Intel's Core 2 Under Linux

Tuesday, August 22 2006 @ 02:47 AM CDT

By now, you've probably heard that Intel has a new chip out that went by the code-name “Conroe” but now goes by the names Core 2 Duo and Core 2 Extreme. So while you may have read about how great these chips are under Windows, I bet you have heard little to no news about Linux performance or even Linux support. As usual, this is where Linux Hardware comes in to fill in the gaps. I have each of these top two new processors from Intel as well as some DDR2-800 and a new NVIDIA GeForce 7600 GT. I'll walk you through the features of this new core processor technology from Intel, take you through support and performance, and end up with a discussion on power, heat, cost, and availability. There's a lot to cover here, so let's get going.

Article notes located at the end of the article updated on: Aug. 23, 2006, 5:30 PM EST.

Intel's “Conroe” Core 2 Processor

Intel's new Core 2 processors represent a change of direction for Intel. Core 2 leaves the NetBurst architecture behind which has been at the center of the Pentium 4 product line. This change comes from the realization that the Pentium 4 line isn't going to be able to go much beyond its current clock speeds due to high heat and power requirements. Intel had banked on being able to take this core up much higher than the last processor released (3.73GHz), even up to 10GHz. Unfortunately once these processors started being produced, they were unable to deliver on the promise.

Core 2 Processor Die
(Click to Enlarge)

With the Core 2 architecture, Intel goes back to the Pentium-M core (which is an evolution of the Pentium 3 core) and refocuses on instructions per clock, power consumption, and heat production. Keep in mind though that Intel has done some major rework to the Pentium-M core and 80% or more of that core has been redesigned and/or optimized for this new line of processors. Intel's architects have been hard at work on this new core and we're about to see how this has panned out. Now, let's take a look at the specs and see what the hard numbers of these processors look like.

In the table above, we have a listing of the processors we'll be reviewing today and the specifications of each. First of all, we have Intel's top processor from the Pentium 4 line. The Pentium Extreme Edition 965 was Intel's last, and highest clocked, NetBurst-based processor and it will be the representative to show what Pentium 4 offered in terms of performance. This processor is a dual-core processor which also has Hyper-Threading. This makes the processor appear as four processors to the OS, which is two virtual processors per physical core. The goal of this technology is/was to maximize core utilization. The two cores of the Pentium 4 are actually two independent processors that were cut off a wafer side-by-side and then wired together to the chipset. Each core features 2MB of L2 cache and the cores are linked to the Northbridge by a 1066MHz front-side bus (FSB).

The next two processors are the new Core 2 processors. They both feature 4MB of shared L2 cache. This design allows each processor to use as much of the cache as necessary, as well as directly access data cached by either core instead of having to go through the other processor. Both Core 2 processors also feature a 1066MHz FSB. All three of the Intel processors here are manufactured with a 65nm manufacturing process and fit in an LGA775 socket. The only real differences between the Core 2 processors are the clock speeds and the ability to change the multiplier on the Core 2 Extreme. These features of the Extreme line are designed to make the Extreme processors more attractive to the enthusiast that has a lot of money to blow on a major system upgrade.

To comment on a few other features of the Core 2 processors, they also feature SpeedStep Technology, Execute Disable Bit, and Intel Virtualization Technology. SpeedStep Technology is a feature that was designed to make the processor scale to the amount of processor utilization and specific run modes. The goal is to lower idle power usage and save on electricity through less power draw and lower cooling costs. Execute Disable Bit is a hardware-based technology that prevents malicious code from running on your system without a user's knowledge. Software must be written to support this feature. Intel Virtualization Technology gives you the ability to run multiple operating systems on a system simultaneously. The details of this process are out of the scope of this article but you kind find out much more information on this through Intel's website.

I could go on from here about more specifics of Intel's new processor line but that kind of information can be found in brief on many other sites or more detailed in Intel's public technical documents. From here I'd like to just move on with the Linux discussion and show you what you really came for.

System Configurations

Below is a table outlining the system configurations for this review. First thing I'd like to mention is that AMD did not have a Socket AM2 Athlon 64 FX-62 available for this review. I wanted to bring that to you here but there is little that can be done about availability from mine or AMD's side. Don't worry, I do have an AM2 platform ready to go for the next processor release. I will also be bringing you a look at the platform through another review that I have lined up. Although I don't have an FX-62 for you, I did want to bring you close to the performance of one as possible so I overclocked the FX-60 on the socket 939 platform. This should give us numbers close to what you would see on the new platform and since the Athlon 64 FX line isn't multiplier locked, I was able to get FX-62 speeds without changing the memory clock speed.

One nice thing about Core 2 processors is that they'll run in some of the same motherboards as the previous generation of Intel processors. If you have a 975X chipset, you should check with your motherboard manufacturer for Core 2 support. Most 975X boards will support the new chips. If the 975X is too rich for your blood, then you can can checkout Intel's latest chipset, the P965, for some lower-priced motherboards. Seeing the backwards compatibility for the 975X from Intel was a pleasant surprise given their recent past of a new chipsets for each new type of processor.

Since we have a new core in this review, I decided it was prudent to pull out both 64-bit and 32-bit versions of Linux. We wanted to find out where Core 2 shows the most performance and how it did versus AMD in each mode. The table below shows the compile options for each OS and processor type. I didn't go all out and be over-aggressive here. It was important to compile the applications for each specific processor type but not be too aggressive and cause instability.

Unfortunately, there doesn't seem to be any new optimizations for the Conroe in GCC yet. For 32-bit, I chose the Pentium M architecture as the closest match, and I added SSE2 support via a secondary compiler flag. For 64-bit, I has only one option for Intel 64-bit processors which was for a “Nocona” core.

Support and Stability

There's good news and more good news in this category from Intel. First of all, since the motherboard hasn't changed for this new generation of chips, all the top-notch support found for the existing 975X support is still there. On the Intel motherboard I used in this review, all components are supported including ACPI, Ethernet, and on-board audio. All I had to do on my existing board to support Conroe was update my BIOS to the latest released on Intel's website. Next, Intel's technologies mentioned above including SpeedStep and Execute Disable Bit, are available in the latest kernels and should work find with the Core 2 processors. Even Virtualization support is being worked on through several packages which can be found through a Google search.

Stability on this system was rock solid. I didn't have one crash from this platform, explained or unexplained. Everything “just worked” and that's always a pleasant experience with new hardware,

Performance

For my performance testing, I try to pick a mix of applications that will showcase various components of the system and test the processors under different situations. In this review you will see many of the applications that I have used in the past but there have been a couple of changes to stay current. I will refrain from any comments on the Pentium EE 965. It is the slowest part here by a large margin and only makes a good showing at SPECViewperf. It is mainly here for reference to show just how far Intel has come.

POVRay Website / POVRay Benchmark

POVRay is a single-threaded application that stresses the floating-point ability of the CPU. POVRay computes image data for a scene described in a text file. This has been an application that has been traditionally strong for AMD. For the first time, AMD takes a loss in this benchmark. It's not by a huge margin but the Core 2 Extreme starts out strong here. Notice that we see a much faster processor in 32-bit mode than 64-bit mode, a whopping 21% increase. Let's see if this holds true for the remainder of our testing.

Ogg Vorbis Website
Our Commands: oggenc input.wav, oggenc -b 192 input.wav

Since we support Open Source projects here, we would never use a proprietary audio CODEC that would require a license. So for our audio encoding benchmark, we turn to Ogg Vorbis which provides excellent audio compression without any annoying proprietary technology. The Oggenc application is single-threaded. Here both Core 2 processors were able to take out the Athlon 64. This time it was the 64-bit side of the Core 2 that shined. The Athlon 64 was about 15% slower than the Core 2 Extreme and about 5% slower than the E6700.

AudConvert is a Linux Hardware Application
Earlier version still used here.

My Audconvert application takes audio encoding and makes it multi-threaded. It uses any set of audio de-encoders and audio encoders to convert audio files from one format to another. The application is configurable to any number of threads and takes a directory of files for its input. For testing, I used a folder of ripped content from my own CD collection (about 21 songs total). Core 2 takes this benchmark by a small margin but, once again, even the second-tier E6700 was able to edge out the 2.8GHz Athlon 64. Again 64-bit was faster than 32-bit.

MPlayer/MEncoder Website
Our Command: mencoder merry_melodies_falling_hare.mpeg -o merry_melodies_falling_hare.avi -ovc lavc -lavcopts vcodec=mpeg4:mbd=2:vbitrate=1800:threads=# -oac copy
The video we used is freely available and can be found here at the Internet Archive.

MPlayer and MEncoder are staples for most Linux users if you play videos or encode videos from any number of sources. In this benchmark we re-encode a freely available movie found on the Net using MEncoder. This benchmark utilizes multiple threads as defined in the command-line. At encoding video, Core 2 is extremely strong and almost 30 seconds separates the Athlon 64 FX from he Core 2 Extreme. Notice that the previous Pentium 4 wasn't too far behind the Athlon 64 but that Core 2 takes a huge leap ahead in this benchmark. Again, AMD takes a loss here to both Core 2 processors.

Firefox Website

The best way to benchmark the ability of a processor to compile software is to grab a large software package that takes at least a couple of minutes to compile and let it fly. For this, I chose Firefox due to its wide availability and large code-base. The build process is managed by Gentoo's portage system and includes uncompressing the distributed tar file, compiling the source, and installing the application. At compiling applications, Intel keeps the lead with both Core 2 processors and with the Core 2 Extreme, the job was almost completed in 10 minutes flat.

Quake 4 for Linux

For gaming benchmarking, I finally had to retire UT2004 due to its 80FPS ceiling. It was to the point that all the numbers were maxing out and none of the CPUs looked slower or faster than any other. I've also switched over to Quake 4 from Doom 3 due to SMP support and ease of benchmarking. For this benchmark, I recorded my own demo and played it back under timedemo mode. This benchmark, as well as instructions on how to run the benchmark, will be released on this site soon.

For this benchmark we ran Quake 4 in SMP mode. At the low resolution of 640x480, we see the only real differences between the processors. Once we go up from there, the graphics card becomes the bottleneck and the processor has little impact. At 640x480, Core 2 wins again with both processor models. In gaming benchmarks we give credit where credit is due, but we also keep in mind that no one plays games at 640x480 anymore.

SPECViewperf Website

The two graphs above are of SPECViewperf, the workstation graphics benchmark. Although version 9 is out, I had to stick with 8.1 due to the lack of a UNIX tar.gz release. In our final benchmark, Core 2 wins again in a benchmark that has always been strong for Intel.

So to sum up the performance category, that's Core 2 Extreme - 7 and Athlon 64 FX-60 – 0. What's even more impressive is that we have Core 2 Duo – 6 and Athlon 64 FX – 1. With this in mind, let's move on to the next category and see how performance factors in to the total package of the Core 2.

Power Consumption, Heat, Cost, Availability

So now let's take a look at the cost of each of these processors from the standpoint of power consumption, heat generation, and purchase price. In the table below, you can see the specifications of each processor in terms of power consumption. The king of power consumption is the Pentium Extreme Edition 965, topping out at 130W. This processor sucks a lot of juice and runs very hot. The next hungriest processors are now the Athlon 64 CPUs, with the latest FX-62 almost matching that of the Intel 965 at 125W. Finally, we see the new Core 2 processors relatively sipping energy at only 65W and 75W for the Duo and Extreme respectively. In fact, the Core 2 Extreme uses 40% less power than the Athlon 64 FX-62. This is a huge turn around for Intel and it should make their processors attractive to anyone concerned about energy efficiency. It's important to mention here that AMD does have low power Athlon 64s available for those interested in low power processing. In fact, you can even get a processor as high as a 4800+ that only sips 65W, the same as the Core 2 Duo. The only catch is that AMD will be charging a small $30 or so premium on these processors and they aren't being offered on their entire line. This means that while you may be able to get a fairly powerful AMD chip at 65W, the top-of-the-line is not currently available.

While the table above is the official specs, I wanted to see how these performed in the real world. In the chart below, I show the maximum amount of power the system actually pulled while idle and running the Firefox emerge and the maximum temperature of the CPU idle and during the benchmark. Note that while the coolers of each CPU are not identical, I did use the coolers provided by each manufacturer.

Power consumption used to be one of AMD's strong points and, according to my testing, it still is. At idle, the Athlon 64 still pulls less power than even the new Core 2 processors. The Pentium 4 was off the charts with its almost 300W under full load. The Core 2 beats out AMD under load, but not by much. This is surprising given the Core 2's lower specs.

At both idle and load, the Core 2 processor puts up impressive numbers and just barely lags behind AMD. We'll give Intel a little slack here though and say that, especially under load, Core 2 is at least equal to Athlon 64 in heat generation. The numbers for Intel would have been more impressive if Intel's stock cooling was a little more substantial. It's funny to see the heat generation of the Pentium 4. At 98C, I might has well pulled out a pan for my heatsink and cooked up dinner!

Finally, the street price of each chip discussed here is listed above. The first thing you'll realize from the chart is just how much of a premium Intel is placing on their top chips. When you take a look at the performance difference, it's hard to justify an over $500 increase in cost for the Core 2 Extreme over the Core 2 Duo. Even the previous Pentium Extreme Edition still costs over $1,000. On the AMD front, they are also imposing a premium for their top chips, although since the Athlon 64 X2 5000+ can be found for $699.00, the premium is no where near as steep.

The Bottom Line

The bottom line here is that Intel has come back to the CPU market with a new core and with a goal to retake the performance crown by force. Anyone who said “the only way Intel will make a come back is to redesign everything,” they were right and that's exactly what Intel did. Intel not only has the fastest chip in their top processor, they even take the performance lead at their second tier chip in six out of seven of our benchmarks. As well as being the fastest thing on the market, it also runs neck and neck with AMD in the heat generation and power consumption race. The Core 2 processor really is a very well designed chip with something for everyone. Here is a list of shipping Core 2 processors with there specs and street prices at time of writing:

It's hard to say exactly where the Core 2 processor would lose its performance edge in this line-up but I'm betting that the E6600 would still stay very close to an FX-62 without much trouble. That's a heck of a lot of performance for less than half the price of AMD's top chip.

So how's AMD going to take back the lead from here? My guess is with more cores and maybe going mainline with their 65W parts. If they could produce all of their CPUs at that level then they would have a lot of headroom on future parts. Also keep in mind that AMD has yet to go to a 65nm manufacturing process. It will be interesting to see when that will happen and how much AMD will gain from the move.

Now is a great time to be CPU shopping because no matter which side of the aisle you look on, you have great choice for both CPUs and motherboards. Along with Intel's chipset offerings, keep in mind that NVIDIA has the nForce series for Intel CPUs which would give you SLI support for all your Quake Wars and UT2007 gaming needs.

Article Notes:

• I wanted to send a special thanks to Kingston for their support on this article. They hooked me up with some top-notch DDR2 at the last minute. Not only did it work, but it worked amazingly well. I was able to use their higher clocked memory for low-latency DDR2 testing. It really performed well under tight timings.
• Other honorable mentions for support: Intel, AMD, Corsair, and NVIDIA.
• All multi-threaded applications that had an option, were run with number of processors + 1 threads. This means 3 threads for Athlon 64 and Core 2 and 5 threads (for Hyper-Threading) for Pentium 4.
• Many people have commented on Core 2 availability. I wanted to mention that here since it seems to be a major concern. I have noticed personally that certain models of Core 2 are available while others are not. The most demanded processors do seem to be in shorter supply. This is common for new hot hardware. Intel will work it out as demand levels off and production ramps up.

Comments (80)

LinuxHardware.org
http://www.linuxhardware.org/article.php?story=06/08/22/0415251