Answered - HD 3850 Performance -What can I expect?

[Gump]

I have trended data from HD 3850 3DMark 06 Scores. I wanted to get this out there for those wondering if it is worth getting the 3850 as an upgrade.

I would be interested in 3850 Owners posting in their individual CPU, SM2 and SM3 scores so I could have the data come to me instead of using ORBs filters. If anyone has 3850 scores in 06 please post and Specify AGP / PCI-E -- 256MB or 512MB mem -- GPU and Mem clocks and your CPU along with the CPU, SM2 and SM3 scores.
I can use them to refine my graphs.

The numbers below are taken directly from ORB and individual submissions. They are mixed - AGP and PCI-E - not really a big deal as this is a mid-range card. I did take measures to ensure no crossfire was included (the delta of SM2 vs SM3 is much higher in a crossfire configuration). I also eliminated data points which varied too far from the average. From the data I gathered, the AGP version holds it own against its PCI-E relatives. The curves are representative of an overclocked 3850. If you have run 3Dmark06 you should be able to apply your CPU score to this graph to estimate what performance you can expect if you had an Overclocked 3850. I used 3DMark 06 as a Baseline because it has a large and readily available database. It also separated out the CPU Score from the Graphic Score allowing me to show how much CPU power is needed before the 3850 hits a ceiling. Thanks to 15th Warlock for getting me a crucial data point to show how far AGP can go. Thanks to Tom's Hardware for dissecting 3Dmark 06 so I could get the formula used to calculate the 3DMarks. Enjoy, and click me a "Thanks" if you find it useful.

 

[Silverel]

Very nice!

Now to force you to work on all the rest of the cards... heheheh...

 

[ghost101]

Nice, but you need a lot more datapoints. Also, the higher end is possibly flattening out due to people switching to quadcores. It also explains the dip where people start using stock q6600s. Dual cores clocked high enough to get a similar score, also benefit the sm2/sm3 scores and therefore actually get a higher score overall.

What you need to do is limit the data to dual cores and use core 2 duo clock speeds as the x axis or do the same for core 2 quads.

edit: an easier way to do this would be to use a controlled setup. So basically, a pc with a hd3850 and core 2 quad and simply vary the clock speed of the cpu and take relevant benches. Far less innacuracy and dont need as many datapoints.

 

[Gump]

Nice, but you need a lot more datapoints. Also, the higher end is possibly flattening out due to people switching to quadcores. It also explains the dip where people start using stock q6600s. Dual cores clocked high enough to get a similar score, also benefit the sm2/sm3 scores and therefore actually get a higher score overall.

What you need to do is limit the data to dual cores and use core 2 duo clock speeds as the x axis or do the same for core 2 quads.

edit: an easier way to do this would be to use a controlled setup. So basically, a pc with a hd3850 and core 2 quad and simply vary the clock speed of the cpu and take relevant benches. Far less innacuracy and dont need as many datapoints.

Good points.
I have tried to filter down using ORB but the filters don't seem to work properly for me (most of the time) when I try multiple filters.
I did have another 30 to 40 data points, but filtered them out due too much deviation from the curve. The curve that best approximates the data points was a 4th Order Polynomial. In a couple earlier versions I have floating around I had used 3rd order polynomial to generate the trendline. I ran through several edits of the data before finally deciding this was the most representative of the data I had.

I would be interested in 3850 Owners posting in their individual CPU, SM2 and SM3 scores so I could have the data come to me instead of using ORBs filters. If anyone has 3850 scores in 06 please post and Specify AGP / PCI-E -- 256MB or 512MB mem -- GPU and Mem clocks and your CPU along with the CPU, SM2 and SM3 scores.
I can use them to refine my graphs.

Thanks

 

[Gump]

Ghost101
I extracted the 3Dmark Scores for the Top Dual Core w/ HD 3850 from the 3DMark 06 3K score thread. Two of the scores I was unable to get a breakdown of their graphic vs CPU so had to skip leaving only 9 data sets.
The Deviation from my trend line was within 10% for all 3DMark 06 scores (most were higher but a couple were under)
The 11688 fell out at 8% deviation.
Another 3 fell out at 5% deviation.
The other 5 data points were within 5% deviation.
The graph was pretty ugly, but held to reasonable error percentages.

Send more scores!

 

[BrooksyX]

Ill post my scores later tonight. I have an Athlon X2 5000+ Black Edition and a Visiontek 3850 512mb.

 

[Gump]

Heres the latest with over 250 data points from 3DMark06.
Find your CPU score on the bottom and look up to the green line to estimate 3DMark 06 Score with a 3850.

 

[Gump]

Final

Final Version showing approximately where each CPU would fall at stock speed.
CPU Scores taken from Tom's Hardware CPU Charts.