Introduction

Intel Core 2 Extreme QX9650, formerly known as Yorkfield, is the first 45-nm desktop CPU from Intel, being a quad-core CPU running internally at 3 GHz and externally at 1,333 MHz, the same clock specs of Core 2 Extreme QX6850. QX9650, however, brings two novelties: the new SSE4 instruction set and a larger 12 MB L2 memory cache, making it the most high-end desktop CPU available today. Did these two new features improve the CPU performance? That is exactly we are going to find out.

click to enlarge
Figure 1: Core 2 Extreme QX9650 engineering sample.

We have already written two articles explaining what is new on the 45 nm manufacturing technology from Intel, also known by the codename Penryn: Details on Intel’s Forthcoming 45 nm Manufacturing Technology and Penryn Core New Features. We highly recommend you to read these two articles in order to understand the manufacturing technology behind Core 2 Extreme QX9650.

Just to clarify, Penryn isn’t the codename of a specific processor, but the codename of Intel’s 45-nm manufacturing process. The codename for desktop CPUs using Penryn technology is Yorkfield. So you might see this CPU being referred by these two codenames.

On the table below we summarized all CPUs included in this review with their main specs. As we mentioned Core 2 Extreme QX9650 has exactly the same clock specs as Core 2 Extreme QX6850 – working internally at 3 GHz, which is obtained by multiplying its external clock by 9 – so we were very curious in comparing these two CPUs.

As you can see, this quad-core CPU continues to use two L2 caches, so its 12 MB L2 cache is in fact two 6 MB caches, the first one being shared between cores 1 & 2 and the second one being shared between cores 3 & 4. For a more detailed explanation on this subject please read our article Intel Quad-Core CPU Overview and Roadmap.

For the record, we always configured our DDR2-1066/PC2-8500 memories running at 1,066 MHz.

CPU	Cores	Internal Clock	External Clock	L2 Memory Cache	SSE4	Platform	TDP	Man. Tech.
Core 2 Extreme QX9650	4	3 GHz	1,333 MHz (333 MHz x 4)	6 MB x 2	Yes	Socket 775	130 W	45 nm
Core 2 Extreme QX6850	4	3 GHz	1,333 MHz (333 MHz x 4)	4 MB x 2	No	Socket 775	130 W	65 nm
Core 2 Extreme QX6700	4	2.66 GHz	1,066 MHz (266 MHz x 4)	4 MB x 2	No	Socket 775	130 W	65 nm
Core 2 Extreme X6800	2	2.93 GHz	1,066 MHz (266 MHz x 4)	4 MB	No	Socket 775	75 W	65 nm
Core 2 Duo E6750	2	2.66 GHz	1,333 MHz (333 MHz x 4)	4 MB	No	Socket 775	65 W	65 nm
Core 2 Duo E6700	2	2.66 GHz	1,066 MHz (266 MHz x 4)	4 MB	No	Socket 775	65 W	65 nm
Pentium 4 550	1	3.4 GHz	800 MHz (200 MHz x 4)	1 MB	No	Socket 775	115 W	90 nm

TDP, Thermal Design Power, is how much power the CPU dissipates, meaning that you must match the CPU with a cooler capable of dissipating that amount of power.

How We Tested

During our benchmarking sessions we used the configuration listed below. Between our benchmarking sessions the only variable device was the CPU being benchmarked.

Hardware Configuration

• Motherboard: ASUS P5K-E/WiFi-AP (0401 BIOS)
• Memory: 2 GB Corsair Dominator TWIN2X2048-8500C5D (DDR2-1066/PC2-8500 with 5-5-5-15 timings), configured at 1,066 MHz
• Hard Disk Drive: Seagate Barracuda 7200.10 160 GB (ST3160815AS, SATA-300, 7,200 rpm, 8 MB buffer)
• Video Card: Gigabyte GeForce 8800 GTS 320 MB
• Video resolution: 1440x900 75 Hz
• Video Monitor: Samsung Syncmaster 932BW
• Power Supply: OCZ ProXStream 1000 W
• CPU Cooler: Thermaltake TMG i1
• Optical Drive: LG GSA-H54N

Software Configuration

• Windows Vista Ultimate 32-bit

Driver Versions

• nVidia video driver version: 163.75
• Intel Inf chipset driver version: 8.3.1.1009
• LAN driver version (Marvell): 9.0.32.3
• LAN driver version (Realtek): 6.1285.215.2007

Used Software

• PCMark Vantage Professional 1.0.0
• VirtualDub-MPEG2 1.6.19 Build 24586 + DivX 6.7
• DriverHeaven Photoshop Benchmark V2
• Cinebench 10
• 3DMark06 Professional 1.1.0 + October 2007 Hotfix
• Quake 4 – Patch 1.4.2

We adopted a 3% error margin. Thus differences below 3% cannot be considered relevant. In other words, products with a performance difference below 3% should be considered as having similar performance.

PCMark Vantage

The new PCMark Vantage program simulates the use of real-world applications and gives scores for the following categories:

• PCMark
• Memories
• TV and Movies
• Gaming
• Music
• Communications
• Productivity
• HDD

For a detailed description of each one of these tests, please download and read the PCMark Vantage Reviewer’s Guide.

You can see the results for each category below. We are not going to compare the results for the Memories and HDD suites.

 

VirtualDub-MPEG2 + DivX 6.7

With VirtualDub we converted a full-length DVD movie to DivX format and saw how long it took for this conversion to be done. DivX codec is capable of recognizing and using not only more than one CPU (i.e. more than one core) but also the new SSE4 instruction set, so we expect that CPUs with SSE4 support reach a higher performance here.

The movie we chose to convert was Start Trek – The Motion Picture: Director’s Cut. We copied the movie to our hard disk drive with no compression, so the final original file on our HDD was 6.79 GB. After compressing it with DivX, the final file was only 767.40 MB, which is quite remarkable.

The results below are given in seconds, so the lower the better.

 

VirtualDub-MPEG2 + DivX 6.7	Score	Difference
Core 2 Extreme QX9650 (3 GHz)	1868.35
Core 2 Extreme QX6850 (3 GHz)	2326.08	24.50%
Core 2 Extreme QX6700 (2.66 GHz)	2480.07	32.74%
Core 2 Extreme X6800 (2.93 GHz)	2890.99	54.74%
Core 2 Duo E6750 (2.66 GHz)	3102.93	66.08%
Core 2 Duo E6700 (2.66 GHz)	3134.10	67.75%
Core 2 Duo E6600 (2.4 GHz)	3453.18	84.83%
Pentium 4 550 (3.4 GHz)	6216.51	232.73%

Photoshop CS2

The best way to measure performance is using real programs. The problem, though, is creating a methodology using real software that provides accurate results. For Photoshop CS2 there is a methodology created by the folks at DriverHeaven that is very accurate. Their script applies a series of 12 filters to a sample image and we write down the time taken to each filter to run. At the end we have the results for each individual filter and we simply add them up to have the total time taken to run the 12 filters from DriverHeaven batch. The results below are given in seconds, so the lower number the better.

Cinebench 10

Cinebench 10 is based on the 3D software Cinema 4D and it is very useful to measure the performance gain given by having more than one CPU installed on the system when rendering heavy 3D images. Rendering is one area that having more than one CPU helps a lot, because usually rendering software recognizes several CPUs – Cinebench, for instance, can use up to 16 CPUs.

Since we were interested in measuring the rendering performance, we ran the test called “Rendering x CPUs”, which renders a “heavy” sample image using all available CPUs (or cores) to speed up the process. Keep in mind that even though the Pentium 4 CPU we included in our review has only one core, it has Hyper-Threading technology, which simulates two CPUs.

3DMark06 Professional

3DMark06 is the latest version of 3DMark franchise, measuring Shader 3.0 (i.e. DirectX 9.0c) performance. We run this software on its default configuration (1280x1024 resolution with no image quality settings enabled) and besides the 3D score given by this program we also compared the results from its internal CPU benchmark.

Quake 4

We upgraded Quake 4 to version 1.4.2 and ran its new multiplayer demo id_perftest at 1280x1024x32 with SMP option enable (which allows Quake 4 to recognize and use more than one CPU), under two scenarios: first with image quality settings configured at “low” and then with image quality settings configured at “high”. You can check the results below, given in frames per second.

Quake 4 1.4.2 - Low	Score	Difference
Core 2 Extreme QX9650 (3 GHz)	202.65
Core 2 Extreme QX6850 (3 GHz)	202.27	0.19%
Core 2 Duo E6750 (2.66 GHz)	198.87	1.90%
Core 2 Extreme X6800 (2.93 GHz)	198.59	2.04%
Core 2 Extreme QX6700 (2.66 GHz)	196.48	3.14%
Core 2 Duo E6700 (2.66 GHz)	195.08	3.88%
Core 2 Duo E6600 (2.4 GHz)	190.26	6.51%
Pentium 4 550 (3.4 GHz)	77.22	162.43%

Quake 4 1.4.2 - High	Score	Difference
Core 2 Extreme QX9650 (3 GHz)	193.59
Core 2 Extreme QX6850 (3 GHz)	193.43	0.08%
Core 2 Duo E6750 (2.66 GHz)	191.22	1.24%
Core 2 Extreme X6800 (2.93 GHz)	190.53	1.61%
Core 2 Extreme QX6700 (2.66 GHz)	188.74	2.57%
Core 2 Duo E6700 (2.66 GHz)	187.06	3.49%
Core 2 Duo E6600 (2.4 GHz)	182.71	5.95%
Pentium 4 550 (3.4 GHz)	75.23	157.33%

Overclocking

This CPU has an outstanding overclocking capability. Since we got an engineering sample, it had its clock multiplier unlocked, so we were able to increase its clock multiplier to x10 for a 3.3 GHz CPU, working just fine. The program that was most benefited by this internal clock increase was Cinebench 10, where we saw an 11% performance increase.

It was also really easy to increase this CPU external clock rate (FSB) up to 410 MHz (1,640 MHz QDR), but we achieved stability (i.e. we could run PCMark Vantage without crashing) only at 397 MHz (1,588 MHz), which made our CPU to run at 3.57 GHz internally keeping its default x9 clock multiplier. At this configuration we saw an 18.42% performance increase on Cinebench 10. Keep in mind that this program is the one that gets most benefit from internal clock rate increase, i.e other programs achieved lower performance gains.

This basically means that you can practically “transform” this CPU into a 1,600 MHz FSB part, which is the next external clock rate for Intel processors.

Conclusions

Since the new Core 2 Extreme QX9650 has the same clock specs as Core 2 Extreme QX6850 we were very curious to see whether this new 45-nm quad-core CPU was faster or not than the previous 65-nm part. As we had already explained the only two differences between these two CPUs besides the manufacturing technology is the size of the L2 memory cache (6 MB x 2 vs. 4 MB x 2) and the addition of the new SSE4 instruction set.

The truth is that on regular applications we saw no performance gain at all, even with a bigger L2 memory cache, except on the Music test suite from PCMark Advantage, where Core 2 Extreme QX9650 was 4.81% faster, and on Photoshop CS2, where it was 3.80% faster.

On Cinebench 10, however, we saw a significant 9% performance increase.

But where QX9650 really made the difference was on DivX compression: it was 24.50% faster than QX6850, reducing the time required to compress our Star Trek movie from almost 39 minutes to 31 minutes. This happened because this codec recognizes the new SSE4 instruction set.

So if you are a professional user especially on the audio and video industry and your favorite program has support for the new SSE4 instruction set, you should see a major performance gain just by enabling SSE4 on your program – DivX 6.7 and TMPGEnc 4.0 Xpress are two examples of audio/video encoding software that use SSE4, and we should expect more programs supporting this new instruction set as more desktop CPUs using it reach the market.

The good news is that Core 2 Extreme QX9650 reached the market costing the same thing as QX6850 – USD 999 for distributors buying at least 1,000 units directly from Intel; the retail price is higher than that –, so if you have the money and are looking for the most high-end desktop CPU available today, QX9650 is definitely your best option. Also unless Intel cuts the price on Core 2 Extreme QX6850 it is today a CPU simply not worth buying, as for the same price you can get a QX9650 that will deliver you a higher performance on forthcoming programs, especially audio- and video-related.

Originally at http://www.hardwaresecrets.com/article/499

---

© 2004-9, Hardware Secrets, LLC. All Rights Reserved.

Total or partial reproduction of the contents of this site, as well as that of the texts available for downloading, be this in the electronic media, in print, or any other form of distribution, is expressly forbidden. Those who do not comply with these copyright laws will be indicted and punished according to the International Copyrights Law.

We do not take responsibility for material damage of any kind caused by the use of information contained in Hardware Secrets.