SETI2 benchmarking WU & average AR of WUs

[Assimilator1]

Ever since I've been running SETI2 I've found it hard to compare different rigs with each other due to the longtime it takes RAC to level out (~2wks) & the varying Angle Ranges of WUs which isn't starightforward to find. RAC is useless for quick comparisions, like when you've just switched client versions, you have temporary use of another rig or want to compare many rigs.

[edit]Go to the 'To summarise' line if you want to skip the explanations,history & complaints

For me this took out some of the fun of SETI, in the old days of SETI classic (as some of us remember) with SETIQ you could open up your browser to SETIQ, checkout what WU times your rigs were getting & what AR WUs they were chewing on. Quick,simple & accurate .

Then it was also known that the average & common AR WU was 0.417 & from that spawned the TLC benchmarking WU & benchmarking tables & others giving accurate comparisions between large numbers of different machines.
Their are many TA names in there too btw.

Yes in your BOINC stats you can check out WU times for each machine (done in seconds :roll: ) but you can't see the AR along with it (you have to go through 2 further links to find out) so you can't make easy comparisions.

Anyway as I mentioned this has been bothering me for ages so I planned to do something about it, so about 2yrs on from when I 1st asked RattleDagger how to capture a WU I believe I finally have enough info on WU ARs. Inccidently I have searched around to see if this info was already out there but all I found was an assumption that the average AR was still 0.417, I wasn't going to assume that is still right

So today & ~3wks ago I trawled through 200 WUs that my Q6600 had done & copied & pasted the AR of each WU into excel (god that was tedious! lol). I know their are millions of WUs out their but I don't have time to look through all those , & I felt that a mix of 200WUs 3 weeks apart would give a fair idea of what's out their.If you have data to show otherwise then send it to me & I'll add it in to the average.

I averaged the AR of all 200 WUs & ended up with a figure of 0.502711 which is almost totally useless as looking through those 200+20 WUs none of them had an AR of ~0.5! Of the 200 that I'd recorded 66 were at 0.39 AR or close (0.38-0.4), 46 at 0.44 or close (0.415-0.45), the rest above or below those groups.

To summarise, in my sample of 200 WUs the 0.39 AR WU was the most common, it's also pretty close to the average of the 2 most common ARs WU groups (0.39 & 0.44) which averaged is ironically close to 0.417 (though I haven't worked that out accurately).

So therefore I suggest that the new benchmark WU for SETI2 should be a WU of 0.39 AR, this gives the additional advantage that most people will simply be able to look through their completed WUs under 'Tasks' & get a very good idea of what times their rigs are doing them in, look for a WU with 73.xx credits. Of course that won't be as accurate as running a benchmark WU but it'll be very close (see Smokes results about that down below).

If I get enough results here from TA members (& others?) then I would like to create a benchmarking table like TLC did, our very own Thunder did 8yrs ago & SETI Spy author Roelof's Unofficial Top 200 Team Lamb Chop Benchmark Results table. (many TA names near the top there , last updated 9/03 before he became too ill RIP).

[edit] Preliminary table available here

Before I grab the appropriate WU I'll add a few results myself.

My rig Q6600 @3GHz, DDR2 @788 4-4-4-15 2T, 1 WU
AR 0.393746 (~73 creds), CPU time ~4412s = 1hr 13.5mins
(running alongside 3 other cores doing DC) AK v8 SSE3 client.

My 2nd rig Sempron 3100 @2.5GHz, DDR @454 3-3-3
AR 0.391117 (73.58 creds), CPU time 8914s = 2hrs 29mins
AK v8 SSE3 client.

Nick's rig E4500 (2.21GHz), 1WU
AR 0.393110 (~73 creds), CPU time ~5220s = 1hr 27mins (other core also doing SETI) Ak v8 SSE3 client.

Ali's rig XP2600 (1.92 GHz), 512KB L2, DDR333 2.5-2-2
AR 0.390633 (~74 creds), CPU time ~23,623 = 6hrs ~34mins :Q (wow I knew that'd be slow but still! lol, no 73 cred results atm )

None of the rest of my mini-fleet have completed normal AR WUs (or any) recently .
Both the Core 2 rigs are running the optimised v8 client.[edit] Darn, but the SSE3 version not the SSSE3 one!

To find ARs, go to your BOINC SETI stats>Computers>tasks(of chosen PC)>Task ID ,on completed WUs only.[edit] Or as Bryan & Smoke have done forget the AR & just look for ~73 credit WUs , that's a lot quicker as you don't have to look into each WU. Though I was hoping for some additional feedback on this to re-confirm that ~73 cred WUs are always WUs with an AR of ~0.39.*

I'm gonna see if I can find the info RD gave me about finding out how to 'grab' a WU.

Comments? questions?

*Credit vs WU AR, times for my Q6600 @3GHz, AKv8 SSE3 app

73.82 cred WU, AR 0.389979 (4165s)
73.82 cred WU, AR 0.390045 (3965s)
73.82 cred WU, AR 0.390062 (4342s)
73.71 cred WU, AR 0.390654 (4382s)
73.61 cred WU, AR 0.390944 (4376s)
73.59 cred WU, AR 0.391073 (3927s)
73.50 cred WU, AR 0.391539 (3924s)
73.50 cred WU, AR 0.391472 (4513s) #
73.40 cred WU, AR 0.392135 (3911s)
73.33 cred WU, AR 0.392577 (3789s) #
73.32 cred WU, AR 0.392662 (4478s)
73.27 cred WU, AR 0.392925 (4392s)
73.23 cred WU, AR 0.393219 (4500s)
72.78 cred WU, AR 0.394307 (4391s)
72.75 cred WU, AR 0.394542 (4046s)

# Biggest time difference ~720s (12mins) ,average those 2 (4151) & it's a variance of about +/- 8% ,hmm not good! Still it's better than RAC.
We definitley need a benchmark WU for accurate scoring though.

 

[bryanW1995]

2843 seconds ~ 73.27 credit x3350 @ 3.6, memory at ddr2 1080 xp sp 3
3253 seconds ~ 73.51 credit p9450 @ 3.264 memory at ddr2 816 4-4-4-12 xp sp 2
3961 seconds ~ 73.61 credit q6600 @ 3.501 memory ddr2 782 5-5-5-15 xp sp 3
5571 seconds ~ 73.27 credit t7250 @ 2.00 memory ddr2 667 5-5-5-15 vista home premium sp 1
11852 seconds ~ 73.73 credit x2 4000+ @ 2.1 memory ddr2 667 5-5-5-15 1t xp sp3

the 4 core 2 based cpus are running the v8.0 optimized app. the x2 is running the kwsn v2.4 old optimized app. I would update to the new one but it's at my old job and I don't really want them to just delete it off the computer

 

[Smoke]

Mate, you've done a good bit of work there and a tip of my cap to ya.

In Classic SETI, we could run that "Test WU" over and over until we got the fastest result and then submit it to the Ars Record Page. That was a lot of fun, especially when you could get the fastest time or at least be near the top of the list. I recall once upon a time I held the top spot for about a week.

My Q6600 is also running at 3.00GHz. Below is one page of my completed WUs. The page I selected had the credits closest to 73 that I could find.

The closest credit I could find was 73.16 (1 hour, 0.6 minutes). Notice below I have six WUs that earned a credit of 73.22 and they ranged from 0 hour, 59.6 minutes to 1 hour, 1.8 minutes. Now that's a spread of only 2.2 minutes which is pretty damn close but there was a difference.

Credit - Seconds - Hr: - Mn:
40.77 - 2,275.52 - 0 - 37.9
50.68 - 2,917.95 - 0 - 48.6
50.72 - 2,940.22 - 0 - 49.0
50.73 - 2,935.47 - 0 - 48.9
50.74 - 2,913.30 - 0 - 48.6
50.74 - 2,911.50 - 0 - 48.5
50.74 - 2,905.02 - 0 - 48.4
52.97 - 3,015.44 - 0 - 50.3
52.97 - 3,106.61 - 0 - 51.8
63.98 - 4,058.61 - 1 - 07.6
73.16 - 3,637.36 - 1 - 00.6
73.17 - 3,680.42 - 1 - 01.3
73.21 - 3,759.56 - 1 - 02.7
73.22 - 3,707.63 - 1 - 01.8
73.22 - 3,633.31 - 1 - 00.6
73.22 - 3,612.48 - 1 - 00.2
73.22 - 3,611.28 - 1 - 00.2
73.22 - 3,632.17 - 1 - 00.5
73.22 - 3,578.92 - 0 - 59.6
73.25 - 3,669.19 - 1 - 01.2



The biggest problem I see with your plan is finding a WU with exactly 73 credits.

All in all, it's going to be hard finding a yard-stick (meter-stick ) that beats the RAC that's provided by BOINC.

BTW, I created a few threads that listed our (the TeAm's) fastest computers running Rosetta. It required a lot of work too. :roll: I had to search each of our "users" for their fastest "host". I also included in the report the type of processor for that host. That report showed not only the fastest systems but also who had the fastest system with the same CPU. This, of course, only included the "users" that allowed their details to be seen by the public. That meant that quite a few of our fastest systems were not in the report.

 

[Assimilator1]

Thanks guys

About looking at the AR, I was hoping for some additional feedback on this to re-confirm that ~73 cred WUs are always WUs with an AR of ~0.39.
If each poster could check at least 2 ~73 credit WUs to see if they are ~0.39 AR then that would add assurance that that observation is consistently true as much as possible.

Bryan
Lol, its always intriged me to see how long 'Captain-less' ships will crunch for, Darkone had about a dozen crunching for him until SETI classic was closed, which was about 5mths after he left that job!:laugh:

Man I've just realised how fast your X3350 is!:Q sweet! ,is that the Penryn core CPU?

Greg
Thanks mate , I hope this will stoke some extra chat & interest about SETI too.
Beating RAC will be a sinch (spelling??), so far it seems going by ~73 credit WUs is already more accurate & quicker than using RAC for individual PC speeds.

Btw I never said I needed a WU of exactly 73 creds, all I need to do is grab a WU with an AR 0.39xxx which will thus be approximatly 73 creds, as long as everyone who wants a precise time runs the same WU it doesn't really matter. Those who don't can just record the time of any WU with 73.xx credits.

Oh & I think it was Roelofs TLC Top 200 (linked above) that you were top in for a while .

Hey how come your Q6600 is so much faster than mine?, I wonder if DPAD is hitting the available bandwidth more than running other SETIs?....
What RAM timings & FSB are you using on that?, my FSB is 1333 MHz.

 

[bryanW1995]

fsb and trd seem to be pretty important for seti, at least with the v8.0 science app.

@ assimilator: my x3350 (xeon version of q9450) is in fact a 45nm quad. it's pretty cool to run seti on 3 cores, f@h on one core (for the gpu2 client) and still play any game I want at max settings. it might be a little overkill, but it was fun building it and price was very reasonable so I have no complaints

what is the AR that you keep referring to?

 

[Assimilator1]

Angle Range of the WU, IIRC correctly that's basically the angle the telescope was recording from (not pointed as it's fixed).
[edit, link removed]Explanation by RD further down makes more sense .

 

[Assimilator1]

I guess everone else is holding out for the more accurate benchmark WU test? .

I've grabbed an appropriate WU now (AR 0.392), I want to run it myself to check
1. It's not a duff WU.
2. What time I get.
3. Confirm it gives 73 creds.

However it was then I realised I didn't know how to insert it into another BOINC instal without it deleting 'my' WU & downloading others!.
I've PMed RD & hopefully he can help but if anyone else knows how then by all means tell .

Another problem is that I won't be able to host the 'benchmark WU' without deleting alot off my website which I don't really want to do, anyone know where I could grab a little hosting space for nothing?

Anyone else who doesn't want precise results to the nearsest second then by all means post your scores .

 

[Alyx]

I'm going to swap a couple computers to SETI, it'll take a couple days for the current que to clear out and to get good results but I'll post them.

Edit: I've moved 2 P-D 3.0Ghz machines, a P4 3.4 Ghz machine and tonight I'll move my quad over.

 

[Rattledagger]

Originally posted by: Assimilator1
Angle Range of the WU, IIRC correctly that's basically the angle the telescope was recording from (not pointed as it's fixed).
Yep, look here.

Hmm, not sure that's the best method to explain Angle Range... Each SETI@home-wu contains 107 seconds of recorded data. The Angle Range is, the angle between the direction the receiver is pointing at start of wu-recording, and the end of wu-recording. Since Arecibo is located on the rotating earth, it's normally not pointing in the same direction after 107 seconds as then it started.

There's 3 "main" groups of Angle Ranges:

1; Follows a specific object. To do this, must constantly compensate for earth rotation. Angle-Range close to zero.
2; Doesn't track anything specific, just follows earth rotation. In theory, this would give:
107 seconds / 86400 seconds in day * 360 degrees rotated in a day = 0.44583 Angle Range.
In practice, you'll get Angle Range in between 0.35 and 0.45, and it seems AR 0.392 is fairly common.
3; Does "basketweave scanning", giving typical Angle Range 1.5-3.



As for re-running same wu. The "easy" method is, just download some work, and before started crunching newly-downloaded work, select "Network activity suspended", exit BOINC, backup boinc....

If wants to run one or more of the wu's multiple times, just copy-in the backup-copy of BOINC, and start it... making sure you leaves "Network activity suspended".

 

[Alyx]

If we want to distributed a work unit to all of us would it work as long as we use network suspend?

 

[Assimilator1]

Yea I believe the WU can be crunched by anyone (once I've sussed out how to get WUs re-crunched that is), it's only the result that has your own ID AFAIK. Which of course you wouldn't bother sending for the benchmark WU.

Btw good to see ya in SETI , what's your quad clocked too? (I'm sure I've already asked in DPAD(?) but I've forgotten).

RD
Thanks for the info .
I did briefly run my 2nd copy of SETI last night, I found that it had all the finished results of my main instal, I thought I'd cleared everything but the bench WU & 3 others.
Where are the results stored?

 

[Rattledagger]

Originally posted by: Assimilator1
Where are the results stored?

Wu and result-files that haven't uploaded yet is stored in boinc-data-dir\projects\project-url

For SETI@home this is boinc-data-dir\projects\setiathome.berkeley.edu

Info about which wu is assigned, downloaded, started, finished crunching and so on, is stored in client_state.xml located in the boinc-data-dir.

The actual SETI@home-result-file doesn't include anything about who has crunched a wu, this is generated based on info in client_state.xml.


Now, it is possible to run SETI@home "stand-alone", but not sure on how this is actually done...

 

[Assimilator1]

For SETI@home this is boinc-data-dir\projects\setiathome.berkeley.edu

Yep I've already cleared that out, results are still showing in the 'tasks' window though & the 4 WUs it does have are not showing up.

Info about which wu is assigned, downloaded, started, finished crunching and so on, is stored in client_state.xml located in the boinc-data-dir.

OMG have I got to edit that file to clear the results & insert 'my' WUs??
Maybe it would be easier to start from a fresh install? though I'd still have to insert 'my' WUs.

Btw what did you mean 'run SETI@home "stand-alone", ?

 

[Alyx]

I've got my first 74 credit unit.

Pentium-D 3.0 GHz with SSE3 app.
CPU time -
7393.406 - 74.03 - AR 0.388636
7599.453 - 74.02 - AR 0.388780
7427.875 - 74.03 - AR 0.388636

Different Pentium-D 3.0 GHz with 2 GB ram (more than other CPU)
6,808.719 - 74.09 - AR 0.388312
6,781.75 - 73.35 - AR 0.392354
6,774.13 - 73.22 - AR 0.393159

Q6600@ 3.28 GHz running the SSSE3x app.
4,290.26 - 74.07 - AR 0.388415
4,035.09 - 73.71 - AR 0.390763
4,193.55 - 73.71 - AR 0.390763
3,959.32 - 74.12 - AR 0.388079

 

[Assimilator1]

Is the only difference between the 2 P4 Ds the RAM quantity?, I'd be surprised if that made much difference to crunching times seeing as it only uses 30-40MB per process.
Different chipset or RAM timings ,maybe?
Interesting to see a time difference of 158s (~4%) between the 0.39 AR WUs on the Q6600 (& a bigger difference on the AR 0.38s , ~8%), shows we do need a specific WU to get accurate comparisions. However its already far more accurate & insightful than RAC .

 

[bryanW1995]

this is fun! fwiw, I think that the memory speed/timings has a greater impact than it does on most apps. alyx's q6600 @ 3.28 is almost = my q6600 at 3.5, but my q6600 is running ram at ddr2 782 5-5-5-15 and trd 8, also it's only single channel.

 

[Assimilator1]

SETI classic was sensitive to memory timings too & cache size (though that varied),sounds like some things don't change

IIRC my Q6600 RAM timings are 4-4-4-15 ,dual channel (2x1GB).

I need to start compiling a table, at least a prelimnary one anyway.

 

[Alyx]

Originally posted by: Assimilator1


Is the only difference between the 2 P4 Ds the RAM quantity?, I'd be surprised if that made much difference to crunching times seeing as it only uses 30-40MB per process.
Different chipset or RAM timings ,maybe?

They were purchased a year and a half apart so I'm sure the memory and chipset are different. I'm not at work today so I can't check but I know the faster one is DDR2 and I believe the other still has DDR1.

Originally posted by: bryanW1995
this is fun! fwiw, I think that the memory speed/timings has a greater impact than it does on most apps. alyx's q6600 @ 3.28 is almost = my q6600 at 3.5, but my q6600 is running ram at ddr2 782 5-5-5-15 and trd 8, also it's only single channel.

mine is 3.284GHz and ram is 4GB of DDR2@730 with 5-5-5-15 2T timings (not sure about trd) but it is dual channel.

 

[Assimilator1]

Ok I've created a partial table, it doesn't include all results yet but it's enough to get a good idea of what it looks like.
See the TA top '100' table here
.......err I have no idea what happened to rank 6!, no time to sort it now.

(Anyone know how I could convert it from an excel file to a table like Roelof's TLC Top 200? )

I've decided to change the credit range for the WU results, we need to stick to 73.xx credit WUs to cut down the time variance, these so far have also all been 0.39 AR WUs.

Alyx
Could you dig out some ~73 cred results for your 1st P4-D?
Sorry about that but I want to keep the results table reasonable accurate, I thought 74creds would be OK but that ups the inaccuracy to 10% which I think is too much.

Greg
Just FYI
Q6600 @3GHz, 4,513.09s 73.50 credits, AR 0.391472

I did see a 72.78 cred WU which was also a 0.394307 AR WU (4391s) & a 73.94 cred WU @ 0.389319 AR (4429s).
Maybe the 0.39 AR WU will range from 72.5 to 73.5??

 

[Assimilator1]

Well I just found out why my Q6600 rig was so much slower than other peoples ,turns out I was running the SSE3 client instead of the SSSE3 version. I'd missed the one 'S' ,stupid Intel naming . I'll post new scores when I get them.

Btw can you all confirm whether you are running the SSE3 or SSSE3 client (called by default 'AK_v8_win_SSSE3x' or 'AK_v8_win_SSE3' in task manager), or SSE4.1 for those lucky enough to be able too .

 

[Rattledagger]

Originally posted by: Assimilator1
OMG have I got to edit that file to clear the results & insert 'my' WUs??
Maybe it would be easier to start from a fresh install? though I'd still have to insert 'my' WUs.

Basically, yes, you'll either need to manually edit client_state.xml, or, can after you've downloaded some wu's see that some of them are the AR you want to keep, pause these wu's, crunch the rest & report the rest, and make backup of only the unstarted wu's...

Btw what did you mean 'run SETI@home "stand-alone", ?

It is possible to run only SETI@home and not BOINC, but for the moment I don't remember the way to do this... Also, in case the application includes graphics, it's normally display these, slowing-down processing.

# Biggest time difference ~720s (12mins) ,average those 2 (4151) & it's a variance of about +/- 8% ,hmm not good! Still it's better than RAC.
We definitley need a benchmark WU for accurate scoring though.

From your small table of wu's with different AR, I've not calculated credit/hour (or credit/second), but it's possible the variation for this fairly low variation in AR is so low that it will be a fairly accurate benchmark.

But of course, having a specific benchmark-wu is an advantage, since everyone will run the same wu... but this also gives the disadvantage, you'll wasting usable crunching-time doing the benchmark, instead of running real wu's...

Even with a benchmark-wu, there'll still be some variation in crunch-time even if runs 4 copies of same wu, or run same wu multiple times alongside other work.

BTW, to see if cpu is memory-bound running multiple instances, example on quad-core it would be interesting to see the difference in cpu-time then runs 4 copies of same wu at once, compared to running a single instance and nothing else...

Atleast with AR 1.5+, there's a noticeable slow-down running multiple copies with high AR, compared to running one VHAR, and one "normal" AR... or another project... How much this slow-down is, probably depends on how fast the memory is, and possibly cache-size.

Btw can you all confirm whether you are running the SSE3 or SSSE3 client (called by default 'AK_v8_win_SSSE3x' in task manager), or SSE4.1 for those lucky enough to be able too .

I've not followed the various threads about optimized application closely, but AFAIK there was something about the SSSE3 is faster than SSE4.1 if computer has slow memory, while SSE4.1 is faster with fast memory... And, which application is best depends on AR, so for some SSSE3 is best while for others SSE4.1 is best...

 

[Assimilator1]

From your small table of wu's with different AR, I've not calculated credit/hour (or credit/second), but it's possible the variation for this fairly low variation in AR is so low that it will be a fairly accurate benchmark.

Yea this is true ,often the variation is 5% or less which is OK ,but at around 8% variation that's getting a little too much, however averaging out 3-4 results would keep the variance low & it would still be a quicker & more accurate benchmark than RAC .
Ideally we need to easily be able to re-run a benchmark WU (with any other cores busy crunching too), maybe once I've got the ideal WU inserted into BOINC I could backup the entire install & have that to share.

you'll wasting usable crunching-time doing the benchmark, instead of running real wu's...
True but we had that for SETI classic & the TLC WU, many people did use it & as mentioned it does give the most accurate score.

Even with a benchmark-wu, there'll still be some variation in [the] crunch-time even if [they] run 4 copies of [the] same wu, or run [the] same wu multiple times alongside other work.

Yea but their will only be significant variation if someone runs the benchmark alongside other CPU intensive programs. I guess that's worth mentioning when (if?) I get to the point that I have a BOINC install ready to share.

BTW, to see if cpu is memory-bound running multiple instances, example on quad-core it would be interesting to see the difference in cpu-time then runs 4 copies of same wu at once, compared to running a single instance and nothing else...

Yep that's the beauty of having a benchmark WU, I'd certainly be interested in giving that a go .

At least with AR 1.5+, there's a noticeable slow-down running multiple copies with high AR, compared to running one VHAR, and one "normal" AR... or another project... How much this slow down is, probably depends on how fast the memory is, and possibly cache-size.
Interesting, I didn't know that ,why would this be? Where did you hear of this?

About the SSE3 vs SSSE3 clients I was just asking people to confirm what client they are running .
Curious about SSSE3 vs SSE4.1, you think that 4.1 would be faster no matter what AR the WU, wierd!, maybe its the same scenario as with multiple VHARs, maybe the memory bus is being saturated due to higher throughput? ,in which case overclocking the RAM should give a decent boost, hmmm I'm pretty sure I could o/c my RAM more .

 

[Rattledagger]

Originally posted by: Assimilator1
At least with AR 1.5+, there's a noticeable slow-down running multiple copies with high AR, compared to running one VHAR, and one "normal" AR... or another project... How much this slow down is, probably depends on how fast the memory is, and possibly cache-size.
Interesting, I didn't know that ,why would this be? Where did you hear of this?

Someone asking about a way to not get VHAR running on all 4 cores at the same time over at the seti-forums. Most likely the variance is due to VHAR being more dependent on memory-speed than "normal" AR, and slow-down likely due to saturated memory-bus.

Don't remember how much the slowdown was, possibly it was around 10% or something.

Did see a slow-down on my dual-core some time ago then had a bunch of VHAR, and running another project on one core sped-up crunching again on VHAR. But, again, don't remember how much variation...

About the SSE3 vs SSSE3 clients I was just asking people to confirm what client they are running .
Curious about SSSE3 vs SSE4.1, you think that 4.1 would be faster no matter what AR the WU, wierd!, maybe its the same scenario as with multiple VHARs, maybe the memory bus is being saturated due to higher throughput? ,in which case overclocking the RAM should give a decent boost, hmmm I'm pretty sure I could o/c my RAM more .

The latest isn't always the fastest, for Amd the SSE2-versions has AFAIK always been faster than SSE3 on internal benchmark-runs, so for this reason there haven't been any SSE3 for Amd before...

The SSE4-version is possibly more dependent on memory-speed than earlier SSE-versions, so with "wrong" memory-timings, you'll get an extra penalty waiting for memory, so even SSE4 is faster, in practice it's wasting too much time on memory and delivers result slower...

 

[Alyx]

Originally posted by: Assimilator1
Alyx
Could you dig out some ~73 cred results for your 1st P4-D?
Sorry about that but I want to keep the results table reasonable accurate, I thought 74creds would be OK but that ups the inaccuracy to 10% which I think is too much.

7472.922 - 73.33 - AR 0.392548
7375.344 - 73.32 - AR 0.392548
7416.75 - 73.34 - AR 0.392399

 

[Assimilator1]

Thanks Alyx ,only slightly less variation! lol.

RD
Interesting , well for now this mini-project is on hold until I get my main rig back online, its out of action atm .