Looking to take the 940 to 4.0 24/7 on air, which mobo?

[TL1000S]

Thanks Chew*.
Great lecture.

[Kingsfan]

It's a revision of the board with SB750 instead of SB600

would it be worth it to upgrade my DFI 790FX with sb600 to the board with sb750?

[crazydiamond]

prolly be a downgrade IMO

[Aquineas]

Sure theres a large wafer, they make quite a few chips out of it lets toss a 1000 as a # of chips made. ( Not really sure how many they can squeeze out of a wafer atm )
<snip>

Great informative post. I learned a lot from reading it!

[Halk]

Sure theres a large wafer, they make quite a few chips out of it lets toss a 1000 as a # of chips made. ( Not really sure how many they can squeeze out of a wafer atm )

I lifted this pic off particle for explanitory reasons, hope you don't mind particle.


So the bottom line last 5 digits are what we are looking at here.......Looks like particles came from batch 80000 and is 74 in that batch of 80000, that batch will go up to 81000. The higher the # the further you get away from the center of the wafer of silicon.........The center of the wafer houses most of the "Cherry" chips in a batch as the silicon at the center of the wafer is of the purest quality.

Now getting a chip in the first 80001 to 80100 increases your chances of getting a good chip but there is a catch. Batch 80000 could have been a poor batch to start with and batch 81000 might be better.........getting a low batch # just increases your chances of getting a higher yield chip in a batch even if the batch is poor at clocking to begin with.

I personally don't buy chips unless they are 01 - 50 and where i get my chips from they have plenty of chips to pick through ( distributors warehouse ), I've rarely gotten duds this way.

In 939 x2 I ended up with one that was under #50 from a poor batch and still did 3.0 under a mach 1 24/7 stable, The other chip I got was # 47 in a decent batch and did 3.0 under water and I later went on to take the x2 3800 WR at 3.5g under DI for a short period of time.......( I still have it and plan to take the current 3.7 WR down with LN2 )

Batch is not god and does not mean you will get the best chip in the world but it will ensure you getting a better piece of silicon out of the whole batch, try to think of it as playing blackjack against the casino versus any other game, You odds are higher with blackjack than any other card game at the casino

Last but not least particle nice chip being in under 100, have fun with it.

I could go further into what the other numbers meant back in the day but alot of things may have changed. I may make a new database for stepping/batch.......last time I was able to determine which chips did well under cold among other things.....this time I might be able to figure out something different but I think I will reserve that for AM3 as the 920/940 appear they will be short lived.

Chew, thanks for taking the time to explain that!

The entire codes on my chip are the same as everybody elses..

0850DPCW
9100082L80280

Meaning I got number 280 which isn't the best. However it's not around the edges (I hope) so it's not the end of the world.

Edit : Since I'm a crack whore for knowledge.... Care to shine any light on how the stepping works? All I know is that some are good, some are bad, but I don't know how they generate the numbers. I remember from Opteron Dual Core that CABCE or such like was good but a couple of letters different was a not so good batch.

[EniGmA1987]

It is weird that it seems EVERYONE got an 0850 and I am the only one who got a retail 0848... Also, my last 5 digits are 80062 and my CPU cant clock the cores worth sh*t.

Just wanted to add one more little tidbit, better silicon does not always equal out to higher oc's, sometimes it will be in the form of a stronger memory controller, or chip capable of higher HTT etc etc.....

So either I got REALLY unlucky, or the 0848's are pretty bad for clocking the CPU. It seems their redeeming factor is I can stably clock the NB and HTT to 2.75GHz

[chew*]

It is weird that it seems EVERYONE got an 0850 and I am the only one who got a retail 0848... Also, my last 5 digits are 80062 and my CPU cant clock the cores worth sh*t.




So either I got REALLY unlucky, or the 0848's are pretty bad for clocking the CPU. It seems their redeeming factor is I can stably clock the NB and HTT to 2.75GHz

Like I said lower batch # doesn't always mean higher core clocks, could be in the form of HT, a stronger mem controller.

By nature amd's earlier batches aren't always the greatest and they get better with time.

[chew*]

Chew, thanks for taking the time to explain that!

The entire codes on my chip are the same as everybody elses..

0850DPCW
9100082L80280

Meaning I got number 280 which isn't the best. However it's not around the edges (I hope) so it's not the end of the world.

Edit : Since I'm a crack whore for knowledge.... Care to shine any light on how the stepping works? All I know is that some are good, some are bad, but I don't know how they generate the numbers. I remember from Opteron Dual Core that CABCE or such like was good but a couple of letters different was a not so good batch.

I could shed some light on older cpu's via a database I had created, It's honestly not worth trying to interpret 920/940 steppings because they will be short lived.

I will start a database and try to interpret AM3 however, I am keeping a personal database though of some of the 920/940's people are posting, I'm going to want to compare them with am3 steppings and batches as I get a funny feeling these are failed am3 cores ( ddr3 controller or HT failed chips ) It's the only logical explanation I can think of for AMD to launch an end of life product like this.

[Halk]

So what you're saying is it's not public what the batch numbers actually mean. The four letters could mean days of the week, night shift, day shift and back shift, they could be which QC team checked it. But as far as we're concerned we can't translate it into anything meaningful other than watching which batch numbers are good, and which are not so good.

[chew*]

So what you're saying is it's not public what the batch numbers actually mean. The four letters could mean days of the week, night shift, day shift and back shift, they could be which QC team checked it. But as far as we're concerned we can't translate it into anything meaningful other than watching which batch numbers are good, and which are not so good.

No they can be interpreted, for instance we interpreted that CDBHD chips had a strong memory controller under cold with 939 and thus did not coldbug or coldbug as bad........You can also identify chips that were failed FX models etc etc.....They definitely are not days of the week/night shift day shift.....there is another line that determines that

Trust me I will go into detail eventually on steppings but first I need a database and results to back up any findings

[Halk]

So AMD label the chips with the this code after they've tested the silicon?

[chew*]

So AMD label the chips with the this code after they've tested the silicon?

Yes btw the last 4 letter in the stepping doesn't mean as much as you think. the 5 letter stepping has more importance....the chip pictured above has a date code in it as well. It was made in 08, and its week 50 If I remember correctly.....thats what the 0850 part means....So this chip was done around the 14th december.....

[Halk]

Yep, I understood the date, it's kind of obvious. All I know about the 5 letter steppings is the same letters seem to get used, from San Diegos to Phenom IIs...

[chew*]

I created a thread on steppings in effort not to derail this thread anymore.

[Titan7171]

Dont they cut the cores off from one side of the wafer and work their way through to the other side? this would mean the first cores assembled would be from the outer edge and be first numbered........they wouldnt wait till they got to the center of the wafer then start numbering them......as an example if theres 144 usable cores on a wafer (smaller scale)and they cut from left to right then move in and cut again. that would be a wafer (that looks like a 12x12 tictactoe grid) from left to right and top to bottom the first cores out would be from the edge, the best cores would be from the center 4 rows leaving 4 rows on the left and right and 4 rows on top and bottom

that leaves 16 cores in the very center numbered 53-56, 65-68, 77-80, 89-92.

the highest numbers off the batch would be the last row cut from wafer.

Unless they start cutting from the middle I dont think cores from the center would have low numbers.

This is just an observation.....i have no idea how AMD cuts their cores.

[chew*]

Dont they cut the cores off from one side of the wafer and work their way through to the other side? this would mean the first cores assembled would be from the outer edge and be first numbered........they wouldnt wait till they got to the center of the wafer then start numbering them......as an example if theres 144 usable cores on a wafer (smaller scale)and they cut from left to right then move in and cut again. that would be a wafer (that looks like a 12x12 tictactoe grid) from left to right and top to bottom the first cores out would be from the edge, the best cores would be from the center 4 rows leaving 4 rows on the left and right and 4 rows on top and bottom

that leaves 16 cores in the very center numbered 53-56, 65-68, 77-80, 89-92.

the highest numbers off the batch would be the last row cut from wafer.

Unless they start cutting from the middle I dont think cores from the center would have low numbers.

This is just an observation.....i have no idea how AMD cuts their cores.

How they cut them out doesn't matter, they could number backwards, the lower batch #rs always comes from the center, I could offer supporting evidence but i fear I would be giving intel guys ammunition so I will refrain.

[carajean]

Ok back on track..... Im gonna take the plunge and get the proc from newegg. As I see more reviews come out im not too sure I will be able to reach my overclock goal of 3.8ghz for 24/7 bummer I know. Here is what I have so far for board, proc, and ram. Do let me know if I went wrong somewhere.

http://www.newegg.com/Product/Produc...82E16813131339

http://www.newegg.com/Product/Produc...82E16820231211

Proc is the X4 940. Im needing fast responses so I can tweak my shopping cart.

P.S. I may go with this board http://www.newegg.com/Product/Produc...82E16813130190 I would save $35 bucks. It doesnt look like a bad board but im NOT a fan of MSI after buying a couple of their boards. Also I may consider that new gigabyte board. Im a Xtreme fan of DFI and ASUS but DFI may be out of my price range this build.

[Klarko]

So the new motherboards will be out when AM3 comes out?

sidenote- what are the best air coolers for deneb? Im looking to spend a little extra to get a little higher clock.

Thanks

[carajean]

So the new motherboards will be out when AM3 comes out?

sidenote- what are the best air coolers for deneb? Im looking to spend a little extra to get a little higher clock.

Thanks

Either you posted in the wrong thread or im gonna tell you to make your own thread. MY THREAD has been sidetracked enough!

[chew*]

Ok back on track..... Im gonna take the plunge and get the proc from newegg. As I see more reviews come out im not too sure I will be able to reach my overclock goal of 3.8ghz for 24/7 bummer I know. Here is what I have so far for board, proc, and ram. Do let me know if I went wrong somewhere.

http://www.newegg.com/Product/Produc...82E16813131339

http://www.newegg.com/Product/Produc...82E16820231211

Proc is the X4 940. Im needing fast responses so I can tweak my shopping cart.

P.S. I may go with this board http://www.newegg.com/Product/Produc...82E16813130190 I would save $35 bucks. It doesnt look like a bad board but im NOT a fan of MSI after buying a couple of their boards. Also I may consider that new gigabyte board. Im a Xtreme fan of DFI and ASUS but DFI may be out of my price range this build.

I'm assuming you cant wait? the asus seems like a solid performer especially for fine tuning. From what i hear with the same equipment it gets better performance #rs at same settings.

[EniGmA1987]

Best air cooler would be the Sunbeam Core Contact, Thermalright Ultra-120 Extreme, OCZ Vendetta 2. The core contact is the the best performance per $. It is equal to and sometimes beats the TRUE, and costs less. Put a good fan on it though, dont use the stock one. Buy a high RPM delta or soemthing.

Also, the Gladiator cooler from OCZ is worse than the Vendetta 2. It has larger heatpipes and costs more, but somehow it performs worse.

[chew*]

I'm using a true 120 Black and it definitely kicks hiney, I don't think theres a cooler out ther that can beat it if you use a push pull setup.....save the TRUE copper. I'm right now about 10 hours into prime with a Q6600 @3.6 gig running small FFT's to generate the most heat to cure some thermal paste quicker and hottest core is 61C, I would imagine AMD's would run far cooler at 45nm process....MY OCZ vindicator was breaking 80C for reference at same clocks, volts, fans etc etc.

Bang for buck wise though the xigamatex 1284EE or 1283 are both excellent coolers for AMD, for intel they need the add on backplate kit..........which makes them cost like $15 less than a TRUE.

[Sentential]

One thing that (we) are running into is the same problem we had with the 9950BEs, alot of the cheaper motherboard's power layout has issues with very high vcore and crashes because the mobo's FETs are overheating.

For example no matter how much voltage or clock speed I use anything over 1.55v instantally crashes. Alot of people like myself on cheaper boards are having this issue ie MSI, Biostar and the like.

I have heard some of the DFI's are less prone to this since they use a digital PWM and Asus is generally of high quality, the same can be said of Gigabyte. In either case if you intend to use a tower sink be sure to have some kind of airflow over the power regulation heatsink as that will honestly be the limiting factor, that and heat as these chips do NOT like getting hot (50c+).

Also these chips seem to have the same quirky behavior in Vista64 where they clock significantly lower than their 32bit cousins. I really dont understand why but it has been well documented.

Finally these chips appear to have "ACC" enabled by default so you do not have to have a SB750 to get the higher clocks, so far the average across all motherboards both amd and nvidia alike are hitting in the same range of about 3.5 on the low end and 3.7-3.8 on the upper end with air.

[carajean]

I'm assuming you cant wait? the asus seems like a solid performer especially for fine tuning. From what i hear with the same equipment it gets better performance #rs at same settings.

No I cant wait because the only reason im upgrading is because my parents need a computer now ( PSU on their crappy dell took the board with it to hell ). What would I be waiting for?

Best air cooler would be the Sunbeam Core Contact, Thermalright Ultra-120 Extreme, OCZ Vendetta 2. The core contact is the the best performance per $. It is equal to and sometimes beats the TRUE, and costs less. Put a good fan on it though, dont use the stock one. Buy a high RPM delta or soemthing.

Also, the Gladiator cooler from OCZ is worse than the Vendetta 2. It has larger heatpipes and costs more, but somehow it performs worse.

I have the Tuniq 120. Im not getting another cooler, if anything I will get another fan for it.

I'm using a true 120 Black and it definitely kicks hiney, I don't think theres a cooler out ther that can beat it if you use a push pull setup.....save the TRUE copper. I'm right now about 10 hours into prime with a Q6600 @3.6 gig running small FFT's to generate the most heat to cure some thermal paste quicker and hottest core is 61C, I would imagine AMD's would run far cooler at 45nm process....MY OCZ vindicator was breaking 80C for reference at same clocks, volts, fans etc etc.

Bang for buck wise though the xigamatex 1284EE or 1283 are both excellent coolers for AMD, for intel they need the add on backplate kit..........which makes them cost like $15 less than a TRUE.

Yeah a friend of mine on my recommendation got the Tuniq for his Q6600 and is sitting nicely at 3.6ghz.

[carajean]

One thing that (we) are running into is the same problem we had with the 9950BEs, alot of the cheaper motherboard's power layout has issues with very high vcore and crashes because the mobo's FETs are overheating.

For example no matter how much voltage or clock speed I use anything over 1.55v instantally crashes. Alot of people like myself on cheaper boards are having this issue ie MSI, Biostar and the like.

I have heard some of the DFI's are less prone to this since they use a digital PWM and Asus is generally of high quality, the same can be said of Gigabyte. In either case if you intend to use a tower sink be sure to have some kind of airflow over the power regulation heatsink as that will honestly be the limiting factor, that and heat as these chips do NOT like getting hot (50c+).

Also these chips seem to have the same quirky behavior in Vista64 where they clock significantly lower than their 32bit cousins. I really dont understand why but it has been well documented.

Finally these chips appear to have "ACC" enabled by default so you do not have to have a SB750 to get the higher clocks, so far the average across all motherboards both amd and nvidia alike are hitting in the same range of about 3.5 on the low end and 3.7-3.8 on the upper end with air.

I really like your reply. With not " having " to have the sb750 kinda opens things a lot more for me. I would like to hear a few more confirmations on this before I go ahead with it and look at other boards.