Why do "slow" processors overclock so well?

[ShawnD1]

I'm looking at the overclockability of some Intel processors sold at the local computer parts shop.

Dual Core E1200 - 100% overclocked
Dual Core E2140 - 95% overclock
Dual Core E4500 - ~85% overclock
Dual Core E7200 - 78% overclock
Dual Core E6600 - 67% overclock
Dual Core E8500 - 46% overclock


This isn't a new thing either. I remember when there was a lot of hype about the original Celeron processors being overclocked by ridiculous amounts. This benchmark shows how an overclocked Celeron can be compared to a stock X6800 in most test areas.
Is Intel just selling the same processor with 20 different names and 20 different preset frequencies?

 

[xxxFrogg]

Generally, If they match up across the board, same core (Ex: Yorkfield), and have the same size caches, they are pretty much the exact same chip with different set multipliers.
We are charged a premium for an increased multiplier. These chips pretty much have the same boundries as far as OC but this is also based on the stepping used as well as luck.
So if they match up, Lower speeds have a lower set Multiplier. Multiplier X FSB gives you the set speed in Hz. Usually, the FSB is Quad pumped which means the base FSB for example 333 is multiplied by 4.
4x333= 1332 which is what is thought of as a 1333 FSB.
When a multiplier is locked, it puts all the OC stress on your motherboard to increase your FSB. Since Speed(Hz)= FSBxMultiplier.
Ex: CHIP1 (Rated at 65watts thermal dissapation)
Multiplier=8 FSB=333
8x333= 2.66Ghz
Same chip, but i'll Call it
Chip2(they usually up the chip number and charge a few hundred dollars extra for.)
A real work example of this is the Q9450 and Q9550, 8 and 8.5 multiplier respectively.
9X333=2.997Ghz approx.= 3Ghz (Rated at 95Watts) increasing speed increases the heat output

but, say u buy CHIP1 and OC the FSB
8x375= 3Ghz

For some overclocks, you have to increase various voltages which increases heat output and increases stress on various components.


With all that being said, not all chips are exactly the same chips. In the case of the ones that are, they mostly all have the same OC potential. In the case of chips who are locked at a low multiplier, there is usually max FSB that they can stabily and safely acheive. In the case where you hit the max FSB stability, you would need to get the chip with a higher multiplier to go any farther.

 

[Foxery]

In simpler terms -- when Intel/AMD create a range of CPUs with several speed grades, e.g. 2.0, 2.33, 2.66, 3.0ghz models of the same architecture, they actually start out exactly the same on the manufacturing line. They are all identical on the production line, and labeled at the end based on some quick stability tests.

Throw some extra voltage into that 2.0 GHz model, and you can push past the imperfections that kept it from being stamped 3.0.

 

[ShawnD1]

Originally posted by: Foxerymodels of the same architecture, they actually start out exactly the same on the manufacturing line. They are all identical on the production line, and labeled at the end based on some quick stability tests.

Interesting. How do they determine which ones get 1mb cache, 2mb cache, or 4mb?

And thanks for the overview Frogg.

 

[xxxFrogg]

No problem Shawn. I tend to get to technical sometimes, but its a habit since i'm an engineer. Im not sure on the cache side of things. I think its just whatever they want to put.
The general idea though i think is that bigger is better but costs more. Maybe somone else can chime in on that side of things.

 

[myocardia]

Usually, for Intel at least, they do it roughly the same way as they do the different speed grades. For instance, all E2xx0's are actually E4x00's, with half of the cache disabled. That way, they can still sell the chips that have a portion of their L2 cache that turned out bad.

 

[Aluvus]

Originally posted by: ShawnD1

Originally posted by: Foxerymodels of the same architecture, they actually start out exactly the same on the manufacturing line. They are all identical on the production line, and labeled at the end based on some quick stability tests.

Interesting. How do they determine which ones get 1mb cache, 2mb cache, or 4mb?

And thanks for the overview Frogg.

It depends.

Sometimes two variants of the same basic design will be produced, with different amounts of L2 cache. A chip with less cache will be physically smaller, meaning more dies can fit on each wafer, which makes it cheaper to produce. So there is merit in this method.

Other times, everything will roll off the same manufacturing line. The cache will be tested for defects, and dies with too many (and the right kind of) defects will have half their L2 cache disabled. This is better than throwing them away, so again there is merit in this method.

When the Core 2 Duo originally launched, the E6300 and E6400 (which have half as much L2 cache as their more expensive brothers) were Conroes with half their cache disabled. Later, Intel released new versions of the E6300 and E6400 based instead on the Allendale core, which is like Conroe but only has half as much L2 cache.


The process that Foxery described is known as "speed binning". It is a standard practice in the world of microprocessors, and is also done for some other electronics.

 

[Duvie]

INtels yields are quite good and I know some will disagree but I think a lot of the stamping of speeds are speed binned for market need....Economy 101...

I have owned 4 intel conroe chips and I have yet to ever have one that wouldn't do 3ghz on stock vcore, and I have never bought one above rated 2.4ghz speed. My quadcores have all done at least 3ghz as well at stock vcore...My older b0 stepping (albeit an ES chip) does 3.2ghz at stock vcore, but MarkFW's fleet of Q6600's I believe have done 3-3.2ghz at stock vcore.

I agree the different cache models can definitely be graded based on certification testing of the cache...The E7300 model with 3mb of cache screams it...

 

[Markfw]

My fleet is dwinding... I sold my 2 B3 stepping Q6600's, and one E6300. The electric bill was out of right !

 

[Duvie]

Originally posted by: Markfw900
My fleet is dwinding... I sold my 2 B3 stepping Q6600's, and one E6300. The electric bill was out of right !

I know I bought one for a friend!!!!

Does 3ghz overnight at .025v below stock setting....

 

[bryanW1995]

Originally posted by: Markfw900
My fleet is dwinding... I sold my 2 B3 stepping Q6600's, and one E6300. The electric bill was out of right !

dude, I really think that Q9400 when it comes out will be a good solution for you. It should overclock as well as your 9450 but will have less power draw due to the lower cache.

The only reason I got the Q6600 (other than the dirt cheap price obviously) is that it's going in mom's house. :evil:

 

[Tempered81]

dang bryan i just looked at your sig and saw that your q6600 oc'd higher than your q9450. What does the speed difference feel like? is the q9450 still faster in benches?

 

[Duvie]

Originally posted by: bryanW1995

Originally posted by: Markfw900
My fleet is dwinding... I sold my 2 B3 stepping Q6600's, and one E6300. The electric bill was out of right !

dude, I really think that Q9400 when it comes out will be a good solution for you. It should overclock as well as your 9450 but will have less power draw due to the lower cache.

The only reason I got the Q6600 (other than the dirt cheap price obviously) is that it's going in mom's house. :evil:

The problem is the cache has some benefit in the world of folding at home....So he will want maximized PPD for the limited boxes he wants to run