The Pinnacle of Desktop CPU 2007 - Core 2 Extreme QX9650

Back in January earlier this year, we've heard Intel telling us about their latest processor technology breakthrough, utilizing a new combination of high-k dielectrics and conductors which allows Intel to fabricate 45nm processors, and shortly after that, we've heard the codename "Penryn" floating around the web. All in all, the Penryn family will consist of Penryn for the mobile platform, Wolfdale for the Core 2 Duo and Yorkfield for the Core 2 Quads.

Yorkfield XE architecture isn't revolutionary compared to the current Kentsfield XE Core 2 architecture as evident from Intel's Tick Tock model but the next generation Nehalem architecture will be. Yorkfield is 45nm based packed with 810M transistors into 214mm² die size compared to 682M transistors in 286mm2 die size, has 12MB L2 cache total (2 x 6MB, 24-way set associative) and the support 47 out of 54 new instructions on Intel SSE4. Other microarchitecture enhancements include a fast shuffle engine that improves performance of SSE4 and even Supplemental SSE3 instructions, enhanced cache line split load that improves performance on unaligned loads and optimized store and load operations, a deep power down technology that reduces power consumption, as well as Enhanced Intel Dynamic Acceleration Technology that improves single-thread operations.
 

Family	Core	Model	Clock Speed (GHz)	Core / Thread	L2 Cache	FSB (MHz)	Socket	TDP
Core 2 Extreme	Yorkfield	QX9775	3.2	4/4	12MB	1600	LGA771	150W
		QX9750	3.2	4/4	12MB	1600	LGA775	136W
		QX9650	3.0	4/4	12MB	1333	LGA775	130W
Core 2 Quad		Q9550	2.83	4/4	12MB	1333	LGA775	95W
		Q9450	2.66	4/4	12MB	1333	LGA775	95W
		Q9300	2.5	4/4	6MB	1333	LGA775	95W
Core 2 Duo E8000 Series	Wolfdale	E8500	3.16	2/2	6MB	1333	LGA775	65W
		E8400	3.0	2/2	6MB	1333	LGA775	65W
		E8300	2.83	2/2	6MB	1333	LGA775	65W
		E8200	2.66	2/2	6MB	1333	LGA775	65W
		E8190	2.66	2/2	6MB	1333	LGA775	65W
Core 2 Duo E4000 Series		TBD	TBD	2/2	3MB	1066	LGA775	65W
	Conroe	E4700	2.6	2/2	2MB	800	LGA775	65W
		E4600	2.4	2/2	2MB	800	LGA775	65W
		E4500	2.2	2/2	2MB	800	LGA775	65W
		E4400	2.0	2/2	2MB	800	LGA775	65W
		E4300	1.8	2/2	2MB	800	LGA775	65W

As a norm, Intel will usually release the extreme edition of a new range of processor prior to the introduction of the mainstream chips as a marketing stance. Today, we will be taking a look at the new Core 2 Quad Extreme QX9650 Processor, other wise known as the Yorkfield XE. Clocked at 3Ghz, the QX9650 is still sporting two separate dies like the QX6850, consisting of two 107mm die packed with 6MB of L2 cache each, essentially totalling up to a L2 cache of 12MB! The QX9650 will run at 333Mhz x 9 at a FSB of 1333MT/s, sporting a TDP of 130W just like the QX6850 once again! At the same time, These new Core 2 CPUs are also what Intel tout as the Industry's first "lead-free" processors, a huge step taken to taken by Intel to further meet EU's RoHS Directive, something which the whole semi-conductor industry is shifting to now to solve the potential build up of toxic e-waste.

The QX9650's beefier 1600MT/s brothers, the QX9770 and QX9775 will be released in Q1 '08, both clocked at 3.2Ghz, with the QX9770 based on the LGA775 socket and the QX9775 based on the LGA 771 socket. Interestingly, the QX9775 will run on Intel's upcoming Skulltrail platform, with support for dual Quads Extreme CPUs and possibly SLI support. As for the mainstream section, we will see the introduction of 2 models (Q9550 & Q9450) sporting a L2 Cache of 12MB and and a 6MB Q9300 coming up in Q1 '08 as well, along with four Dual Core C2Ds. Refer to the chart above for more information!.

Benchmarking Setup

Hardware Setup

• Intel Core 2 Extreme QX9650 (3GHz, 333MHzx9)
• Intel Core 2 Extreme QX6850 (3GHz, 333MHzx9)
• Intel Core 2 Duo E6850 (3GHz, 333MHzx9)
• Gigabyte GA-X38-DQ6 - Intel X38 Chipset
• 2 x 1GB GSkill PC8000 DDR2 Memories
• Sparkle GeForce 8800 GT (Core/Shader/Mem : 675 /1728/900Mhz)
• Verudium Power Proxima 550W
• Seagate 7200.10 400GB
• Windows Vista SP1 Beta (Build 2175)
• Nvidia Forceware 169.01

Overclocking Setup

• ASUS P5E3 Deluxe Wifi-AP/N - Intel X38 Chipset
• Corsair Dominator PC3-1800C7D 2GB Kit (1333Mhz, 7-7-7-20)

Benchmark Settings

• Crysis Demo : 1280x1024, Low Quality, CPU Benchmark.bat
• UT3 Demo : 1280x1024, Low Quality, DM-Shangrila Flyby
• ET Quake Wars v1.1 : 1280x1024, Low Quality, Custom Timedemo
• Lost Planet Extreme Condition : 1280x768, Low Quality, Cave
• CineBench R10 : Multi-CPU Rendering
• POVRay 3.7 : Chess2.pov
• WinRAR 3.7 : Benchmark Test
• VirtualDub 1.7.6 w/ DivX 6.7
• DivX Converter 6.5
• MainConcept H.264 Encoder

It will be very interesting to check out all how all the 3GHz/FSB1333 processors fare in the benchmarks. Lets see how the 4MB, 8MB and 12MB L2 cache processors perform and see how the micro-architecture enhancements on the Yorkfield will aid it in the real world benchmarks.

Gaming : Crysis, Unreal Tournament 3, Quake Wars & Lost Planet

 

 

 

I would say gaming performance improvements aren't too significant with the built-in timedemos but a larger cache and more cores still help. However, we should be seeing more significant performance delta between these processors during actual in-game play since the games these days uses fairly amount of real-world physics which is pretty CPU intensive.

CPU Intensive Apps : WinRAR, POV-Ray & CineBench

 

 

We check out some really CPU intensive and cache hogging applications like WinRAR, POVRay and CineBench. As evident, when the applications gets more CPU intensive, the performance delta between the 3GHz processors widen. More cores and cache certainly helps here.

Video Encoding : DivX Converter, H.264 Encoder, VirtualDub

DivX converter is pretty much single-threaded but still QX9650 is able to encode slightly faster. Having more cores certainly helps speed things up in H.264 Encoder and VirtualDub apps.

Updated : We re-ran VirtualDub 1.7.6.1 again with Experimental SSE4 full search enabled with the QX9650 and SSE2 for the rest of the processors. Now we are able to observe quite a boost  when we compress a MPEG file into higher compression DivX using VirtualDub 1.7.6 w/ DivX 6.7 codec. Certainly, the 2 new SSE4 instructions, PHMINPOSUW and MPSADBW on QX9650 speed up the whole encoding process. We will re-examine the applications again in greater details and hopefully we can get to test out upcoming TMPEGnc Xpress 4 that supports SSE4 on next week. More SSE4 supported applications to come along next year.

Overclocking Tests

Core 2 Extreme QX9650 is extremely stable at 4.42GHz speed using a good CPU cooler from Scythe paired up with a good overclocking board like ASUS P5E3 Deluxe.

Significant performance boost in 3DMark06 CPU and SuperPI 8MB tests. Up to 43% boost in 3DMark06 CPU performance and SuperPI 8M calculations finished almost 50 seconds faster! QX9650 is screaming to be overclocked!

Conclusion : 45nm Era Starts Now

For Gamers : Core 2 Extreme QX9650 performs quite decent in the slew of new games and will especially help when they require more physics calculations during game play. However, if you are already having a QX6850 or E6850, then it is unwise to dump them and get this processor.

For Enthusiasts : Definitely a must to get if you don't have any budget constraint and be an early adopter of this new cool processor technology. Of course, you must overclock the hell out of it to squeeze any last bit of performance out of it and the benchmark scores you'll be getting is most rewarding. Be sure to get a good cooler and a nice board to pair up this processor.

For Video Encoders : More SSE4-enabled applications are available next year so if you can afford to wait a little longer, you can actually get cheaper Wolfdale based processors like E8000 and E40000 series that similarly support SSE4.

For CPU Intensive Apps : For guys who do plenty of files zipping/unzipping like WinRAR, WinZIP etc., tons of Folding, CPU/Modeling rendering software like POVRay, 3ds Max, Lightwave look no further, QX9650 with more cache really speed things up here.