Intel i5 non K clocks - why such a large increase with KBL?

[escrow4]

I was wondering today after getting properly liquored up, KBL has a steep boost in clocks for the highest locked i5s:

i5 2500 non K - 4C - 3.4GHz | 1C - 3.7GHz
i5 3570 non K - 4C - 3.6GHz | 1C - 3.8GHz
i5 4670 non K - 4C - 3.6GHz | 1C - 3.8GHz
i5 4690 non K - 4C - 3.7GHz | 1C - 3.9GHz
i5 6600 non K - 4C - 3.6GHz | 1C - 3.9GHz
i5 7600 non K - 4C - 3.9GHz | 1C - 4.1GHz

Now what Intel giveth, Intel also taketh, but after hovering around the same clocks KBL got coked up. Why? Will this continue for the next i5 or will clocks droppeth once more?

 

[LTC8K6]

Intel doesn't seem to list official 4 core turbo speeds for KL chips.

I think overall the answer is that KL's 14nm+ process is indeed more efficient.

 

[dullard]

That is why I think the 7600 is probably the best Intel chip value for most users. Sadly, non-custom OEMs (Dell, HP, Lenovo, etc) don't sell it.

 

[sm625]

It's not that big of an increase considering its the same architecture as skylake. We are forgetting that the i5-7600 should have been a 10nm chip, with 10% more IPC, clocked at probably 3.7GHz. Instead of that, they just bumped the frequency by 10%. In the end I guess it is the same thing.

 

[LTC8K6]

It's not that big of an increase considering its the same architecture as skylake. We are forgetting that the i5-7600 should have been a 10nm chip, with 10% more IPC, clocked at probably 3.7GHz. Instead of that, they just bumped the frequency by 10%. In the end I guess it is the same thing.

Well, they didn't just bump the frequency.

The 14nm+ process must have allowed the bump in frequency and considerably better turbo performance.

 

[imported_jjj]

I was wondering today after getting properly liquored up, KBL has a steep boost in clocks for the highest locked i5s:

i5 2500 non K - 4C - 3.4GHz | 1C - 3.7GHz
i5 3570 non K - 4C - 3.6GHz | 1C - 3.8GHz
i5 4670 non K - 4C - 3.6GHz | 1C - 3.8GHz
i5 4690 non K - 4C - 3.7GHz | 1C - 3.9GHz
i5 6600 non K - 4C - 3.6GHz | 1C - 3.9GHz
i5 7600 non K - 4C - 3.9GHz | 1C - 4.1GHz

Now what Intel giveth, Intel also taketh, but after hovering around the same clocks KBL got coked up. Why? Will this continue for the next i5 or will clocks droppeth once more?

Because Kaby Lake doesn't offer any increases in IPC , to make it better than the 6600, they could only boost clocks.

If you do the math, the gain is small, 8.3% in base clocks and 5.1% in max turbo.

 

[dullard]

Because Kaby Lake doesn't offer any increases in IPC, to make it better than the 6600, they could only boost clocks.

If you do the math, the gain is small, 8.3% in base clocks and 5.1% in max turbo.

The specifications gain is even smaller than that in simple terms. Base 7600 is 3.5 GHz, base 6600 is 3.3 GHz, a 6.1% gain not 8.3%. However, the 7600 should spend significantly more time at turbo (up to 4.1 GHz) when the 6600 has to throttle back to base (3.3 GHz). In those moderately heavy use cases, the 7600 is 24.2% faster. Not great, but not bad for basically the same price.

Plus, if they are relevant to your needs, HD 630 is 11% faster than HD 530 in benchmarks and you have access to Optane.

 

[imported_jjj]

The specifications gain is even smaller than that in simple terms. Base 7600 is 3.5 GHz, base 6600 is 3.3 GHz, a 6.1% gain not 8.3%. However, the 7600 should spend significantly more time at turbo (up to 4.1 GHz) when the 6600 has to throttle back to base (3.3 GHz). In those moderately heavy use cases, the 7600 is 24.2% faster. Not great, but not bad for basically the same price.

Plus, if they are relevant to your needs, HD 630 is 11% faster than HD 530 in benchmarks and you have access to Optane.

I was going by the clocks the OP listed.
The 24.2% is utterly absurd , you just to the math for 3.3GHz vs 4.1GHz, because somehow the 6600 has no turbo at all and the 7600 has 4.1GHz turbo "in moderately heavy loads"

 

[dullard]

I was going by the clocks the OP listed.

The OP didn't list base clocks though.

The 24.2% is utterly absurd , you just to the math for 3.3GHz vs 4.1GHz, because somehow the 6600 has no turbo at all and the 7600 has 4.1GHz turbo "in moderately heavy loads"

There are three possible scenarios:

1) The chips are both cool and they run at turbo. For single core uses that would be 4.1 GHz vs 3.9 GHz, a 5.1% difference.

2) The chips are both hot and they run at base. That would be 3.5 GHz vs 3.3 GHz, a 6.1% difference.

3) The main benefit of Kaby Lake is that it runs at turbo more often, for longer amount of time, and ramps to turbo faster than Skylake. This is when a chip is under load, but not yet at a hot state (I called that moderately heavy load). In this situation, the Skylake chip has to down throttle to 3.3 GHz because it is too hot when the Kaby Lake stays at turbo 4.1 GHz, a 24.2% increase (single core). For quad core this drops to 3.9 GHz vs 3.3 GHz, an 18.2% difference.

How much of the time you spend at scenario #3 depends on your particular usage. For some people it will be only a short amount of time. But, it does happen.

There are not a lot of 7600 vs 6600 benchmarks. Based on specifications only, you would think there would be just a 5% or 6% difference in speed. But, in benchmarks, the difference is often more than that. For example the 7600 is 15.1% faster in PassMark:
https://www.cpubenchmark.net/cpu.php?cpu=Intel+Core+i5-7600+@+3.50GHz&id=2920
vs.
https://www.cpubenchmark.net/cpu.php?cpu=Intel+Core+i5-6600+@+3.30GHz&id=2594
Explain that with just the clock speeds.

In FutureMark, the difference is 10.6%. Again this is more than you would get just from clock speeds alone since the 7600 spends more time at turbo than the 6600 does. https://www.futuremark.com/hardware/cpu/Intel+Core+i5-7600/review

 

[imported_jjj]

The OP didn't list base clocks though.

There are three possible scenarios:

1) The chips are both cool and they run at turbo. For single core uses that would be 4.1 GHz vs 3.9 GHz, a 5.1% difference.

2) The chips are both hot and they run at base. That would be 3.5 GHz vs 3.3 GHz, a 6.1% difference.

3) The main benefit of Kaby Lake is that it runs at turbo more often, for longer amount of time, and ramps to turbo faster than Skylake. This is when a chip is under load, but not yet at a hot state (I called that moderately heavy load). In this situation, the Skylake chip has to down throttle to 3.3 GHz because it is too hot when the Kaby Lake stays at turbo 4.1 GHz, a 24.2% increase (single core). For quad core this drops to 3.9 GHz vs 3.3 GHz, an 18.2% difference.

How much of the time you spend at scenario #3 depends on your particular usage. For some people it will be only a short amount of time. But, it does happen.

There are not a lot of 7600 vs 6600 benchmarks. Based on specifications only, you would think there would be just a 5% or 6% difference in speed. But, in benchmarks, the difference is often more than that. For example the 7600 is 15.1% faster in PassMark:
https://www.cpubenchmark.net/cpu.php?cpu=Intel+Core+i5-7600+@+3.50GHz&id=2920
vs.
https://www.cpubenchmark.net/cpu.php?cpu=Intel+Core+i5-6600+@+3.30GHz&id=2594
Explain that with just the clock speeds.

In FutureMark, the difference is 10.6%. Again this is more than you would get just from clock speeds alone since the 7600 spends more time at turbo than the 6600 does. https://www.futuremark.com/hardware/cpu/Intel+Core+i5-7600/review

Full throttle eh?
You never encounter scenario 3 and you know it. Maybe read again your claim before reacting and repeating it.
As for your so called benchmarks the explanation is simple , different system configs and not a CPU only benchmark.
Only replied for the sake of others, won't engage you again so you can keep going.

 

[escrow4]

Full throttle eh?
You never encounter scenario 3 and you know it. Maybe read again your claim before reacting and repeating it.
As for your so called benchmarks the explanation is simple , different system configs and not a CPU only benchmark.
Only replied for the sake of others, won't engage you again so you can keep going.

Scenario 3 would be gaming. A multi threaded title (like Andromeda) would hammer the cores at 3.9GHz and keep them there and Kaby lake would be less likely to throttle.

 

[LTC8K6]

Plus, if they are relevant to your needs, HD 630 is 11% faster than HD 530 in benchmarks and you have access to Optane.

HD530 and HD630 are identical in graphics performance, clock for clock. You will see no performance difference in desktops, and the difference in mobile is due to 630 being able to run faster for longer periods. That is. clock speeds. The only other difference is in video encoding/decoding where 630 has more features.

 

[jihe]

I was wondering today after getting properly liquored up, KBL has a steep boost in clocks for the highest locked i5s:

i5 2500 non K - 4C - 3.4GHz | 1C - 3.7GHz
i5 3570 non K - 4C - 3.6GHz | 1C - 3.8GHz
i5 4670 non K - 4C - 3.6GHz | 1C - 3.8GHz
i5 4690 non K - 4C - 3.7GHz | 1C - 3.9GHz
i5 6600 non K - 4C - 3.6GHz | 1C - 3.9GHz
i5 7600 non K - 4C - 3.9GHz | 1C - 4.1GHz

Now what Intel giveth, Intel also taketh, but after hovering around the same clocks KBL got coked up. Why? Will this continue for the next i5 or will clocks droppeth once more?

Because ipc increase was nonexistent.

 

[dullard]

HD530 and HD630 are identical in graphics performance, clock for clock. You will see no performance difference in desktops, and the difference in mobile is due to 630 being able to run faster for longer periods. That is. clock speeds. The only other difference is in video encoding/decoding where 630 has more features.

Desktop difference in graphical commands is 11% in this synthetic benchmark:
http://gpu.userbenchmark.com/Compar...-Intel-HD-530-Desktop-Skylake/m178724vsm33102

 

[dullard]

You never encounter scenario 3 and you know it.

Other than Intel itself saying that scenario 3 exists, and they advertise it.
http://images.anandtech.com/doci/10968/1-7.png

Only replied for the sake of others, won't engage you again.

Good, we don't need uneducated people spreading misinformation.

 

[LTC8K6]

Desktop difference in graphical commands is 11% in this synthetic benchmark:
http://gpu.userbenchmark.com/Compar...-Intel-HD-530-Desktop-Skylake/m178724vsm33102

I can't recall reviews showing an 11% increase in gaming performance. It would have been news that 630 was 11% faster in games than 530 clock for clock.

 

[dullard]

I can't recall reviews showing an 11% increase in gaming performance. It would have been news that 630 was 11% faster in games than 530 clock for clock.

There just aren't many reviews that compare them. More synthetics on SPECviewperf 12 Geometric Mean (do note this is the Xeon P630 vs P530 at the same clocks although the comparison should be similar):
http://www.intel.com/content/www/us...d-graphics/hd-graphics-performance-guide.html
9.4% gain and 10.6% gain.

The very few gaming benchmarks I can find are closer to 2.7% to 5.7% gain.

 

[LTC8K6]

Can we rule out clock speed differences, though? KL may be able to boost the IGP for slightly longer periods, even in a desktop.

As far as I know, no one has ever pointed out architectural differences between HD530 and HD630.

All I can see is that 630 apparently supports Open CL 2.1 whereas 530, and all other current graphics chips, are only 2.0 or below.

 

[dullard]

Can we rule out clock speed differences, though? KL may be able to boost the IGP for slightly longer periods, even in a desktop.

Ok, we have a definition difference here. I was saying same clock speeds as in "same limits on the spec sheet". You are saying "running at the same speed for the same amount of time on both chips during the benchmark test".

Yes, they have the same limits on the specification sheets. But KL runs in a faster boost mode more often and for longer periods than SL. No, there isn't an architectural difference between the two other than the fin tweak to produce less waste heat (which allows for more time in boost mode).

The intensive games would produce more heat and thus be in base speeds more often (the 3% to 5% gains) vs synthetics that can isolate the GPU and run at max speeds more often (the ~11% gains).