Trinity; 5.1Ghz achieved on air.

[Frakk]

First response the the question of Piledrivers overclocking ability.

http://hexus.net/tech/news/cpu/46197-amd-trinity-a10-5800k-apu-overclocked-73ghz/

It would be great if we see plenty of something that around the forums from users.

But how usable is it at that.

 

[EarthDog]

Nice! I also wonder if that is 'benchmark' stable or 'daily driver' stable.

Question for someone in the know - Why is the voltage so high with AMD CPU's? 1.45 for 4.5Ghz seems a lot. Even my TERRIBLE IB does 1.28v at 4.5Ghz. I have to assume its in the architecture or something?

 

[ssjwizard]

AMD chips have always used a slightly higher vCore than Intel chips as far as I recall. Its just part of the design, Im unsure if it has to do with architecture or transistor design.

 

[BeepBeep2]

This 5.1 GHz is only "CPU-Z" stable. Not stable at all, but CPU-Z Validation only for HWBOT.

Daily driver stable, clocks are to be expected around 4.2-4.4 GHz.
You might think, "doesn't turbo VID go up to 1.475v possibly, and isn't turbo 4.2 GHz?" but about 1.45-1.475v is all that is recommended AT ALL for this architecture and manufacturing process.

AMD is pushing clocks a lot here, and just like BD release I'm sure there will be defective CPUs right off the bat (unstable at stock without LLC on, you know, Aero problems and failing Prime95) ...aaaand I know a few guys that have tested several chips and average stable clocks are about 4.4 GHz which is about ~200 MHz lower than Zambezi (Bulldozer FX).

EarthDog, FX-41/61/81xx series CPUs have needed about 1.375-1.45v on average for 4.5 GHz. I don't believe the 1.45v for 4.5 GHz there is even stable, but more "will run a bench test suite for hours, then crash randomly at idle sometime" stable ...so most of the way there.

Intel's IB is a different animal, it is a 22nm CPU, AMD's Bulldozer and Piledriver CPUs are on the 32nm node. Why it needs less voltage at 22nm should make a little more sense after you read the PM I sent you.

 

[Robert17]

AMD chips have always used a slightly higher vCore than Intel chips as far as I recall. Its just part of the design, Im unsure if it has to do with architecture or transistor design.

It's been that way a long time. AMDs always were more troublesome to keep cool than Intel in my experience.

 

[EarthDog]

Ugh, I need to go home. For some reason I thought BD was on 22nm.. that explains it a bit.

But hell, even when Intel was on 32nm process, 2600k, 1.45v at 4.5Ghz was a TERRIBLE chip. 1.35, usually less, was about right.

 

[BeepBeep2]

Ugh, I need to go home. For some reason I thought BD was on 22nm.. that explains it a bit.

But hell, even when Intel was on 32nm process, 2600k, 1.45v at 4.5Ghz was a TERRIBLE chip. 1.35, usually less, was about right.

Yeah, but AMD's FX-8100 series is a bit harder to cool, overclock a little less (you won't see stable 8150s over 4.8 GHz, and that is on $500 water cooling) and use a lot more power than a 2600K

Then the FX 4 core and 6 cores are just castrated 8 cores most of the time, so they do not clock as well as the 8 cores regardless, some needing pushing upwards of 1.5v for 4.5-4.6.

My first FX-8150 needed 1.43v for 4.6 GHz, but was unstable at stock frequency and stock voltage, also did not scale for benching with voltage over 1.45v.
My second FX-8150 needed 1.41v for 4.6 GHz, scaled for benching up through 1.55v, and just barely gets too hot for 4.8 GHz (1.48v) stable on water.

 

[Robert17]

This is interesting. I can't read German very well (from the follow up link) so don't know if (guessing it would have to be) LN2 if not performed on Pluto.

http://hardocp.com/news/2012/10/04/amd_trinity_a105800k_apu_overclocked_to_73ghz

 

[BeepBeep2]

Yes, LN2.
It was done in Japan.

Here is PcCI2iminal's (the overclocker who achieved this') thread in English at Xtremesystems.org

The source that the (sometimes quite "slow" media) are posting about at PCGH.de is simply a repost of the original.

 

[Frakk]

This 5.1 GHz is only "CPU-Z" stable. Not stable at all, but CPU-Z Validation only for HWBOT.

Daily driver stable, clocks are to be expected around 4.2-4.4 GHz.
You might think, "doesn't turbo VID go up to 1.475v possibly, and isn't turbo 4.2 GHz?" but about 1.45-1.475v is all that is recommended AT ALL for this architecture and manufacturing process.

AMD is pushing clocks a lot here, and just like BD release I'm sure there will be defective CPUs right off the bat (unstable at stock without LLC on, you know, Aero problems and failing Prime95) ...aaaand I know a few guys that have tested several chips and average stable clocks are about 4.4 GHz which is about ~200 MHz lower than Zambezi (Bulldozer FX).

EarthDog, FX-41/61/81xx series CPUs have needed about 1.375-1.45v on average for 4.5 GHz. I don't believe the 1.45v for 4.5 GHz there is even stable, but more "will run a bench test suite for hours, then crash randomly at idle sometime" stable ...so most of the way there.

Intel's IB is a different animal, it is a 22nm CPU, AMD's Bulldozer and Piledriver CPUs are on the 32nm node. Why it needs less voltage at 22nm should make a little more sense after you read the PM I sent you.

Daily driver stable, clocks are to be expected around 4.2-4.4 GHz.

Well that sux, it's even far worse than Bulldozer, hell 4.2 - 4.4, if that's the case it doesn't overclock at all, absolutely pointless being unlocked.

That will be the next bashing target then, IB overclocks worse than SB, Piledriver almost no overclocking at all (at best as if it is unstable at stock with LLC then your would have to down clock it)

I'm running my 2 year old Phenom at 4.2 for crying out loud.

Are you sure about that? it would make it a worse clocker than Phenom II, a step down from Bulldozer in every way

This can't possibly get any worse.

 

[bmwbaxter]

These chips using the resonate clock mesh? Not sure if it has been implemented or not yet, but IIRC I heard it being used conversations regarding AMD.

 

[BeepBeep2]

Well that sux, it's even far worse than Bulldozer, hell 4.2 - 4.4, if that's the case it doesn't overclock at all, absolutely pointless being unlocked.

That will be the next bashing target then, IB overclocks worse than SB, Piledriver almost no overclocking at all (at best as if it is unstable at stock with LLC then your would have to down clock it)

I'm running my 2 year old Phenom at 4.2 for crying out loud.

Are you sure about that? it would make it a worse clocker than Phenom II, a step down from Bulldozer in every way

This can't possibly get any worse.

4.4 is a target. 4.6 pushing things. Mind you, BD usually only does about 4.6 stable on average so this is about 200 MHz worse.

Second, no, it will not be the next bashing target. The stock speed is 3.8 GHz. It does overclock, and if chips come out that are unstable at stock WITHOUT LLC then they are defective and will be RMA'ed. Some FX-8150s acted that way, I saw a few on newegg reviews and I had one that also did that. (1.18v after vdroop, with 1.275v VID, was unstable at 3.6 GHz, defective CPU.)

Congrats. 4.2-4.3 on Phenom II is about as good as it gets when you overclock within limits of safety and longevity.

Yes, I'm sure. I don't know if you understand how these chips work, but the Turbo is at 4.2 GHz, at up to 1.475v. The turbo voltage will be anywhere between 1.4 and 1.475v. 1.475v is the maximum amount of voltage anyone should put into the CPU however if they care at all about safety and longevity. That means "If you want to ensure your CPU lives or does not degrade significantly over more than a year, or at least a decent amount of time, like multiple years". Make sense?

Yes I'm sure about overclock targets. 4.2-4.6. There are FX-4100s that barely hit 4.4 GHz. Don't you know?
No, this is not a worse clocker than Phenom II. If you do simple math, 4.4 average is higher than...about 4.0 average. This is not a replacement for Phenom II either...so I don't know what you're thinking about. Furthermore you're mixing terms. ("Phenom II" is to "FX" as "STARS" is to "Bulldozer")...("Llano is to Trinity" as "Deneb" is to "Zambezi", as "Zambezi" is to "Vishera") We are talking about TRINITY , the APU part with 2 CUs, 4 "cores" and an iGPU and no L3 cache... this is a mainstream part that is optimized for power efficiency, not overclocking or brute performance, and a CPU TDP of about 70w+30w GPU TDP = 100w TDP.

This can't get any worse? Don't get ahead of yourself, Frakk. Don't you remember Llano, you know, the A8-3850, A8-3870 that have STARS cores but only hit 3.4-3.6 GHz max overclock on FM1 and the 32nm node?

These chips using the resonate clock mesh? Not sure if it has been implemented or not yet, but IIRC I heard it being used conversations regarding AMD.

Yes, which is why clocks are up and TDP is lower. Overclocking not quite as well as FX-4/6/8 series on AM3+ but that is a completely different platform.

 

[Frakk]

The thing boosts to 4.2Ghz, so overclocking is basically turning it to permanent boost

If the Piledriver Vishera core at about the same performance clock for clock as Phenom II and only 4.2 / 4.4 Ghz then there is no upgrade path from Bulldozer or Phenom II.

Square one.

 

[Aldakoopa]

Yes I'm sure about overclock targets. 4.2-4.6. There are FX-4100s that barely hit 4.4 GHz. Don't you know?

I've been afraid to push mine past 4.2, I'm still an OC noob. I have it running cool, stable, more than fast enough for what I need it for, and I don't want to change that.

I was hoping PD would be even more of an overclocking monster than BD. However, keep in mind people that this is Trinity! This is AMD's budget-oriented A-Series APUs that, even though they are using Piledriver's core architecture, the actual enthusiast-level PD chips may actually be higher-binned and much better overclockers, or at least we can all hope.

 

[BeepBeep2]

I'm not going to argue with your idiosyncrasy Frakk...but I'll throw this at you for a second time.

This time I'll break it up into chunks...

4.4 is a target. 4.6 pushing things. Mind you, BD usually only does about 4.6 stable on average so this is about 200 MHz worse.

Second, no, it will not be the next bashing target.

The stock speed is 3.8 GHz.

It does overclock, and if chips come out that are unstable at stock WITHOUT LLC then they are defective and will be RMA'ed.
Some FX-8150s acted that way, I saw a few on newegg reviews and I had one that also did that. (1.275v VID, 1.18v after vdroop, was unstable at 3.6 GHz 1.18v, so defective CPU.)

Congrats. 4.2-4.3 on Phenom II is about as good as it gets when you overclock within limits of safety and longevity.

Yes, I'm sure. I don't know if you understand how these chips work, but the Turbo is at 4.2 GHz, at up to 1.475v. The turbo voltage will be anywhere between 1.4 and 1.475v.

1.475v is the maximum amount of voltage anyone should put into the CPU however if they care at all about safety and longevity.

That means "If you want to ensure your CPU lives or does not degrade significantly over more than a year, or at least a decent amount of time, like multiple years". Make sense?

Yes I'm sure about overclock targets. 4.2-4.6. There are FX-4100s that barely hit 4.4 GHz. Don't you know?

No, this is not a worse clocker than Phenom II. If you do simple math, 4.4 average is higher than...about 4.0 average.

This is not a replacement for Phenom II either...so I don't know what you're thinking about.

Furthermore you're mixing terms. ("Phenom II" is to "FX" as "STARS" is to "Bulldozer")...("Llano is to Trinity" as "Deneb" is to "Zambezi", as "Zambezi" is to "Vishera")

We are talking about TRINITY , the APU part with 2 CUs, 4 "cores" and an iGPU and no L3 cache...

...this is a mainstream part that is optimized for power efficiency, not overclocking or brute performance, and a CPU TDP of about 70w+30w GPU TDP = 100w TDP.

This can't get any worse? Don't get ahead of yourself, Frakk. Don't you remember Llano, you know, the A8-3850, A8-3870 that have STARS cores but only hit 3.4-3.6 GHz max overclock on FM1 and the 32nm node? The ones that are still selling?

Overclocking not quite as well as FX-4/6/8 series on AM3+ but that is a completely different platform.

I don't think you're using any logic when you think about this...
Trinity is not Vishera.

 

[Frakk]

I'm not going to argue with your idiosyncrasy...but I'll throw this at you for a second time.
We're talking about Trinity, not Vishera.

Trinity has no L3 cache and half the cores. How many times must I repeat myself.
Clock for clock Trinity loses to Llano but overclocks to 4.4 GHz vs 3.4-3.5 GHz of Llano. Llano is improved (5% IPC increase) Phenom II X4 on 32nm where it did only 3.4 GHz vs 4-4.2 GHz.

I don't think you're using any logic when you think about this...

How much difference it L3 likely to make?

My problem is 'if' we can't expect higher performance clock for clock on Vishera then to give us an upgrade path it must run at higher clocks.

Which it does (Stock) yet if my CPU being 3.2Ghz stock and (lets for argument say 4Ghz overclocked 24/7) for me to justify spending £180 or whatever it is on it it would have to overclock significantly higher, 4.6Ghz on my cooler is what i would want.

I'm not side stepping for that sort of money, and i doubt i'm alone in that.

Let me ask you this, what do you think the chances are of it hitting 4.6Ghz on my cooler?

 

[BeepBeep2]

L3 can make a difference of 0 to 10% (Over Trinity). To give you an idea, go look at the difference in performance between Athlon II and Phenom II. They are exactly the same, one lacking L3 cache.

Performance clock for clock on Vishera vs Zambezi is expected to be about 5% higher.
Overall performance is estimated to be 10% to 15% higher overall, including multi-threading.

Clock speeds are supposed to be 4 GHz / 4.2 GHz turbo for the FX-8350. This is due to 4 GHz stock frequency over the FX-8150 which is a 3.6 GHz CPU, plus estimated 5% IPC increase resulting in up to the full 15% performance increase.

...
Why are you asking me the chances of overclocking a CPU that is not released, and has not been tested publicly and then even more specifically, a certain cooler that I don't have experience with?
That question is no different than asking what kind of overclock you will get out of Steamroller...how would I know the behaviors of a CPU that does not have any behaviors yet?

 

[Frakk]

L3 can make a difference of 0 to 10% (Over Trinity). To give you an idea, go look at the difference in performance between Athlon II and Phenom II. They are exactly the same, one lacking L3 cache.

Performance clock for clock on Vishera vs Zambezi is expected to be about 5% higher.
Overall performance is estimated to be 10% to 15% higher overall, including multi-threading.

Clock speeds are supposed to be 4 GHz / 4.2 GHz turbo for the FX-8350. This is due to 4 GHz stock frequency over the FX-8150 which is a 3.6 GHz CPU, plus estimated 5% IPC increase resulting in up to the full 15% performance increase.

...
Why are you asking me the chances of overclocking a CPU that is not released, and has not been tested publicly and then even more specifically, a certain cooler that I don't have experience with?
That question is no different than asking what kind of overclock you will get out of Steamroller...how would I know the behaviors of a CPU that does not have any behaviors yet?

So you see my point, i would need to get pretty high clocks out of it to be a viable upgrade.

 

[BeepBeep2]

No, I don't see your point at all.
I told you 10-15% performance increase at stock over FX-8150 at same TDP.

That is ALL I can tell you right now. We don't know a damned thing about how Vishera overclocks.

 

[Frakk]

The problem is this



15% does not catch up with what i have.