AMD Phenom II X6 | Turbo CORE 101

[Hans de Vries]

Some good news for overclockers in here:

The big achievement is how they manage to get six cores running at
3.2 GHz at the same TDP as a quad core 3.2 GHz Phenom II X4 955.
The 140W 965 is only just now going EOL being entirely replaced by
a 125W version.

It's almost comparable with a full process node improvement like Intel
did by going from a 130W quad core 3.33 GHz 45nm Nehalem to the
130W six core 3.33 GHz 32nm Westmere.

http://www.tomshardware.com/reviews/...i-x6,2604.html
http://www.techreport.com/discussions.x/18731
http://anandtech.com/show/3641/amd-d...o-core-enabled
http://hothardware.com/News/AMD-Turb...logy-Revealed/

The first two sides mention low-k and THG specifically mentions it as
going between the metal lines so most likely the speed improvements
are indeed due to to lower RC-wire delays.

This looks to be very good news for overclockers since the RC delays
are basically temperature independent. You can't improve them by cooling,
so LN and LHe results are for a large part determined by RC delays.


Regards, Hans.

[Manicdan]

how would turbo be affected if i were to up the bus speed from say 200mhz to 220mhz? for my current PII with CnQ when it drops cores to 4x the speed is 880mhz instead of 800. so can i safely assume that if i buy a 1035T, i can OC it just a hair, and have it perform about the same as 1095T? (the only thing i would worry about is what voltage it would be at, being a 95W part, i assume its about .5v less at max speed compared to the 125W parts)

[RaV[666]]

I think that the more important question is, will it enable to use higher multipliers on NON-BE chips when manually overclocking.

[Hornet331]

Its marked as 3.33Ghz stock speed, yes but with turbo this speed is never kept is it its always around 3.46Ghz. In essence stock speed is never really used for anything is it?

The point is that when turbo overclock all cores its not fair, turbo should overclock when the core count is less than total cores like the 3.6ghz for 2 cores is acceptable.



Yes thats exactly what i am trying to say it never really suns at 3.33Ghz and is advertised as a 3.33Ghz cpu.

Among all the ES's i had time to spent i9 aka i7 980x was one of the best but at the same time the i had a horrible time with Lynnfield cpu's and it was all because of Intel's turbo implementation. I dont want to go in detail but the fact is Turbo soon will not hold as much importance that it does today.



See the above responses and to add to it how do you predict Turbo i mean how do you know when it should start/stop/etc. Intel has taken a very grainy road on this and i dont like it one bit.

The Turbo speed can flex "specially in all cores mode" when ever the cpu wants it to based on the environment. This gives one benefit that is when the environment has not warmed enough or is still in the process the speed of the cpu will work at 3.46Ghz but once the environment is hot enough the speed drops to stock.

Also the power consumed and TDP change with stock freq. on all cores and the turbo freq. on all cores take out a wall meter and check for yourself.

so basically your saying its bad to you get more then you pay for...

I don't see a "grainy road" at all, you pay for a 3.0ghz cpu and you get that. If the conditions are right (temp/consumption) you get more performance. How can this be bad thing... for most reviews it was disabled anyway or tested with and without turbo so even compareabillity is given...

Imho ever since intel introduced turbo certain people where in knee-jerk mode -> turbo = bad...
At least there "arguments" have run out of full now...

[Oliverda]

Some good news for overclockers in here:

The big achievement is how they manage to get six cores running at
3.2 GHz at the same TDP as a quad core 3.2 GHz Phenom II X4 955.
The 140W 965 is only just now going EOL being entirely replaced by
a 125W version.

It's almost comparable with a full process node improvement like Intel
did by going from a 130W quad core 3.33 GHz 45nm Nehalem to the
130W six core 3.33 GHz 32nm Westmere.

http://www.tomshardware.com/reviews/...i-x6,2604.html
http://www.techreport.com/discussions.x/18731
http://anandtech.com/show/3641/amd-d...o-core-enabled
http://hothardware.com/News/AMD-Turb...logy-Revealed/

The first two sides mention low-k and THG specifically mentions it as
going between the metal lines so most likely the speed improvements
are indeed due to to lower RC-wire delays.

This looks to be very good news for overclockers since the RC delays
are basically temperature independent. You can't improve them by cooling,
so LN and LHe results are for a large part determined by RC delays.


Regards, Hans.

Sounds good but what is "RC-wire"?

[tbone8ty]

960t will oc nicely !!!!

[Planet]

Cant wait to get one of these and see how they do.

[demonkevy666]

Because you run into the gpu limit...

Anyway, i wana see banchmarks.

it's on legit reviews, it was using a 5970 at 2560 x 1600.

6% difference up at that resolution.

is there such thing as a balanced system anymore ?....

[64NOMIS]

what about Cold bug 64NOMIS?

[FatAlbert]

ok

thanks

[OhNoes!]

Its marked as 3.33Ghz stock speed, yes but with turbo this speed is never kept is it its always around 3.46Ghz. In essence stock speed is never really used for anything is it?

The point is that when turbo overclock all cores its not fair, turbo should overclock when the core count is less than total cores like the 3.6ghz for 2 cores is acceptable.



Yes thats exactly what i am trying to say it never really suns at 3.33Ghz and is advertised as a 3.33Ghz cpu.

Among all the ES's i had time to spent i9 aka i7 980x was one of the best but at the same time the i had a horrible time with Lynnfield cpu's and it was all because of Intel's turbo implementation. I dont want to go in detail but the fact is Turbo soon will not hold as much importance that it does today.



See the above responses and to add to it how do you predict Turbo i mean how do you know when it should start/stop/etc. Intel has taken a very grainy road on this and i dont like it one bit.

The Turbo speed can flex "specially in all cores mode" when ever the cpu wants it to based on the environment. This gives one benefit that is when the environment has not warmed enough or is still in the process the speed of the cpu will work at 3.46Ghz but once the environment is hot enough the speed drops to stock.

Also the power consumed and TDP change with stock freq. on all cores and the turbo freq. on all cores take out a wall meter and check for yourself.

Ohnoes! Ohnoes! I ordered a 3.33GHZ cpu and de evil Intel gave me a cpu that runs at 3.46GHZ! Ohnoes!

Seriously, dude!

If turbo was such a bad idea, why did AMD implement it? Surely there must be a conspiracy somewhere! Wait...... ah I see..... AMD's Turbo Core does not boost power on all 6 cores. That could be the only reason.

Fact is turbo boost is a genius feature and AMD would have been stupid not to implement it seeing that they're staking their future on many core (just as Intel) in a world dominated by lazy programmers fixated on single-thread.

[Lightman]

I like this answer!
Was laughing hard!

[ajaidev]

so basically your saying its bad to you get more then you pay for...

I don't see a "grainy road" at all, you pay for a 3.0ghz cpu and you get that. If the conditions are right (temp/consumption) you get more performance. How can this be bad thing... for most reviews it was disabled anyway or tested with and without turbo so even compareabillity is given...

Imho ever since intel introduced turbo certain people where in knee-jerk mode -> turbo = bad...
At least there "arguments" have run out of full now...

Ohnoes! Ohnoes! I ordered a 3.33GHZ cpu and de evil Intel gave me a cpu that runs at 3.46GHZ! Ohnoes!

Seriously, dude!

If turbo was such a bad idea, why did AMD implement it? Surely there must be a conspiracy somewhere! Wait...... ah I see..... AMD's Turbo Core does not boost power on all 6 cores. That could be the only reason.

Fact is turbo boost is a genius feature and AMD would have been stupid not to implement it seeing that they're staking their future on many core (just as Intel) in a world dominated by lazy programmers fixated on single-thread.

Oh No i did not say i dont like turbo i just said that the all core boost is not fair that's it. It alters the speed of all cores if a correct environment is presented and that is grainy in it self.

The box should have been marked 3.46Ghz instead then should it not, different users of same cpu with same multi thread app will get different performance depending on the turbo mode.

As for AMD not supporting all core turbo, look at my posts when i had a intel Bloomfield ES and i reported odd bench results in cinebench, i hate all core turbo since then. I supported less than 4 core turbo from that time because single threaded app's worked better as a result.

[Humminn55]

Oh No i did not say i dont like turbo i just said that the all core boost is not fair that's it. It alters the speed of all cores if a correct environment is presented and that is grainy in it self.

The box should have been marked 3.46Ghz instead then should it not, different users of same cpu with same multi thread app will get different performance depending on the turbo mode.

As for AMD not supporting all core turbo, look at my posts when i had a intel Bloomfield ES and i reported odd bench results in cinebench, i hate all core turbo since then. I supported less than 4 core turbo from that time because single threaded app's worked better as a result.

Oh Nooooes....it's the "It's not fair!" complaint. Or the "Intel cheats!" complaint.

And, no, the cpu in question is marked 3.33GHz, and it runs at 3.33GHz base. Who cares how much boost it gives different people in differing situations? It's guaranteed to provide a base speed of 3.33GHz....and you get it. The performance boost is gravy.

Or do you prefer the AMD solution......downclocking cores but not turning them off and shoving whatever overvoltage the turbo'd cores need into the downclocked cores, despite them not needing anywhere near that voltage? Seems to me that AMD's solution is almost half-baked and a knee-jerk response to Intel's rather elegant solution.

[Manicdan]

Seems to me that AMD's solution is almost half-baked and a knee-jerk response to Intel's rather elegant solution.

i think they are just different. AMD is staying in its power envelope, inlets is about getting out as much perf as possible. how about the i7 laptops where they had to underclock due to heat, thats not very elegant

edit: somehow i put in a thumbs down, didnt mean too, lol

[accord99]

i think they are just different. AMD is staying in its power envelope, inlets is about getting out as much perf as possible.

Intel's turbo also considers the power envelop, but the i7 can get a reasonable measurement of the actual power usage through an onboard chip.

how about the i7 laptops where they had to underclock due to heat, thats not very elegant

That's not really anything to do with turbo; more of the laptop designers not designing the laptop to handle the maximum possible load situations.

[informal]

Intel's turbo also considers the power envelop, but the i7 can get a reasonable measurement of the actual power usage through an onboard chip.


That's not really anything to do with turbo; more of the laptop designers not designing the laptop to handle the maximum possible load situations.

AMD's method also uses dedicated HW which AMD is not publicly specifying .The core is a new Shanghai/Istanbul revision code named Pharaoh Hound btw.
AMD guarantees each chip won't cross over its maximum electrical specification in any circumstance. Also,we now know the changes in clocks are dynamically distributed over various cores and are very fast and thus should be "immune" to poor thread scheduling of Vista/Windows 7 OS.

[Hans de Vries]

Sounds good but what is "RC-wire"?

http://en.wikipedia.org/wiki/Repeater_insertion

The RC wire delay is the delay through a wire due to it's (R)esistance
and parasitic (C)apacitance.

A low-k dielectric reduces the parasitic capacitance.


Regards, Hans

[Oliverda]

http://en.wikipedia.org/wiki/Repeater_insertion

The RC wire delay is the delay through a wire due to it's (R)esistance
and parasitic (C)apacitance.

A low-k dielectric reduces the parasitic capacitance.


Regards, Hans

Thanks.

If I'm not mistaken GF's 32nm SOI process tech will use high-k dielectric instead of low-k dielectric.

[TJ TRICHEESE]

The key words here are: 'up to' and 'depending on CPU model'

[Hans de Vries]

Thanks.

If I'm not mistaken GF's 32nm SOI process tech will use high-k dielectric instead of low-k dielectric.

It will use both

High-k in the HKMG transistors and low-k to embed the wires.


Regards, Hans

[Oliverda]

It will use both

High-k in the HKMG transistors and low-k to embed the wires.


Regards, Hans

Good to know.

Hasn't the AMD/GF utilized the low-k dielectric in any SOI process tech so far?

[Hans de Vries]

Good to know.

Hasn't the AMD/GF utilized the low-k dielectric in any SOI process tech so far?

Yes, for many years actually. This one has a just a lower k as the previous one.
I think it should be at around 2.4 this time.

Regards, Hans

[Manicdan]

The key words here are: 'up to' and 'depending on CPU model'

yes, some models are good for 400mhz, some 500, but sofar i dont know if that means we will always see the max turbo, or no turbo, or if there are stages, but since its a P state, i assume it means its max or no turbo, just how many cores is the only variable

[Oliverda]

Yes, for many years actually. This one has a just a lower k as the previous one.
I think it should be at around 2.4 this time.

Regards, Hans

Thanks again.