E4 vs E6 - Does it make a difference

[Topweasel]

The jist of what I am going to ask came from a question about purchasing a 3800+ Toledo. I got curious because AMD wasn't going to have any Toledos be rebinned as 3800+s (this is what i heard from some including AT). After some research it apeared to me that some of the confusion was X2 3800+s being sold that were E6 which up till now meant that it was a Toledo. After Referencing AMDs sight I found that actually many CPUs where being created under E6 including several Semprons. CPUs made on the E4 stepping apear to be limited to all San Diego and Manchester cores, with the 3800+ being the only dual core part that is manufactured under E4 and E6 steppings (while the 3500+ SC is produced under both as well).

The Big thing I see is that with the FX 55 and FX 57 being the fastest CPUs AMD sells, it makes me wonder if the San Diego cores haven't been moved to E6 because it might not allow for the best top end speed. E6 might have a better yields but the E4 might have a better chance of hitting a top end of 2.8 or more.

I ask my question becuase it appears that they are moving all of produciton towards this point and it makes me wonder if for overclocking would it be best to make sure a person were to get a E4 3800+ while the getting is good.

For reference Please refer to this thread and this website.

 

[Gronich]

I was just posting in the other thread so I repeat my entry here:


I believe I have a Toledo core with half the cache disabled:

X2 3800+ 2.8Ghz@1.375v

Prime95 stable @ 13 Hours with a slight voltage bump to 1.4v.

Which in comparison with some of the other post I have seen they are averaging around 2.6Ghz@1.5v.

I would guess that because the surface area of the chip is bigger compared (Toledo) 199mm2 vs (Manchester) 147mm2 (35% bigger) this is allowing the CPUs heat to be dissipated more efficiently.

 

[Furen]

Here's my understanding of this whole mess (I might be wrong about some of my points, just so you know):

In single-core CPUs:

E3 is the initial "Venice" revision.
E4 is the initial (and only) "San Diego" revision, which corrected some errata in the E3
revision as well.
E6 is the revised "Venice".

In dual-core CPUs

E4 is the Manchester core, this also includes the corrections on the E4 single-core revision
E6 is the Toledo core, same as the above.

So basically the only difference between E4 and E6 in dual cores is the cache. What anandtech said in its review of the 3800 is that AMD would be producing true manchesters, not just Toledos with disabled cache (which they claimed is what they had been doing until then).

 

[Furen]

Originally posted by: Gronich
I would guess that because the surface area of the chip is bigger compared (Toledo) 199mm2 vs (Manchester) 147mm2 (35% bigger) this is allowing the CPUs heat to be dissipated more efficiently.

You just got a really good CPU. Silicon substrates, like all non-cubic crystals, has an Anisotropic thermal conductivity. This makes their "vertical" thermal conductivity better than their "horizontal", meaning that heat can move upwards (towards the heatsink) much easier than it can move sideways (towards other crystals). Basically, if your extra cache is not producing heat it doesnt really change the thermal properties of your CPU.

 

[Topweasel]

Looking over it more I see that even semprons are being made on E6 (which would mean .9 sempies makeing .65 939s even more likely) makes me think that everyone who thinks the have a 3800+ Toledo really has a 3800+ manchester. Trust me AMD would get better value with their Toledos that can't run a 2.2GHz+ turning them into slower DC Opterons, and I doubt you would find a CPU that could not reach 2.2 or use the full 2MB cache, not when almost every 3800+ sees at least 2.4GHz and possibly more.

I think the only way to answer all questions would be for somebody to pull the spreader on both a E6 3800+ and a 4400+ and compair core sizes.

 

[Diogenes2]

oops..

 

[bjc112]

Originally posted by: Topweasel
Looking over it more I see that even semprons are being made on E6 (which would mean .9 sempies makeing .65 939s even more likely) makes me think that everyone who thinks the have a 3800+ Toledo really has a 3800+ manchester. Trust me AMD would get better value with their Toledos that can't run a 2.2GHz+ turning them into slower DC Opterons, and I doubt you would find a CPU that could not reach 2.2 or use the full 2MB cache, not when almost every 3800+ sees at least 2.4GHz and possibly more.

I think the only way to answer all questions would be for somebody to pull the spreader on both a E6 3800+ and a 4400+ and compair core sizes.

Not all 3800+ hit 2.4ghz on STOCK voltage..

That's the same thing with 3000+'s then.. Why would they sell my Venice 3000+ (did 2.8ghz on 1.49v) for $150 when it easily could have done 3800+ speeds on stock voltage!

 

[SubtleIntelFreak]

AMd still allows overclocking? How silly, overclocking kills CPU"s within one week. Intel chipsets allow something like overperformanceperwattclocking

 

[n7]

I stole my own post from another thread.

There are actually two different X2 3800+ out there.

One is an E4 Manchester.
The other is a less seen E6 Toledo with half the cache disabled. Most likely they are 4400+ that had half the cache fail, & were down-binned.

If you look on AMD' site, you will see that for each Manchester X2, there are two versions, the E4, & the E6.

Now more than likely, the E6 4200+ is a cache-failed 4400+, & the E6 4600+ is a cache-failed 4800+.

That why the E6 3800+ is so interesting to me.
As of the present, there is no 4000+, so the E6 3800+ has to be a cache-failed 4400+. The only downside of that is that is may also be a speed-failed 4400+, which isn't so good for us OCers.

However, it seems both E4 & E6 3800+ are hitting very nice OCs.

I think the OCing potential is higher with the E6 3800+ myself, since i'm assuming they were designed to be 4400+, which should mean a slightly higher OC.

Either way though, both E4 & E6 3800+ seem to hit 2.4GHz at the lowest, which means a free upgrade to 4600+

Here's the links to some great info on the X2s & the different revisions:
http://balusc.xs4all.nl/Ned/har-cpu-amd-k9.php

On AMD's compare site, select X2, & more details will appear -
http://www.amdcompare.com/us-en/desktop/Default.aspx








Here's what i copied onto notepad for my own reference:

More detailed info from AMD.com:

Processor AMD Athlon 64 X2 Dual-Core
Model 3800+
Ordering P/N (Tray) ADA3800DAA5CD
Ordering P/N (PIB) ADA3800BVBOX
Stepping E6 - Toledo (half cache disabled)

AMD Athlon 64 X2 Dual-Core Details
Processor AMD Athlon 64 X2 Dual-Core
Model 3800+
Ordering P/N (Tray) ADA3800DKA5BV
Ordering P/N (PIB) ADA3800BVBOX
Stepping E4 - Manchester



Detailed info from The BalusC Server:

X2 3800+ Toledo
OEM: ADA3800DAA5CD
BOXED: ADA3800CDBOX

X2 3800+ Manchester
OEM: ADA3800DAA5BV
BOXED: ADA3800BVBOX

Honestly, since i firmly believe the X2 3800+ E6 is a cache-failed 4400+, i'd rather have it than the E4.

 

[Furen]

Originally posted by: bjc112
That's the same thing with 3000+'s then.. Why would they sell my Venice 3000+ (did 2.8ghz on 1.49v) for $150 when it easily could have done 3800+ speeds on stock voltage!

If all of AMD's CPU did 2.8GHz on air at stock voltage they'd still have to make 3000+ because the demand for lower speed grades is much greater than the demand for the higher speed grades. Just because AMD is selling you a Toledo 3800+ X2, it doesnt mean it necessarily was a 4400+ that was cache-failed and didnt reach 2.2GHz. Also, AMD's tests at each speed grade are much more extreme that the kind of testing that we, ourselves, do when overclocking.

 

[Topweasel]

All I am saying is look at the E6 stepping chart I linked to. If they are making sempies on the E6 why is it so hard to believe that a E6 3800+ X2 may actually be a Manchester instead of a toledo.

 

[Hacp]

Originally posted by: SubtleIntelFreak
AMd still allows overclocking? How silly, overclocking kills CPU"s within one week. Intel chipsets allow something like overperformanceperwattclocking

OMFG QFT!

 

[Furen]

Originally posted by: Topweasel
All I am saying is look at the E6 stepping chart I linked to. If they are making sempies on the E6 why is it so hard to believe that a E6 3800+ X2 may actually be a Manchester instead of a toledo.

Because cache-disabled toledos are revision JH-E6, the same as regular Toledos (meaning that there is no separate revision for 1MB ones) while Manchester cores are BH-E4.

E3 and E6 semprons have the same core revision as regular venices because they're cut-down Venices.

 

[Topweasel]

Originally posted by: Furen

Originally posted by: Topweasel
All I am saying is look at the E6 stepping chart I linked to. If they are making sempies on the E6 why is it so hard to believe that a E6 3800+ X2 may actually be a Manchester instead of a toledo.

Because cache-disabled toledos are revision JH-E6, the same as regular Toledos (meaning that there is no separate revision for 1MB ones) while Manchester cores are BH-E4.

E3 and E6 semprons have the same core revision as regular venices because they're cut-down Venices.

Very Interesting Do you have a source for that info to compare.

 

[RichUK]

so are the 3800+ manchesters poor overclockers and the 3800 + toledos better overclockers ?

 

[CrimsonCutie]

*bump*

 

[Topweasel]

Originally posted by: RichUK
so are the 3800+ manchesters poor overclockers and the 3800 + toledos better overclockers ?

I would guess that the Manchesters are better overclockers, since the use the same process as the San Diegos that hit 3.0GHz+

 

[n7]

See, i don't believe that's correct, Topweasel.

Manchester = two Venices
Toledo = two San Diegos

Or at least as i understand it.

So it doesn't make sense, especially since we now have E4s as San Diego, & E6 as the new Venice.
That's bascially the opposite of what i understand for the X2s...

 

[rise]

i think alot of people have toledo cores and think they are manchesters.

i know my bro bought one that was listed as a manchester at ZZF and it shows as a manchester in cpuz but the stepping is a toledo.

i dunno, i'm all confused. i know he's running 10x255 so i don't think he cares

 

[Topweasel]

Originally posted by: n7
See, i don't believe that's correct, Topweasel.

Manchester = two Venices
Toledo = two San Diegos

Or at least as i understand it.

So it doesn't make sense, especially since we now have E4s as San Diego, & E6 as the new Venice.
That's bascially the opposite of what i understand for the X2s...

See same here, I know that a Toledo is pretty much 2 San diegos but they are using a the same process as the venices now are. The Manchesters are using the same stepping as a San Diego.

This is my theory, the E4 stepping is more of a performance stepping, higher potential speeds but a lower yield rate. When selling an FX for a $1000 they basically can make up in one chip sale 250 bad chips on a wafer (this is at the $40 per chip cost). Ones that Don't Make it are resold apperently as 3500+ (only "Venice" on E4). So wasted chips are very easily made up for. This would also make sense for E6 Venices, these aren't the fastest CPUs so instead of try to super clock them all they care about is yields since these cpus sell for at max i believe $300.

I believe at this point the had an option for X2s (and therefore DC opterons), They had to choose what process for each, with the X2 I would make sense because over top speed isn't an issue and any a higher range of speeds on the 2* 1MB cores (opterons go down to 1.8 I believe) and the fact that they are almost twice the size of the 512K models for them to use the safer E6 stepping. Think about If its twice the size and has lower clocks that they can be marked at it would make sense to have them use the stepping that had less DOAs. On the other hand with the 512 models they have a very low span (2.0-2.4 GHZ) and getting cores that clock to those hights is more important then the number of ones that don't hit that (because they can't be binned lower or disable any cache as they don't have any models for it). Then you take size into account the impact of bad (or un binnable parts) will be less since the are half the size.

The more I think about it, I was completely wrong. 3800+ X2s and other 512k X2s can be Toledos (nuetered) and the E6 selection shows it. But this makes me think for multiple reasons that the Manchester cores on the E4 are Probably the best overclockers. But I think we will see more of a march towards the E6 stepping as it is tweaked more and more, and that any future progressions will be based on that. Since we are only expecting a increase in speed of 200MHZ and maybe 400MHz prior to .65 it would make more sense to keep the yields high then to dig for extra performance.

Edit: notice some of my views change as I was writing this, edited it to flow with the rest of the post.