP (thermal displacement profile), whereas the 2.66 GHz Penryn Q9400 processor has a TDP of 95W. That’s a pretty significant difference, folks. Especially when considering that my server will essentially be running in a 24/7 configuration consuming 95W of electricity vs. Nehalem’s 130W, the energy savings add up quickly. So at least between the two comparable processors, the energy savings are pretty stark (the advantage quickly shifts as you get to higher clock speeds, though). Again, I’m really over-simplifying the power consumption comparison issues here, but for my server, the electric bill is the bottom line.
Ok, so the processor I chose has better power savings than the new version. What about performance? After all, Nehalem was supposed to be a brand spankin’ new architecture. The changes that Intel made to Nehalem were certainly significant (I won’t go into them in detail in this series), but what would the real-world impact be? Overall (again, simplifying here) Nehalem shows an average of a 10-20% performance advantage over Penryn. Is that enough by itself to justify an upgrade?
Well, let’s consider the hedonic contrast I’d experience going from a single-core 1.8 GHz Opteron to a quad-core 2.66 GHz Penryn. At least from initial usage, the difference is like night and day. In some cases, I’d probably be experiencing anywhere from a 100-400% increase in performance (single-thread to quad-thread processes). So in comparison, 10-20% isn’t really a major difference. So at least from a performance angle, Nehalem doesn’t offer a big enough improvement to justify itself as an upgrade.
Ok, so what about price? After all, I’m building computers on a very tight budget, so I don’t have the advantage of deep pockets. I paid a relatively inexpensive $199 for the Penryn Q9400 (normally it runs closer to $270, but I got it cheaper). A new Nehalem processor at the same clockspeed will likely cost close to $300. So right away it becomes unjustifiable to buy a processor with a 10% performance increase for a 50% increase in price.
As I alluded to in my previous post, the inclusion of Windows Server 2008 gives the entire decision-making process of building a server a clear goal: provide a stable platform to adequately run Windows Server 2008 for several years. This brings up several points: stability, reliability, and performance. As I’ve discussed, the performance comparison is pretty much a non-issue with a 10-20% performance gap. The stability and reliability points, however, are another matter.
New technologies are a double-edged sword. On the one hand, they offer (usually) a technologically superior improvement over existing technologies. On the other hand, they are often plagued with bugs, unforeseen issues, and compatibility problems. Nehalem is even too new to tell what bugs or unforeseen problems may arise, but the compatibility issue is clear enough. Nehalem processors will require an entirely new socket. All existing motherboards that work with Penryn (and before them, Mermon) processors will be incompatible with Nehalem.
I have few doubts as to the stability of Nehalem, as Intel’s processor development and manufacturing have been firing on all cylinders since the release of Mermon (the first of the Core series architecture). But as to the stability of the individual motherboards that will play host to the Nehalem processors, it’s very unclear at this point which actual boards will be good candidates for an install of a non-transferrable license of Windows Server 2008.
Penryn (and the P45 chipset boards that support them) have been around long enough to provide a decent number of reviews testifying to the reliability or unreliability of certain motherboards. As such, choosing a P45 board on NewEgg.com that has received over 800 reviews gives me a great deal of data points from which to construct an idea for how reliable a particular motherboard is. The board I chose (which I will get to in another part of this series) was chosen because I had good reason to believe that it would be a stable platform that could run a license of Server 2008 for several years without dying.
So, I’ve determined that the Penryn 2.66 GHz Q9400 is superior to the equivalent Nehalem 2.66 GHz quad-core processor in power consumption, price, and reliability (mostly due to motherboard compatibility). What the Penryn processor lacks compared to the Nehalem version is performance. Does the 10-20% increase in performance justify a 50% increase in price, a 40% increase in power consumption, and a currently unknown reliability rate? Methinks not.
I don’t want to sound like I’m bashing Nehalem here. It’s a significant improvement in Intel’s processor micro-architecture; one that will benefit future iterations of Intel’s processor families. But right now, it’s still new technology with the disadvantages of being expensive, largely untested (again, more of a motherboard issue), and power hungry (although future releases of Nehalem processors should have lower TDPs).
So, did I make a good choice with regard to the processor I decided to purchase? I think so, given the fact that my choice was largely a debate between Penryn and Nehalem. I didn’t mention AMD in this post because while Phenom does have some positive attributes, the high TDP (130W for most of the quad-core chips) and lackluster performance compared to Intel’s Penryn quad-cores pretty much ruled it out as a possible candidate for my server.
In my next article, I’ll talk briefly about the heatsink I chose, but then move on to the more important (and more interesting) choice in motherboards.