Tweaking Core 2 For More Performance July 16, 2006 Brandon Sandman Bell

Introduction

Page:: ( 1 / 16 )

In our testing we found that Intel’s new processors had no problem outperforming AMD’s flagship Athlon 64 FX-62, in some cases by percentages as high as 28% at 1600x1200 in gaming, while DivX conversion was almost two minutes faster for the Core 2 Extreme X6800 versus Athlon 64 FX-62. If you didn’t see our Core 2 performance article, you’ll definitely want to check it out if only to see how well single card and dual-GPU CrossFire configurations scale across a variety of resolutions.

[image]

Quest for more performance issue 1: the memory

The number one issue holding back the performance of the Core 2 systems we tested last week was without a doubt the memory subsystem, more particularly, when it comes to system memory.

As we’ve seen in the past with previous Intel processors, they tend to be very hungry when it comes to memory bandwidth. Therefore we were a little surprised when Intel decided not to provide official support for DDR2-800 in their high-end 975X chipset, as AMD has done with their new AM2 Socket Athlon 64 processors. Instead 975X is officially limited to DDR2-667. Intel’s newer P965 chipset supports DDR2-800, but it’s considered to be Intel’s chipset for the mainstream market and therefore lacks support for high-end features like ATI’s CrossFire dual-GPU technology.

We’re not exactly sure why Intel decided not to provide an updated version of 975X with DDR2-800 support. Our guess is that since an official 800MHz spec hasn’t been ratified by JEDEC they didn’t want to roll out another new chipset, but since motherboard manufacturers and even Intel themselves had to redo their own 975X motherboards earlier this year in order to support Core 2 anyway we think it would have been a perfect opportunity to add DDR2-800 support to 975X.

In other words, this means that the numbers we presented you with on Friday were with a Core 2 system running at the DDR2 memory speed of 667MHz, even though we were running memory modules perfectly capable of faster speeds.

The number one area we knew we could improve performance would be to run the DDR2 system memory at faster frequencies. We didn’t want to stop at 800MHz however, instead we were shooting for 1,066MHz! The following chart illustrates why 1,066MHz memory for Core 2 would be ideal: more bandwidth.

DDR2 Peak Memory Bandwidth
Memory	Peak Bandwidth
DDR2-533	8.5
DDR2-667	10.6
DDR2-800	12.8
DDR2-1066	17

With DDR2-1066 providing over 6GB/sec of additional bandwidth over DDR2-667, that should be more than enough to keep Core 2 and its 1,066MHz bus fed with data. In order to hit that kind of speed though, you’re going to need the right kind of memory. That’s why we decided to go with Cosair’s TWIN2X2048-6400 memory modules. To the best of our knowledge, these are the only DDR2 modules currently available on the market that provide CAS3 latency and fast memory timings at 800MHz. We wanted to test how latency affects performance, so this made the Corsair modules the perfect choice, plus we’ve heard reports that these modules are pretty good at hitting speeds of 1,066MHz as long as you aren’t too aggressive with the timings.

[image]

The motherboard

Page:: ( 2 / 16 )

Quest for more performance issue 2: the motherboard

This is not because Intel’s desktop motherboards are bad, on the contrary they’re good boards, but they’re just not designed to wring every last bit of performance out of the CPU. Instead they’re focused on delivering solid reliability, particularly when it comes to system stability, and this often directly contradicts with getting more performance out of the CPU. As a result the Core 2-ready Intel D975XBX motherboard we tested with on Friday doesn’t offer BIOS options for tweaking many common system parameters like front-side bus speeds, as this can obviously impact the stability of the system. Instead it offers just enough to let you dabble a little bit with tweaking, but not enough where you can really damage the motherboard or its components, including the Core 2 CPU inside.

If you want a motherboard built for tweaking, you’ve really got to go with one of the third-party motherboard manufacturers like ASUS, abit, DFI, Gigabyte, or MSI. All of these manufacturers build motherboards that are not only more feature-rich than Intel’s D975XBX, but they’re also highly tweakable. Inside the BIOS of these motherboards you’ll find a wealth of options for fine-tuning the performance of not only your CPU, but other system components such as the graphics card and memory.

[image]

System Setup

Page:: ( 3 / 16 )

System Setup

Intel Core 2 Extreme Edition X6800
Intel Core 2 Duo E6700

ASUS P5W DH Deluxe (975X)

2GB Corsair TWIN2X2048-6400C3

ATI Radeon X1900 XTX and Radeon X1900 CrossFire with CrossFire 6.6

Windows XP Professional with Service Pack 2

DirectX 9.0c

Benchmarks

LAME MT MP3 Encoding (MS Compiler)
DivX Converter
Windows Media Encoder 9
3DMark06
F.E.A.R. 1.05
Pacific Fighters 4.04
Quake 4 1.2
Oblivion
SiSoft Sandra 2007
Lock On: Modern Air Combat
Half-Life 2 Lost Coast
Cinebench 9.5

Notes

Over the course of the following pages, you’re going to see lots of different system configurations covering a wide range of benchmarks. The graphs are going to look a little crowded because we tested 11 different CPU/memory speed and timing combinations. Here’s what it all means:

Core 2 Extreme DDR2-1066C5: This is the Core 2 Extreme X6800 running with 1,066MHz DDR2 memory with timings of 5-5-5-15

Core 2 Extreme DDR2-800C5: This is the Core 2 Extreme X6800 running with 800MHz DDR2 memory with timings of 5-5-5-15

Core 2 Extreme DDR2-800C3: This is the Core 2 Extreme X6800 running with 800MHz DDR2 memory with timings of 3-4-3-9

Core 2 Extreme DDR2-667C5: This is the Core 2 Extreme X6800 running with 667MHz DDR2 memory with timings of 5-5-5-15

Core 2 Extreme DDR2-667C4: This is the Core 2 Extreme X6800 running with 667MHz DDR2 memory with timings of 4-4-4-12

Core 2 Duo DDR2-1066C5: This is the Core 2 Duo X6800 running with 1,066MHz DDR2 memory with timings of 5-5-5-15

Core 2 Duo DDR2-800C5: This is the Core 2 Duo X6800 running with 800MHz DDR2 memory with timings of 5-5-5-15
Core 2 Duo DDR2-800C3: This is the Core 2 Duo X6800 running with 800MHz DDR2 memory with timings of 3-4-3-9

Core 2 Duo DDR2-667C5: This is the Core 2 Duo X6800 running with 667MHz DDR2 memory with timings of 5-5-5-15

Core 2 Duo DDR2-667C4: This is the Core 2 Duo X6800 running with 667MHz DDR2 memory with timings of 4-4-4-12

Core 2 Duo DDR2-667C3: This is the Core 2 Duo X6800 running with 667MHz DDR2 memory with timings of 3-4-3-9

Media encoding

Page:: ( 4 / 16 )

Microsoft Windows Media Encoder 9

DivX Converter

LAME MT MP3 Encoding

Cinebench 9.5

Notes

Going from DDR2-667 CAS5 to DDR2-1066 CAS5 slashed eight seconds off our encode time for both Core 2 Duo and Core 2 Extreme. Adjusting memory timings also played tangible dividends, going from CAS 5 DDR2-800 to CAS3 modules cut encode time by six seconds for the Core 2 Duo, while DDR2-800’s performance improved by four seconds for Core 2 Extreme. In fact, the lower latency DDR2-800 modules were faster at encoding than the higher latency DDR2-1066 modules. We saw nice improvements in DivX conversion as well, although MP3 encoding is probably close enough to be within the margin of error.

3DMark 06

Page:: ( 5 / 16 )

Futuremark 3DMark05

Notes

While we see slight gains in 3DMark 06, the results are really nothing to write home about. This isn’t too surprising though considering the nature of 3DMark.

F.E.A.R.

Page:: ( 6 / 16 )

F.E.A.R.

Notes

In F.E.A.R. we see very slight gains when upgrading from the default DDR2-667 to faster DDR2 modules, each speed grade bump bought us about 1% in performance on average.

Quake 4

Page:: ( 7 / 16 )

Quake 4

Notes

In Quake 4 we see an improvement of 3% going from your typical stock DDR2-667 CAS5 setup to DDR2-800 CAS5 for the Core 2 Duo, and 4% for Core 2 Extreme. Going from DDR2-667 to DDR2-1066 bought us an additional 8% for the Extreme CPU and 6% for the Core 2 Duo. We also saw a nice speedup when moving from CAS 5 DDR2-667 to CAS 3 memory, 3% in the case of the Core 2 Duo while the Core 2 Extreme’s performance improved by 2% with the use of DDR2-800.

Upgrading from CAS5 DDR2-667 to CAS3 DDR2-800 improved performance by 5% for Core 2 Duo and 4% for the Core 2 Extreme.

At 1600x1200 we’re bottlenecked by the graphics card and therefore results are the same.

Lock On: Modern Air Combat

Page:: ( 8 / 16 )

Lock On: Modern Air Combat

Notes

After seeing our results in Pacific Fighters in the original Core 2 performance article, we decided to substitute Call of Duty 2 for a second flight sim, in this case, LOMAC. LOMAC is a pretty graphically-intensive game on top of being CPU-dependent, as you can see by the lower frame rates.

In this test though, we really don’t see any benefit from the faster memory, performance is basically unchanged regardless of configuration.

Pacific Fighters

Page:: ( 9 / 16 )

Pacific Fighters

Notes

Like Quake 4 we see some gains from the faster memory subsystem in Pacific Fighters. Going from DDR2-667 to DDR2-800 bought us an additional 2% in performance for both Core 2 Duo, and Core 2 Extreme. Transitioning up to DDR2-1066 from DDR2-800 also improved performance by 2% for both processors. Surprisingly, we even see slight gains at 1600x1200 for both CPUs as well.

Oblivion

Page:: ( 10 / 16 )

Oblivion

Notes

Oblivion doesn’t seem to benefit from the faster DDR2-800 or even DDR2-1066 memory. The results are pretty similar regardless of the memory settings used, whether we’re in the city filled with NPCs or out in the mountains checking out the countryside.

HL2 Lost Coast

Page:: ( 11 / 16 )

Lost Coast

Notes

Valve’s Half-Life 2 Lost Coast is another app that takes advantage of the faster DDR2 memory. Performance is up 3% for Core 2 Duo when going from DDR2-667 CAS5 to DDR2-800 CAS5, while the Extreme’s performance is up by 4%. The Core 2 Extreme and Core 2 Duo’s performance improves by another 4% when transitioning from DDR2-800 CAS5 to DDR2-1066 CAS5. The cumulative effect is an 8% improvement in performance from DDR2-667 to DDR2-1066 for Core 2 Extreme and 6% for Core 2 Duo (with its higher clock speed, it shouldn’t be too surprising to see the margin greater for Core 2 Extreme).

Just going from DDR2-667 CAS5 to DDR2-800 CAS 3 improved performance by 6% for both Core 2 CPUs.

SiSoft Sandra 2007

Page:: ( 12 / 16 )

Notes

Not surprisingly the only SiSoft Sandra test that shows a clear performance benefit from the faster DDR2 memory are the memory bandwidth tests. Sandra’s memory tests actually show double-digit gains when going from DDR2-667 to DDR2-1066.

CrossFire w/Oblivion

Page:: ( 13 / 16 )

Oblivion

Notes

When running two Radeon X1900 cards in CrossFire mode, Oblivion’s performance still doesn’t scale as you add faster memory based on our results.

CrossFire w/Shooters

Page:: ( 14 / 16 )

F.E.A.R.

Notes

When running CrossFire at 1600x1200 with 4xAA/8xAF, we saw no real tangible gains from the faster memory modules with the first-person shooters we typically test with.

CrossFire w/Flight simulators

Page:: ( 15 / 16 )

Pacific Fighters

Conclusion

Page:: ( 16 / 16 )

We saw nice gains in our video encoding and DivX conversion tests; and we’re only dealing with a 165MB 720p video, the time saved would be even greater for a larger file.

In our gaming tests, the results were a little more mixed. In games like Quake 4, Pacific Fighters, and Half-Life 2 Lost Coast we saw gains ranging anywhere from 2-4% when going from DDR2-667 to DDR2-800, with the cumulative effect of going from DDR2-667 to DDR2-1066 in the high single-digits. In F.E.A.R. the DDR2-800 barely registered more than a 1% increase over DDR2-667, while Oblivion and Lock On: Modern Air Combat saw no performance improvement at all.

Based on the data provided in this article you can hopefully plan your upgrade accordingly if you’re tempted to pick up a Core 2 processor when they go on sale later this month. If you’re into video encoding/conversion the faster memory could definitely come in handy, especially if you’re dealing with larger files. Our tests also revealed that lower memory timings can also reap significant performance dividends.

For gamers who are into overclocking, picking up faster memory modules would probably be a good idea if you can afford it. While you won’t see performance gains at 1600x1200 with the image quality settings cranked up and the AA/AF turned on (with the exception of Pacific Fighters), buying faster memory should come in handy when you’re overclocking. Typically the premium memory modules scale to higher clock frequencies than the generic modules, and with lower timings as well. In theory, as you hit higher clock speeds with your overclock, memory bandwidth is going to be increasingly important as well. And besides, clearly the third-party 975X motherboards can handle the higher memory speeds (even though technically it is considered overclocking) and there’s a wealth of DDR2-800 and DDR2-1066 modules out there from high-end memory manufacturers like Corsair and OCZ, and they’re selling for lower prices everyday.

If you’re a more casual gamer and you don’t plan to overclock your system, clearly you should stick to regular DDR2-667, as the performance benefits we mentioned above won’t apply to you once you crank up the graphics settings in your games, and you can save a little money in the process that you can put towards something else.

Taking Core 2 to new performance heights isn’t tough if you’ve got the right components. In fact, based on the overclocking reports that are out there so far, Core 2 seems like Intel was a little conservative on more than just the memory speeds. We’ll be exploring that next…