The Test

CPU: AMD Phenom 9850 (2.5GHz)
AMD Phenom 9750 (2.4GHz)
AMD Phenom 9550 (2.2GHz)
AMD Phenom 9600 (2.3GHz)
AMD Phenom 9500 (2.2GHz)
Intel Core 2 Quad Q9300 (2.50GHz/1333MHz)
Intel Core 2 Quad Q6600 (2.40GHz/1066MHz)
Intel Core 2 Duo E8200 (2.66GHz/1333MHz)
Intel Core 2 Duo E6750 (2.66GHz/1333MHz)
Motherboard: ASUS P5E3 Deluxe (X38)
MSI K9A2 Platinum (790FX)
Chipset: Intel X38
AMD 790FX
Chipset Drivers: Intel 8.1.1.1010 (Intel)
AMD Catalyst 8.3
Hard Disk: Western Digital Raptor 150GB
Memory: Corsair XMS2 DDR2-800 4-4-4-12 (1GB x 2)
Corsair XMS3 DDR3-1066 7-7-7-20 (1GB x 2)
Video Card: eVGA GeForce 8800 GT SSC
Video Drivers: NVIDIA ForceWare 169.25
Desktop Resolution: 1920 x 1200
OS: Windows Vista Ultimate 32-bit

 

Our Stance on Testing with the TLB Bug

The B2 stepping Phenoms suffer from the infamous TLB erratum which, if left unpatched, could potentially result in system instability or silent data corruption. Thus far AMD has only seen negative after effects from unpatched B2 processors in very isolated cases, described to AnandTech as the following:

1) Windows Vista 64-bit running SPEC CPU 2006
2) Xen Hypervisor running Windows XP and an unknown configuration of applications

While these are both isolated cases, they are by no means the only scenarios in which the TLB bug could rear its ugly head. All of the latest Socket-AM2+ motherboards have been updated to fix the TLB bug, at the expense of sometimes significant performance degradation. The table below summarizes our findings in our initial B3 stepping article:


  SYSMark 2007 DivX CineBench R10 3dsmax 9 WinRAR
AMD Phenom 9600 (B2 Stepping) - TLB Fix Disabled 117 74.3 fps 7396 7.20 1348 KB/s
AMD Phenom 9600 (B2 Stepping) - TLB Fix Enabled 105 72.0 fps 7031 6.47 367 KB/s
Performance Impact -10.3% -3.1% -4.9% -10.1% -72.8%

 

Since the bug could prove to be a problem in usage scenarios that haven't yet been discovered, we feel that it's best to test these B2 stepping chips with the TLB fix enabled (the default state on all motherboards now). Obviously this doesn't impact the new xx50 CPUs since they aren't plagued by the TLB erratum.

Wolfy, How Fast Art Thou? Overall System Performance - SYSMark 2007
Comments Locked

65 Comments

View All Comments

  • aju - Thursday, March 27, 2008 - link

    Ok, the fastest Phenom is still not quite as fast as the Q6600. The problem is that the cost issues is really much larger than it is made to seem in the article. The review does not really figure the total system price into the equation. The exact parts listed in the review for the test AMD system with an Phenom X4 9850 would cost $864.97. The exact parts listed for the Intel system with a Core2Quad Q6600 would cost $1273.96. Were talking about a difference of $405.99 here. For that price difference, you could forgo the 8800GT and put in 4 Radeon HD 3870s in its place and have quad CrossFire for a total of $1314.94. That MSI board supports 4 PCI Express 2.0 slots. Then we would be comparing systems at a similar price point. I wonder if the Intel system could keep up on the games then.
  • coldpower27 - Thursday, March 27, 2008 - link

    The situation would not change dramatically if Intel was changed back to DDR2-800. Intel processors don't benefit significantly if at all from the extra memroy bandwidth.

    This is a performance of the processor without limitation of other components, not a price/performance article.
  • IvanAndreevich - Thursday, March 27, 2008 - link

    Can we have a bench with the Q6600 running the same FSB and clockspeed as the Q9300? Would be an interesting comparison.
  • Schugy - Thursday, March 27, 2008 - link

    Seems like q'n'q 2.0 still isn't working as good as the specs on paper tell us.
  • Nihility - Thursday, March 27, 2008 - link

    I'm really disappointed by Intel's 45 nm Q9300.
    Doesn't overclock as well, less cache and only marginally better performance over the Q6600.
    Intel is obviously holding back because AMD can't deliver. I am not amused.
    The updated phenoms are nice and all but as an overclocker I'll have to pass on this entire generation from BOTH manufacturers. That and WOW does AMD get owned at the gaming benchmarks.
  • coldpower27 - Thursday, March 27, 2008 - link

    From an overclockers perspective yes, but what's not to like if your buying this product for stock performance, faster then the Q6600 with less cache, and much more efficient energy wise.

    The only chips that AMD and Intel sell that are geared toward overclocking in mind are AMD's Black Series, and Intel's Extreme Series..

    They have no obligation to sell you cheap overclockable processors. If they do it's just very well a bonus.
  • Nihility - Friday, March 28, 2008 - link

    From a stock perspective, it's more expensive than a Q9300 but offers marginal performance gains.
    I don't like marginal processor upgrades. It's a bad sign when a year later you get sold the same speed processors instead of something that is 50% faster. They could obviously be releasing these processors with much higher clock rates but they choose not to so they have that option to crank performance up another useless 5% if they feel like it.
    I don't like being toyed with, can you blame me?
  • nubie - Thursday, March 27, 2008 - link

    Hot Damn! Now I am torn between a 45nm Intel and a Tri/Quad AMD.

    I guess I can afford to sit back and wait, but this is just awesome news, it seems for professional apps the AMD is actually a better value (well duh), I think the Opteron line will be in high demand, and it will probably be very competitive.

    Finally, something that is nearly clock-for-clock competitive with Intel. Now if AMD can only get Dual-core models out in 45nm, then they might be able to compete on level ground in the mainstream segment.

    I just don't know which to buy, I hope that the promised AM2 compatibility will finally be here, if not there will be a lot of unhappy motherboard owners (my DFI Infinity M2 sorely needs one of these.)
  • strikeback03 - Thursday, March 27, 2008 - link

    umm, they aren't really clock-for-clock competitive, and no one said they were. Depending on pricing they may be price competitive, but the Q9300 seems to hold a decent performance advantage over the 9850 in most tests shown.
  • mczak - Thursday, March 27, 2008 - link

    What was the stepping of the Q6600 core used here? IIRC G0 had significantly lower idle power consumption, and somewhat lower load power consumption than B3.

Log in

Don't have an account? Sign up now