Turbo...why so little clock gain?

[boxleitnerb]

I always wonder about the single/dual thread turbo when looking at the majority of CPUs. All the Ivy Bridge CPUs (without T/S-models) increase clock only by about 10%. The S-models 20% and the T-models almost 50%.

Why not have those 50% all the time? Why doesn't my 2600K clock to 5GHz when lightly loaded, but only to 3.8?

Same with AMD (half the modules loaded):
FX8150: 16%
FX6100: 18%
FX4100: 5%

• Intel and AMD cannot go higher without the power consumption going through the roof. Unlikely though, because you have up to 50% of your TDP budget to spend on higher clocks, depending on the model. The T-models show it can be done.
• Intel wants to maximize profits and holds back

So what is it? Or are there other points to consider?

 

[Kenmitch]

For Intel its power consumption and guaranteed reliability topped off with longevity. Not like they test every single CPU and tweak it for vcore and speed.

If you have a K chip just baby over clock it if you want more speed under loads. I think my 2550k will get to about 4.2-4.3ghz before it hits the 95w TDP under something like Intel Burn Test.

 

[boxleitnerb]

You sure about power consumption? I mean single/dual core turbo, not all core turbo. So 50-75% of the CPU should be idle.

 

[ShintaiDK]

So what is it? Or are there other points to consider?

Product lifespan guarantee.

Product reliability.

Working in all envionments.

Power consumption limits.

Heat limits.

And could add alot more. But they dont add as much drama as: "X company holds back on purpose to maximize profit."

 

[greenhawk]

Given this is the second version of turbo and some of the inital turbos where a joke (1 speed bin), then future ones will be hopefully better.

As to the T and S models, the turbos are so much greater as for the same rating, intel has the base clocks far far slower than the standard parts. So not suprise there. Besides, if you check the specs, the similar named parts all turbo to about the same level so not a issue to the end user (who does not know anything about turbo).

Second issue with turbo is the TDP, Turbo is designed to go over this number for a short period of time. Time generally set in the bois and amount (my board defaults to 93W TDP for Ivy) also adjustable on some motherboards

As to why not have all chips turbo a large amount, that is due to chip quality and market segmentation. Same as any feature in a product.

As to intel holding back, just how many cpus at $200/$300 would they sell if they sold a dual core with Hyper threading that could turbo from 2GHz to 5Ghz and stay there for 30 minutes, all for a lowly $50?

 

[ShintaiDK]

Just because you can OC to X Ghz doesnt mean its reliable.

Its like saying the speedlimit should be 200km/h. Because I think I can handle that when driving past schools.

 

[IntelUser2000]

As to the T and S models, the turbos are so much greater as for the same rating, intel has the base clocks far far slower than the standard parts.

The T and S models have higher Turbo gain because they have lower TDP. It's about maximizing responsiveness within a given TDP and in light workloads. The regular voltage models with high base clocks and low Turbo gives you good guaranteed performance.

Notice how we had bigger Turbo gain in the Lynnfield chips compared to Sandy Bridge and Ivy Bridge. The Nehalem/Westmere architecture were less power efficient than Sandy Bridge, so to keep the same 95W TDP, they had to resort to Turbo, which is an opportunistic frequency gain, rather than a guaranteed one. In Sandy Bridge because they improved power efficiency, they were able to increase base clocks.

You sure about power consumption? I mean single/dual core turbo, not all core turbo. So 50-75% of the CPU should be idle.

It is about power usage too. Very much so. Keeping single and dual core Turbo limited means it won't have to use TDP power in most scenarios.

Also consider this. Why should Intel/AMD notch up clock speeds to overclocking maximum when:
-If you need extra performance you can overclock it anyway?
-Really high frequencies mean more verification time required, or increased amount of people complaining about instability(which will happen when you ship hundreds of millions of processors), when again they can overclock it themselves?

 

[Kenmitch]

You sure about power consumption? I mean single/dual core turbo, not all core turbo. So 50-75% of the CPU should be idle.

Yep. The new trend is to smash as much power into the smallest place while using the least amount of energy as possible.

With a single core turbo speed of 5ghz most likely the chip would exceed it's TDP. If Intel shipped chips out of the factory with max turbo of 5ghz you'd never see it anyways as single core turbo speed is almost impossible to achieve.

Not sure how Ivy Bridge compares but my Sandy can suck down some watts once I crank her up!

24/7 overclock @4.8ghz Watt reading is from RealTemp and Hardware Monitor. They read pretty much the same so figure semi-accurate at least. The readings are peak wattage while playing around with Intel Burn Test. The harder the load the higher the wattage.

1 core load = 73w
2 core load = 97w

As you can see the first core is the one that eats all the power. Loading additional cores only adds what that core needs to the equation.

 

[borisvodofsky]

Yep. The new trend is to smash as much power into the smallest place while using the least amount of energy as possible.

With a single core turbo speed of 5ghz most likely the chip would exceed it's TDP. If Intel shipped chips out of the factory with max turbo of 5ghz you'd never see it anyways as single core turbo speed is almost impossible to achieve.

Not sure how Ivy Bridge compares but my Sandy can suck down some watts once I crank her up!

24/7 overclock @4.8ghz Watt reading is from RealTemp and Hardware Monitor. They read pretty much the same so figure semi-accurate at least. The readings are peak wattage while playing around with Intel Burn Test. The harder the load the higher the wattage.

1 core load = 73w
2 core load = 97w

As you can see the first core is the one that eats all the power. Loading additional cores only adds what that core needs to the equation.

What????

the first core doesn't EAT all the power. It's because the power delivery sends the same voltage to all cores.

 

[coffeejunkee]

Given this is the second version of turbo and some of the inital turbos where a joke (1 speed bin), then future ones will be hopefully better.

Bloomfield had 2 speed bin turbo for single core loads, but many never saw it because their mobo had c3/c6 disabled by default. Lynnfield had more aggressive turbo boost than SB/IB speed bin wise, with up to 5 bins for i7 870. But as I understand it SB/IB turbo is better in the sense it can exceed TDP for longer periods. Never really noticed this myself, both my previous i5 750 and my current i5 3570K ran/run full turbo all the time regardless. Might be different in a laptop though.

With a single core turbo speed of 5ghz most likely the chip would exceed it's TDP. If Intel shipped chips out of the factory with max turbo of 5ghz you'd never see it anyways as single core turbo speed is almost impossible to achieve.

You mean with 5GHz? Because I always see max turbo in single core loads.

 

[kevinsbane]

I always wonder about the single/dual thread turbo when looking at the majority of CPUs. All the Ivy Bridge CPUs (without T/S-models) increase clock only by about 10%. The S-models 20% and the T-models almost 50%.

Why not have those 50% all the time? Why doesn't my 2600K clock to 5GHz when lightly loaded, but only to 3.8?

Same with AMD (half the modules loaded):
FX8150: 16%
FX6100: 18%
FX4100: 5%

• Intel and AMD cannot go higher without the power consumption going through the roof. Unlikely though, because you have up to 50% of your TDP budget to spend on higher clocks, depending on the model. The T-models show it can be done.
• Intel wants to maximize profits and holds back

So what is it? Or are there other points to consider?

The T-Models also have a base clock of 2.5 GHz. I'm not sure how you can apply the "50% turbo boost" to a base clock that is already 36% faster. It doesn't get the 50% turbo from a higher turbo mode, but from a lower base mode.

As for a 5ghz single thread turbo boost; I doubt Intel can get their desired power, reliability, stability and thermal characteristics within their specified TDP limits with a 5GHz turbo.

 

[Smoblikat]

I didnt even know anyone used that stupid turbo boost. its virtually useless on a desktop since you can just OC to whatever speeds you want and still have all of your cores.

 

[beginner99]

Bloomfield had 2 speed bin turbo for single core loads, but many never saw it because their mobo had c3/c6 disabled by default. Lynnfield had more aggressive turbo boost than SB/IB speed bin wise, with up to 5 bins for i7 870. But as I understand it SB/IB turbo is better in the sense it can exceed TDP for longer periods. Never really noticed this myself, both my previous i5 750 and my current i5 3570K ran/run full turbo all the time regardless. Might be different in a laptop though.

Yeah my i7-875k at stock always ran at 3.2 Ghz (2.93 Ghz offical without turbo) under BOINC with HT on and all 8 Threads fully loaded. maybe it also depends on CPU temp? No idea.

 

[boxleitnerb]

The T-Models also have a base clock of 2.5 GHz. I'm not sure how you can apply the "50% turbo boost" to a base clock that is already 36% faster. It doesn't get the 50% turbo from a higher turbo mode, but from a lower base mode.

As for a 5ghz single thread turbo boost; I doubt Intel can get their desired power, reliability, stability and thermal characteristics within their specified TDP limits with a 5GHz turbo.

Why not? Why should turbo be only enabled on by default lower clocked CPUs?

Compare these two:
2500T: base clock 2.3GHz, turbo clock for 2 threads: 3.2GHz, TDP 45W
2500K: base clock 3.3GHz, turbo clock for 2 threads: 3.6GHz, TDP 95W

Why shouldn't the ratio of base clock/turbo clock stay the same? Pay more, get more. You can overclock manually, but then you are constrained by stability limits for ALL threads. If only two threads are loaded, there is significantly more TDP headroom, allowing for higher voltage and higher clock. I don't see a problem here to turbo within reason. At least 30%, because you can actually feel 30%. Forget the measly 10% the Intel turbo currently has on most models, that is only cosmetic and basically useless. No one notices 10% in real world use.

 

[Kenmitch]

What????

the first core doesn't EAT all the power. It's because the power delivery sends the same voltage to all cores.

That's the whole point. Hitting 1 core with a heavy sustained load pretty much can push the chip close to it's TDP depending on the speed the chip is running at.

 

[kevinsbane]

Why not? Why should turbo be only enabled on by default lower clocked CPUs?

It's not.

Compare these two:
2500T: base clock 2.3GHz, turbo clock for 2 threads: 3.2GHz, TDP 45W
2500K: base clock 3.3GHz, turbo clock for 2 threads: 3.6GHz, TDP 95W

Why shouldn't the ratio of base clock/turbo clock stay the same? Pay more, get more.

Because considering the technical costs of doing so, it is not economical to actually do it. Besides of which, you're not paying for turbo. You're paying for performance. I guess for you, Intel should have sold you a 2.4 GHz 2500k so you could get your 50% turbo.

 

[IntelUser2000]

Why shouldn't the ratio of base clock/turbo clock stay the same? Pay more, get more. You can overclock manually, but then you are constrained by stability limits for ALL threads.

You can do the same you are proposing with overclocking. You don't have to overclock all of them. There are motherboards with options in BIOS so you can set Turbo based on amount of cores active.

I don't see a problem here to turbo within reason. At least 30%, because you can actually feel 30%. Forget the measly 10% the Intel turbo currently has on most models, that is only cosmetic and basically useless. No one notices 10% in real world use.

No, really? That's why you can overclock it. They are here to sell you parts, and they sell millions and millions of them. Even if 1% of the chips stop working, that's heck of a lot. So they can just bin it conservatively and let users decide whether higher clocks are worth it(aka overclocking), add a note on the box saying overclocking voids warranty blah blah blah and avoid complaints.

The Pentium III 1.13GHz was recalled because they went over the limit of reliable operation. Sure, you've seen overclocks of 1.2-1.3GHz with the previous 1GHz part. But the very fact that it was recalled since it was clocked too high at only 13% higher(rather than 20-30% with overclocks) shows overclocking and stability with millions of parts are totally different animals.

 

[boxleitnerb]

It's not.


Because considering the technical costs of doing so, it is not economical to actually do it. Besides of which, you're not paying for turbo. You're paying for performance. I guess for you, Intel should have sold you a 2.4 GHz 2500k so you could get your 50% turbo.

I typed too quickly. I meant why not have higher turbo on the non-T models.
Lol, that is not what I want. I want a turbo for single/dual threads of 30+% for every quad-thread CPU model, not just the T-models.

You can do the same you are proposing with overclocking. You don't have to overclock all of them. There are motherboards with options in BIOS so you can set Turbo based on amount of cores active.

No, really? That's why you can overclock it. They are here to sell you parts, and they sell millions and millions of them. Even if 1% of the chips stop working, that's heck of a lot. So they can just bin it conservatively and let users decide whether higher clocks are worth it(aka overclocking), add a note on the box saying overclocking voids warranty blah blah blah and avoid complaints.

The Pentium III 1.13GHz was recalled because they went over the limit of reliable operation. Sure, you've seen overclocks of 1.2-1.3GHz with the previous 1GHz part. But the very fact that it was recalled since it was clocked too high at only 13% higher(rather than 20-30% with overclocks) shows overclocking and stability with millions of parts are totally different animals.

I know about the changeable turbo ratios. However, I cannot change voltage for each of these turbo modes separately. Meaning I would vastly overshoot the TDP when 3-8 threads are used. That is not the purpose of the turbo. In my opinion, it should use all the TDP in every situation (1-8 thread load) and adjust clocks and voltages of all cores appropriately and quickly. Maybe make this not the default setting but an option in the bios.

My reasoning behind this is, that there are applications/games where much CPU power is needed but not all cores are utilized. Think about the Total War series, that only effectively uses 2 threads. What else is the turbo for, if not for that? But the implementation is too half-hearted for my taste. I don't expect double the clocks for half the threads, I know that is unrealistic. But 10% is not enough.

 

[IntelUser2000]

I don't expect double the clocks for half the threads, I know that is unrealistic. But 10% is not enough.

That may be so, but if the clocks you are proposing cause Pentium III 1.13GHz recalls again then it'd not be worth it at all. High overclocking headroom could also be translated as very low headache for the CPU manufacturer.

 

[boxleitnerb]

The Pentium III 1.13GHz was a single core CPU. The whole CPU had to run at that clock and be stable. Here only 1/4 or 1/2 is required.

 

[IntelUser2000]

The Pentium III 1.13GHz was a single core CPU. The whole CPU had to run at that clock and be stable. Here only 1/4 or 1/2 is required.

I think you are missing the point. It's about reliably operating at set frequencies for long time(at least warranty period) with hundreds of millions of processors. Pentium III 1.13GHz is a good example of over-speccing it.

 

[kevinsbane]

I typed too quickly. I meant why not have higher turbo on the non-T models.
Lol, that is not what I want. I want a turbo for single/dual threads of 30+% for every quad-thread CPU model, not just the T-models.

Technical infeasibility within a given cost limitation.