Test Setup

Gigabyte GA-965P-S3 / GA-965GM-S2 Testbed
Processor: Intel Core 2 Duo E6300
Dual Core, 1.86GHz, 2MB Unified Cache
1066FSB, 7x Multiplier
CPU Voltage: 1.3250V
Cooling: Zalman 9500 Air Cooling
Power Supply: Corsair CMPSU-620HX 620W
Memory: OCZ Flex XLC PC2-6400 (2x1GB) (ProMOS Memory Chips)
Memory Settings: 3-4-4-9 1.90V
Video Cards: MSI X1950XT
Video Drivers: ATI Catalyst 7.2
Hard Drive: Western Digital 74GB 10,000RPM SATA 16MB Buffer
Seagate Barracuda 7200.10 500GB SATA 16MB Buffer
Optical Drives: Plextor PX-760A, PX-B900A
Case: Cooler Master CM Stacker 830
Operating System: Windows Vista Home Premium 32-bit

Our two boards are from Gigabyte and each one offers similar features, BIOS options, and board level components. The street price for each board is around $105 at this time. We would like to mention that Foxconn and ASUS have matching sibling pairs but our Gigabyte GA-965GM-S2 was already in testing so it was chosen in the interest of saving time. Our Foxconn G9657MA-8KS2H offers a wider selection of BIOS options including the ability to change several voltage options, so it may perform slightly better overall but not enough to dramatically alter our findings. We set up each board with the exact same BIOS settings and found that the Gigabyte GA-965P-S3 ran slightly tighter memory sub-timings. We thought about using MemSet to adjust the sub-timings but figured in the end that most users would not utilize this utility to change timings on boards in the price sector.

A 2GB memory configuration is standard in our entry level Vista test beds as we highly recommend this amount of memory now. Our choice of mid-range OCZ Flex XLC PC-6400 memory represents an excellent balance of price and performance and offers a very wide range of memory settings at stock voltages. Our memory timings are determined by finding the best memory bandwidth via MemTest 86 along with test application results for each board. We only optimize the four main memory settings (CAS, tRCD, tRP, and tRAS) with sub-timings remaining at Auto settings. This will benefit the P965 platform slightly as we found the memory sub-timings were generally tighter on this board than on the G965 variant.

We are utilizing an MSI X1950 XTX video card to prevent our 1280x1024 resolutions from becoming completely GPU bound for our motherboard test results. All of our tests are run in an enclosed case with a dual optical/hard drive setup to reflect a moderately loaded system platform. Windows Vista is fully updated and we load a clean drive image for each system to ensure driver conflicts are kept to a minimum.

We are not going to address overclocking capabilities of the two boards in today's test results. However, the G965 is limited to 350FSB due to the X3000 graphics core and shared memory pipeline. The realistic overclock levels we have seen on our best G965 samples have been around 325FSB. Based upon this, if overclocking is a priority then we definitely recommend the P965 based boards in this price sector.

Synthetic Memory Performance

Click to enlarge

In our Sandra memory test, we see the P965 based board scoring slightly higher in both the buffered and unbuffered scores due to tighter sub-timing settings for the memory. We will see this slight performance improvement in the unbuffered scores generally give the P965 board an advantage in the memory sensitive benchmarks. When we equalize the settings, the P965 still scored about 40 points better on each result - less than a 1% advantage.

We know Intel made a few concessions with the memory controller due to the on-board X3000 graphics core but this primarily affects overclocking capabilities. With additional BIOS tuning from Gigabyte in the memory and MCH timing settings we are sure the end results could be even. The results with SuperPi 1.5 also reflect a slight but unnoticeable difference between the two boards.

Index General System Performance


View All Comments

  • 8steve8 - Tuesday, March 20, 2007 - link

    nice to look at the g965, long overdue.

    but u didnt even touch on integrated graphics.

    lots of people out there dont play many 3d games.

    you should have compared performance between g965 onboard with g965 discrete with p965 discrete

    power/heat numbers would be interesting, then we could see how efficient the integrated x3000 is...

  • xsilver - Saturday, March 17, 2007 - link

    are there ANY G965 boards that DO overclock well? Reply
  • JarredWalton - Saturday, March 17, 2007 - link

    Meaning beyond 325MHz bus speeds? No, because the X3000 IGP really limits the chipset's maximum speed. Where on the P965 you're able to overclock the chipset to 500 MHz and above, the IGP can really only take about a 25% overclock - certainly not more than 33% or so. You might be able to hit 350 MHz on some boards, but that's about it as far as I'm aware. Other IGP chipsets on the other hand... we'll have to see Gary's mATX roundup for that information. Reply
  • Treripica - Saturday, March 17, 2007 - link

    I'd like to echo MrNeutrino's sentiment. How massive of a mATX roundup can we look forward to in the near future? Reply
  • kobymu - Saturday, March 17, 2007 - link


    These systems represent about 90% of the personal computers sold in the North American market each year and just a little more worldwide.

    I just wanted to say thanks for saying that.

    While enthusiast discussion is important (I'm PC enthusiast myself), it is always more important to keep a wider, proportional point of view of the industry as a whole, and that sentence has achieved that goal, so when the enthusiasts start discussing higher-end PC components it can take a more practical, sensible approach, and maybe, hopefully, with time we will see a decrease in the "OMG company X is going down!11" department in particular and in radical fanboism in general.
  • sdsdv10 - Saturday, March 17, 2007 - link


    when the enthusiasts start discussing higher-end PC components it can take a more practical, sensible approach

    Do you really expect enthusiasts to be either practial or sensible...

    That would be kind of an oxy-moron!
  • MrNeutrino - Saturday, March 17, 2007 - link

    Thanks for taking the time to post the mATX performance update, Gary!

    My performance concerns based on architectural scrutiny of how memory bandwidth is shared between two hungry processors in an IGP chipset vs. its ATX counterpart, were always nagging me as I started looking for a mATX build. Add to that the lack of always reliable reviews - if at all - on the web, and you have the perfect recipe for a burning desire to be sure - and soon - whether almost a thousand some odd dollars of investment in a quality mATX vs. ATX (read: G965 vs. P965) will be worth the money.

    At least in my case, the lingering questions were mostly extinguished by your forum reply with quantifiable benchmark data, after the 690G review. What little concern remained (more like intrigue), has been put to rest with these benchmarks and additional comparison data in this article.

    Impressive, that there is this little difference between the two chipset / system architecture variants. I suppose then, thanks goes in large part to the hunk of cache on the C2D (compared to CPUs of just a few years back), better predictors / prefetchers, shorter pipelines and a myriad of other uArch improvements.

    Anyway, techincal topics aside, all I can say is, HURRAY!!

    I'm just happy that this is concrete, published proof that us SFF / mATX fans can have our cake and eat it too!

    Now if only we all could also get those April C2D price drops soon enough... :)
  • Renoir - Saturday, March 17, 2007 - link

    Agreed this was a good comparison of the performance between G965 and P965 when BOTH are using a discrete card. As for the question of how sharing memory bandwidth between IGP and CPU compares to using a discrete card, it has only been touched upon briefly by Gary in the forum thread you mentioned. More detailed info on this aspect would be much appreciated in the upcoming roundup (already planned?). Bit-Tech.net did just that with the 690G but they used XP so I'm very interested in the results when using the more bandwidth hungry Aero in Vista. Reply

Log in

Don't have an account? Sign up now