Exclusive: First benchmarks of Intel's Lakefield Hybrid CPU - Innovative or unnecessary?

[Redaktion]

The Samsung Galaxy Book S with Intel's new Lakefield Hybrid processor has arrived in our office. We have first benchmark results for the new hybrid CPU.

https://www.notebookcheck.net/Exclusive-First-benchmarks-of-Intel-s-Lakefield-Hybrid-CPU-Innovative-or-unnecessary.477524.0.html

[dasodkoas]

By looking at your results I am pretty much certain the scheduler is only using the small cores for now and will get an update later on. So, in this light, I think the results are pretty good and will get even better when the big core will start to be used.

[piokos]

These results are meaningless without the Windows version.
Scheduling for Lakefield will be added in W10X.

If this uses existing W10, you're essentially testing a 4-core Atom.

[Joel]

By looking at your results I am pretty much certain the scheduler is only using the small cores for now and will get an update later on. So, in this light, I think the results are pretty good and will get even better when the big core will start to be used.

It should be possible to confirm this by monitoring HWINFO while the benchmarks are running.

[Andreas Osthoff]

These results are meaningless without the Windows version.
Scheduling for Lakefield will be added in W10X.

If this uses existing W10, you're essentially testing a 4-core Atom.

We ran the benchmarks with Windows 10 Home that was shipped with the device. Windows 10X is probably not coming in 2020, so it does not make much sense to wait for it. If Intel releases the chip, they should take care of the software support as well.

The faster Sunny Cove Core is definitely working, otherwise we would not see clocks of up to 2.4 GHz.

[DavidC1]

We ran the benchmarks with Windows 10 Home that was shipped with the device. Windows 10X is probably not coming in 2020, so it does not make much sense to wait for it. If Intel releases the chip, they should take care of the software support as well.

The faster Sunny Cove Core is definitely working, otherwise we would not see clocks of up to 2.4 GHz.

Andreas, that's not right. The Tremont core can clock up to 2.8GHz in ST. It was revealed by Andreas Schilling on twitter.

You can see from the Task Manager one core is permanently off.

[Andreas Osthoff]

You are correct, Tremont can reach up to 2.8 GHz via Boost. However, Cinebench Multi still uses 5 Threads during the calculation, so the core is not completely off.

[Jack O.]

Can you run some benchmarks in Linux as well?

I'm wondering whether the Linux kernel is supporting the chip fully already. If it does, it should give us a better sense of what kind of performance Samsung / Windows are leaving on the table.

(Plus, the only reason I'd get this over the ARM version is for the better Linux compatibility)

[_MT_]

You are correct, Tremont can reach up to 2.8 GHz via Boost. However, Cinebench Multi still uses 5 Threads during the calculation, so the core is not completely off.

Number of threads in itself means nothing. It could be spawning 100 threads and they could all run on the same core. It's a quite reasonable explanation that the scheduler isn't utilizing it properly. It wouldn't be the first time Windows scheduler was caught with its pants around its ankles, so to speak. Threadripper also wasn't properly supported. Standard desktop Windows have never encountered a big-little configuration nor such a funky number of cores. You could try asking Microsoft, you know. Or Intel. Is it normal, what time frame can we expect for a patch, etc.

While Intel might have access to Windows source code and can provide samples, it's ultimately in the hands of Microsoft and there is nothing they can do about it. With Linux, they can contribute any necessary changes. It's also not their fault if Samsung supplies their device with an OS that's not fully compatible. It's their decision to sell a half-baked product.

[justin6]

Even once the scheduling issues are fixed with Windows 10X, it looks like Intel and Qualcomm will just be competing to create a poor-man's Apple Silicon. The 8cx is almost two years old now, the next version of it will likely be much faster than Lakefield for native ARM code and comparable from emulated x86, which is a bad look for Intel once Microsoft finishes their x64 to ARM64 translator.

While I much prefer Windows to MacOS, it's really incredible just how Apple's lower power chips are years ahead of what Qualcomm and Intel can put out. The 2 year old A12X in web benchmarks completely destroys both the 8cx and what we see here from Lakefield. With the power of Apple' ARM chips, we're going to see laptops with the form factors and battery runtimes of Qualcomm/Lakefield devices but with the performance if i7 U series chips.

[rluker5]

Is there any way the core affinity of cinebench could be set to the Sunny Cove core in task manager before running the single core test?
This could also give us an idea if a single Sunny Cove core power throttles.

Also, if power plans are available on this device, perhaps different power plans affect the propensity of usage of the high power core.

[Mister Ed]

While I much prefer Windows to MacOS, it's really incredible just how Apple's lower power chips are years ahead of what Qualcomm and Intel can put out. The 2 year old A12X in web benchmarks completely destroys both the 8cx and what we see here from Lakefield. With the power of Apple' ARM chips, we're going to see laptops with the form factors and battery runtimes of Qualcomm/Lakefield devices but with the performance if i7 U series chips.

"With the power of Apple ARM chips, we're going to see new Apple mobile products with the form factors and...etc"

There, fixed that for you.

[jetstreamblue]

I just received my Samsung Galaxy Book S today in Canada.  Got all the Windows updates and quickly ran a few benchmarks.  Geekbench 5 is about the same as that shown here, but I got better results with Cinebench R20.

Single Core - 237 pts
Multi Core - 553 pts

[S.Yu]

Let's see who falls for this...