7750 let us see what happens!

[Archer0915]

hold on i will see if I can pull the specs on the slower of the 2 3.3 benches

 

[Archer0915]

Rick I ran the test at 3.2 as that is what I will be running at.

 

[Archer0915]

View attachment 71065

I have a couple more marks to post I will then do a little study of the data and post my findings.

I will hopefully finish this over the weekend and post a final pic. If there are any other questions (Rick) from anyone about any of these tests I will be glad to answer.

After I post the timings total results I will post a summary of what speeds I attained and at what voltages. I will not post anymore new marks unless someone wants some for comparison as Rick did. I will not post over 3.3Ghz as this experiment was on air and at 3.4+ and the voltage required to attain those speeds I do not feel comfortable posting results and will not recommend clocking at those speeds and voltages without water.

 

[KTE]

View attachment 71065

I have a couple more marks to post I will then do a little study of the data and post my findings.

Nice little endeavor here

I'm not sure if you missed these testings, but for K10 65nm, Ganged vs. Unganged, 533 vs. 1066, and, K8 vs K10 Cache/Mem differences have been thoroughly examined in these excellent articles.
http://ixbtlabs.com/articles3/cpu/amd-phenom-x4-9850-ganged-unganged-p1.html
http://ixbtlabs.com/articles3/cpu/amd-phenom-x4-9850-ddr2-533-p1.html
http://ixbtlabs.com/articles3/cpu/rmma-phenom.html

K10 was a big boost in many bus, link, interconnect and latency limitations of K8.

Are you able to run this test? http://www.xtremesystems.org/forums/showpost.php?p=3631328&postcount=1

I cannot vouch for its accuracy but the components and datapaths its polling limitations of is the exact aspect which gives any CPU working power for each type of code. Its just removing software/driver/operating system limitations and trying to poll the real power of each CPU.

 

[Archer0915]

Nice little endeavor here

I'm not sure if you missed these testings, but for K10 65nm, Ganged vs. Unganged, 533 vs. 1066, and, K8 vs K10 Cache/Mem differences have been thoroughly examined in these excellent articles.
http://ixbtlabs.com/articles3/cpu/amd-phenom-x4-9850-ganged-unganged-p1.html
http://ixbtlabs.com/articles3/cpu/amd-phenom-x4-9850-ddr2-533-p1.html
http://ixbtlabs.com/articles3/cpu/rmma-phenom.html

K10 was a big boost in many bus, link, interconnect and latency limitations of K8.

Are you able to run this test? http://www.xtremesystems.org/forums/showpost.php?p=3631328&postcount=1

I cannot vouch for its accuracy but the components and datapaths its polling limitations of is the exact aspect which gives any CPU working power for each type of code. Its just removing software/driver/operating system limitations and trying to poll the real power of each CPU.

All of my current tests are @ 3.0 if you want 3.2 or possibly 3.3 let me know.

 

[QuietIce]

Just for grins I'd like to see just a couple of tests, maybe one Sandra and one encoding, run with the CPU at 228x14 (if it'd run that).

I'm guessing your low NB speed is limiting the RAM performance at the faster speed. Too bad you don't have a pair of DDR2-800 sticks sitting around ...

 

[KTE]

All of my current tests are @ 3.0 if you want 3.2 or possibly 3.3 let me know.

Thanks for the test data.

I'll have to look into it later once I can get CodeAnalyst and Vtune setup to test the application. I have results from Deneb (x4), Agena (x4), Brisbane (x2) and Penryn (x2/x4), but they're all at 2.50GHz.

 

[Archer0915]

WOW!!!!! just got a 23+ min difference at 533-4-4-4 to 667-5-5-5

Will do ICE

I will do the same DivX encode as before.

 

[Archer0915]

Ice of corse the top # is per your request the other 3.2 I threw in for comparison it is lower mem speed 533 4-4-4 timings. As for the encode it might have to wait as I believe the old lady might have marching orders for me. I will get it done asap.

 

[Archer0915]

The timings are titer to make the marks as close as I can.



The DivX was 2m 57.4s

If you like I can post a score @The 533 mem sp 4-4-4 @ 3.2 Ghz as my others are at 3.0 Ghz and I think my titer timings at the same proc speed will win out. I really think the only way to actually see an advantage at Xmem speed is to have is to either have enough data to cause a bottle neck at the lower clocked faster timed memory or equal or better timed faster memory.

This as I have said before is system specific but as I also said I remember when you might if you were lucky have a clock gen jumper on the mobo and the only way to get more performance was to buy 60ns ram and titen the timings.

Hope this data serves you well.

 

[Kuroimaho]

Check the latency, tighter timing with lower clocks can have the same lat as higher clocks with looser timings.

 

[Archer0915]

Check the latency, tighter timing with lower clocks can have the same lat as higher clocks with looser timings.

Already did that I have not done a full analysis of all of the data.

As you can see a lot of the synth is inconclusive and you as well as I know that you can pump #'s through just to get bandwidth usage however if there is any amount of complexity to the data calculations timings come into play and they can play a big part as they impose a WS on the CPU. Slack timings=lower efficiency. on the other end of the scale if you are moving too much data the pipe becomes saturated and timings ain't gonna help this data is above.

 

[QuietIce]

The timings are titer to make the marks as close as I can.
<snip pic>

The DivX was 2m 57.4s

If you like I can post a score @the 533 mem sp 4-4-4 @ 3.2 Ghz as my others are at 3.0 Ghz and I think my titer timings at the same proc speed will win out. I really think the only way to actually see an advantage at Xmem speed is to have is to either have enough data to cause a bottle neck at the lower clocked faster timed memory or equal or better timed faster memory.

This as I have said before is system specific but as I also said I remember when you might if you were lucky have a clock gen jumper on the mobo and the only way to get more performance was to buy 60ns ram and titen the timings.

Hope this data serves you well.

Thanks for the data!

I thought the other was done at 3.2 GHz? - my bad. If the other data is 3.0 GHz then 250x12 (can you run a 250 clock?) or 231x13 would be good for a comparison. I'm just trying to see the difference at a higher clock speed but the same CPU speed. I know the higher clock raises RAM speed as well, but that's kinda' the point - as long as the RAM sticks are the same. Sandra numbers from this test show an obvious jump in bandwidth (as they should) but it's the CPU performance using various RAM timings/speed - especially "real world" - that I'm trying to figure out. The theory I'm trying to prove/disprove is that NB speed was the limiting factor in your first run tests and why the 533 RAM speed was better or equal to 667 RAM speed performance ...

 

[Archer0915]

Thanks for the data!

I thought the other was done at 3.2 GHz? - my bad. If the other data is 3.0 GHz then 250x12 (can you run a 250 clock?) or 231x13 would be good for a comparison. I'm just trying to see the difference at a higher clock speed but the same CPU speed. I know the higher clock raises RAM speed as well, but that's kinda' the point - as long as the RAM sticks are the same. Sandra numbers from this test show an obvious jump in bandwidth (as they should) but it's the CPU performance using various RAM timings/speed - especially "real world" - that I'm trying to figure out. The theory I'm trying to prove/disprove is that NB speed was the limiting factor in your first run tests and why the 533 RAM speed was better or equal to 667 RAM speed performance ...

Did all of my stability testing before installation of new proc. 238 is highest completely stable I will go with the 231.

As to you wondering about the timing v/s clock speed look at it this way: The timings and the clock speed have an indirect correlation to each other but a direct correlation to total system performance. Let us say you are running a 667 Mhz colum strobe of 5 is not bad however so the CPU sends a fetch every 5 cycles of the FSB so 200/5=40x1Mhz=4000000 now we move to cas 4=5000000 now realize that this is independent of the memory and is controlled by the cpu through bios (limited by memory capability) however and that you also have row and column to row. Now if we take the cpus ability to fetch data that is not in cache and write that data back that is where latency comes into play as long as your throughput is >= mem speed then you will always be faster moreover if mem speed and or timings are too slow total system performance is hurt.

 

[Archer0915]

3m3.3s not bad for encode








I threw in the ddr3 for comparison

If you look at the rs780 in the final benchie that is also one of mine now it shows a big lead in throughput but 4.7 sec difference in encode time yes it was faster but in the end unless you only do work with files that can take advantage of that bandwidth you are left with no real gain.

 

[QuietIce]

IMO - anything ±2% is no difference as that falls into the statistical error range for most tests (unless you run a LOT of them to create a big dataset). I'll have to do some playing with my rigs to see what differences, if any, there are for Crunching. I won't be using a PhenTwo, though - I only have PhenOne's for now. Maybe one of the Folders will be interested enough to donate some folding time for testing as well ...

 

[Archer0915]

Did all of my stability testing before installation of new proc. 238 is highest completely stable I will go with the 231.

As to you wondering about the timing v/s clock speed look at it this way: The timings and the clock speed have an indirect correlation to each other but a direct correlation to total system performance. Let us say you are running a 667 Mhz colum strobe of 5 is not bad however so the CPU sends a fetch every 5 cycles of the FSB so 200/5=40x1Mhz=4000000 now we move to cas 4=5000000 now realize that this is independent of the memory and is controlled by the cpu through bios (limited by memory capability) however and that you also have row and column to row. Now if we take the cpus ability to fetch data that is not in cache and write that data back that is where latency comes into play as long as your throughput is >= mem speed then you will always be faster moreover if mem speed and or timings are too slow total system performance is hurt.

FSB 231 770 mem speed cas4 231/5=46.2x1Mhz=46200000 column address strobes per sec= ~15% increase so the path to faster memory is to try to get 1-1 =1-2 due to reported speed ddr/2

I feel stupid the secret to finding your zenith is first find your fastest timings then max fsb. Set timings at titest possible under clock ram then ramp it back up with fsb until max fab is reached.

EDIT: going to try 7xht multix266=1862ht 533/3=177666666 strobes going to try 3-4-4-12 starting at mem 233?

EDIT did not boot at 266 went to 233 results inconclusive ??? did I miss something am I just getting old and my mind failing me???

Did I miscalculateDoes it matter I am running out of energy must sleep mus........

My head hurts timed ram @667 to same as 533 and results inconclusive HMMMMM? I think that it is possible that at 3.0Ghz an increase in memory speed does no good next step try 400 timings at 533 and see what happens.

Evidently there is no need for high speed ram for a dual unless you bench for a living I have tried several faster settings all results inconclusive. I will now begin an examination of how quads access memory and the actual benefits to a quad we all know there are definite benefits and no apparent drawbacks I am more interested in the architecture I might just buy one and do a full workup and test a little theory I have.

Thanks for the input Guy's

 

[Archer0915]

Haed dont hurt now

started 533 tests titest timings Highest 3d score yet with ~3.0 Ghz.






I will do a little more testing and enter data to sheet when finished we can draw conclusions.

Edit: let me redo that but the timings are not completely changed. and I misread data.

 

[Archer0915]

OK tightening timings even more than last post was inconsequential. I guess this cpu is at Zenith

Now that is how I define a successful OC and tweak session any questions

 

[aries2110]

The Phenom II X3 numbers are getting exciting though. They perform exactly like they should; like a tri core Phenom II. Which is awesome

I'm thinking a 710 or a 720 BE and an HD4870 will be in the plans soon