As of the writing of this, there are two main processors offered by Intel which are still considered to be top of the line. What I will be basing my discussion on is the  differences in two made up market segments by Intel: the desktop processor and the server processor.

For purposes of this discussion I will look at the Intel i7 980X and the Intel Xeon X5680. Both of these processors were launched in Quarter 1 of 2010. Of course other high end Intel processors exists such as the Xeon 7500 series but for the purpose of this discussion the Intel i7 980X and Intel Xeon X5680 will be explored as they are very comparable processors and still very good processors at the time of this writing.

Looking at the specifications of these processors they are very similar. Both use a LGA 1366 socket, have 6 cores, 12 threads, a clock speed of 3.33 GHz, a max turbo frequency of 3.6 Ghz , 12 MB of cache, a bus/core ratio of 25, 32 nm, a bus/core ratio of 25, and a QPI speed of 6.4 GT/s. The first noticeable difference is the # of QPI links which for the i7 980x is 1 and the Xeon X5680 is 2.

One of the differences between these two processors is the QPI bus. The i7 980X has 1 QPI bus between the CPU and Input/Output Hub (IOH) and hence can only be installed as a single processor on a motherboard. The Xeon X5680 on the other hand has 2 QPI buses which allow one for data between the CPU and the IOH and the other for data between 2 different CPUS. This means the Xeon X56680 can be used with another in a dual processor motherboard.

Another difference between these two processors is the memory. The Intel i7 980X has a maximum memory supported of 24 GB where as the Intel Xeon X5680 has a maximum memory supported of 288 GB. Both have three memory channels. The i7 980X has an official memory support of DDR3-1066 where as the Xeon X5680 has official memory support of DDR3-800/1066/1333.  The other memory differences are the i7 980X has a maximum memory bandwidth of 25.6 GB/s with 36-bit physical address extension and no EEC memory support where as the Intel Xeon X5680 has a maximum memory bandwidth of 32 GB/s with 40-bit physical address extension and supports EEC memory. The i7 980X has a suggested retail price of $999 while the Xeon X5680 has a suggested retail price of $1663 (in quantities of 1,000). This price difference is quite large.

So there are three differentiators between the i7 980X and Xeon X5680: 1) the ability to run dual processors or not, 2) maximum memory size supported, and 3) ECC memory. Hence Intel charges a premium of over $600 for the ability to have dual processors, much more memory, and ECC memory.

However, a question needs to be asked, is the i7 980X or the Xeon X5680 faster? It turns out that the i7 980X is actually marginally faster than a Xeon X5680. According to a Performance Test benchmark which is generated from  thousands of high end CPU’s (updated on February 3, 2011) the i7 980X has a score of 10,469 and the Xeon X5680 has a score of 9,668. Hence a single i7 980X tends to be a few percentage points faster than a single Xeon X5680.

Another issue needs to be looked at, and that is the issue of ECC vs. non ECC memory. ECC stands for “Error Correction Codes” and is memory primarily intended for servers that are running 24/7 or for mission critical applications. ECC memory helps prevent errors from occurring in the memory chips. For office or home use ECC memory is not generally needed or used. Because of this, ECC memory is actually slower in peformance than non ECC memory.

The i7 980X only supports non ECC memory. However, does that mean that the Xeon X5680 will work for large amounts of memory with non ECC memory?  The answer to that is no. The Intel S5520HC server motherboard has 12 DIMM slots and six memory channels (three channels per processor)  but it only supports ECC Registered or ECC Unbuffered DDR3 memory. The manual for the board says “Non-ECC memory is not supported and has not been validated in a server environment.” So let’s look at another motherboard,  the supermicro X8DAH+-F.  While this motherboard supports up to 288 GB of ECC Registered DDR3 memory it only supports up to 48 GB of non ECC memory. So it would seem you would have a tough time if even possible to use a Xeon X5680 with large amounts of memory without using ECC memory.

So why does this make a difference? Well ECC memory is awfully more expensive than non-ECC memory. For example currently on Newegg, Patriot Memory has a 12 GB (3 x 4 GB) DDR3 1333 ECC Registered triple channel kit for $359.99. Also on Newegg, Patriot Memory has a  12 GB (3 x 4 GB) DDR3 1333 non ECC memory for $134.99. The difference in price here between non ECC and ECC memory, for the same amount of memory and same speed (voltage slightly different) is over $200.

Another issue with these two processors is overclocking. I run faster memory than the specification for the i7 980X as I run DDR3 1600 MHz at work. Motherboards for the i7 980X tend to be much more overclocking friendly than for the Xeon X5680.

I personally, along with other potential customers, require large amounts of memory which exceed 24 GB for intensive engineering calculations. I would much rather be able to have access to a faster processor (intel i7 980X) which is also cheaper and also be able to use cheaper memory (non ECC) as I don’t require ECC memory for my use.

However, there is a problem, Intel has decided to force a market segmentation and in doing so is hurting customers like myself. Since I am upset about this I called Intel customer service to voice my concerns and hear what they had to say.

Intel has their so called ‘desktop’ market and their so called ‘server’ market. Someone or some group of people at Intel have decided to impose a maximum memory limitation of 24 GB on the first generation i core series chips such as the i7 980X. The second generation i7 series chips has a 32 GB maximum memory limitation. This memory limitation is actually artificial and is deliberately hard factory sealed onto each chip so that the ‘desktop’ market and the ‘server’ market have some differences which allows Intel to charge a premium. According to Intel technical/customer support there is no way to unlock the imposed memory limitation of the i core series processors.

I disagree with Intel in their decision to enforce two separate markets and feel it hinders people like myself by forcing us to spend more money and get less performance in the process. I feel Intel needs to increase the maximum memory limitation on their processors substantially. I understand the need to want to differentiate between being able to use dual processors or just one processor at a time and the ability to use ECC memory. However, customers like myself really just want everything that is in the ‘desktop’ market but just with the ability to support more memory. Perhaps this is a third separate market that Intel could potentially differentiate. Or perhaps they will see the need and rationale to substantially increase the amount of maximum memory supported in their ‘desktop’ processor market. Of course motherboard manufactures will also need to incorporate these changes.

Sources (all accessed February 3, 2011):