Less impressed with processor performance increases than graphics

[mikk]

Gemini Lake is nothing more then AL with 2MB additional L2 cache.

This is nonsense. You are so wrong, better do some homework before posting this.

https://fuse.wikichip.org/news/679/...ements-setting-the-stage-for-a-32-core-model/
https://www.cnx-software.com/2017/1...oarchitecture-used-in-gemini-lake-processors/

 

[Insert_Nickname]

This is nonsense. You are so wrong, better do some homework before posting this.

OK, so GL is AL plus a faster divider, a 4th execution port and 2MB extra L2. Still has the same 3 issue wide front end. That's not what a "major" improvement means to me. That's a welcome generational improvement. Also, Intel calls the core Goldmont+. They're not even bothering with a new microarchitecture code name.

We can at least agree to disagree here, and before anyone ask, my laptop is AL powered. So I am familiar with the hardware.

 

[mikk]

Agree to disagree? You claimed GL is nothing more than AL with just 2 MB bigger L2 cache which just proves you have no clue. The only thing you should agree is that you were wrong.

 

[IntelUser2000]

OK, so GL is AL plus a faster divider, a 4th execution port and 2MB extra L2. Still has the same 3 issue wide front end. That's not what a "major" improvement means to me. That's a welcome generational improvement. Also, Intel calls the core Goldmont+. They're not even bothering with a new microarchitecture code name.

I would, agree with mikk.

The performance improvements in Geekbench is 25% overall, but if you break down into Integer and FP, you get gains in the range of 30%. That's substantial. Their main core line for a good gain was 15-20%. The gains with Goldmont Plus are similar to what we had with Goldmont over Silvermont.

I think you're thinking about the previous generation Bay Trail/Braswell Atoms.

Nope. Apollo Lake, if you really look into the numbers, are quite a bit of ways away from the Core 2 range. You can see roughly how far Bay Trail is from the Core 2 chips using Anand Bench.

Also: https://www.anandtech.com/show/6936...cture-revealed-getting-serious-about-mobile/6

you should expect a 2.4GHz Silvermont to perform like a 1.2GHz Penryn.

Intel says on their press release site, they expect Pentium Silver notebook chips to be 58% faster than Braswell(Bay Trail cores), and the desktop versions to be 68% faster. Now, the notebook chip comparisons are roughly equal in frequency, while Pentium Silver on desktop is ahead of Braswell desktop in frequency.

Bay Trail/Cherry Trail --> Goldmont = ~25% gain
Goldmont --> Goldmont Plus = ~25% gain

Let's say using Merom/Conroe as a comparison instead of Penryn knocks quoted comparison to 2.3/1.2GHz. If you assume 90% scaling with clocks, you end up with needing 2.3GHz Goldmont Plus for a 2GHz Merom/Conroe. 2.3/1.88GHz if you assume 100% scaling. That roughly 15-20% gap is what Core 2 has against AMD Athlon 64 based cores.

 

[IntelUser2000]

OK, so GL is AL plus a faster divider, a 4th execution port and 2MB extra L2. Still has the same 3 issue wide front end. That's not what a "major" improvement means to me. That's a welcome generational improvement. Also, Intel calls the core Goldmont+. They're not even bothering with a new microarchitecture code name.

Expanding on this point. If you read about architectures, they state expansion of the decoders alone doesn't do much. The engineers add decoders if they see a reason for doing so. That reason usually comes with expanding other parts of the core. Beefier execution units, larger OoOE resources, more cache bandwidth, better branch prediction.

Actually, if you read into details about Goldmont Plus, it extends resources nearly everywhere except adding a decoder.

Pentium II/III is a 3-issue decode architecture. Pentium M, Core Duo, and Athlon 64s are much faster despite using the same 3-issue width.

 

[mikk]

I would like to have that minor 30% IPC improvement for Icelake ....or even better for every minor refresh release.

 

[Phynaz]

Still has the same 3 issue wide front end. That's not what a "major" improvement means to me.

How did you determine the front end needs improvement?

 

[tential]

I would, agree with mikk.

The performance improvements in Geekbench is 25% overall, but if you break down into Integer and FP, you get gains in the range of 30%. That's substantial. Their main core line for a good gain was 15-20%. The gains with Goldmont Plus are similar to what we had with Goldmont over Silvermont.



Nope. Apollo Lake, if you really look into the numbers, are quite a bit of ways away from the Core 2 range. You can see roughly how far Bay Trail is from the Core 2 chips using Anand Bench.

Also: https://www.anandtech.com/show/6936...cture-revealed-getting-serious-about-mobile/6



Intel says on their press release site, they expect Pentium Silver notebook chips to be 58% faster than Braswell(Bay Trail cores), and the desktop versions to be 68% faster. Now, the notebook chip comparisons are roughly equal in frequency, while Pentium Silver on desktop is ahead of Braswell desktop in frequency.

Bay Trail/Cherry Trail --> Goldmont = ~25% gain
Goldmont --> Goldmont Plus = ~25% gain

Let's say using Merom/Conroe as a comparison instead of Penryn knocks quoted comparison to 2.3/1.2GHz. If you assume 90% scaling with clocks, you end up with needing 2.3GHz Goldmont Plus for a 2GHz Merom/Conroe. 2.3/1.88GHz if you assume 100% scaling. That roughly 15-20% gap is what Core 2 has against AMD Athlon 64 based cores.

So Gemini Lake is quite "fast".

Gemini Lake and the successor will be quite interesting products for the tablet/low end range of computing.

According to the Anandtech article GeminiLake turbos anywhere from 2.4 ghz to 2.8 which is faster than the 2.3 ghz you quoted.

I may be picking up a desktop and mobile vesion of Gemini Lake if it can match my penryn system.

 

[whm1974]

So Gemini Lake is quite "fast".

Gemini Lake and the successor will be quite interesting products for the tablet/low end range of computing.

According to the Anandtech article GeminiLake turbos anywhere from 2.4 ghz to 2.8 which is faster than the 2.3 ghz you quoted.

I may be picking up a desktop and mobile vesion of Gemini Lake if it can match my penryn system.

So how is Linux support with Gemini Lake?

 

[Arachnotronic]

OK, so GL is AL plus a faster divider, a 4th execution port and 2MB extra L2. Still has the same 3 issue wide front end. That's not what a "major" improvement means to me. That's a welcome generational improvement. Also, Intel calls the core Goldmont+. They're not even bothering with a new microarchitecture code name.

We can at least agree to disagree here, and before anyone ask, my laptop is AL powered. So I am familiar with the hardware.

Far, far more than that.

This is a "major" generational change.

 

[Arachnotronic]

Gemini Lake die shot:

 

[Dayman1225]

Gemini Lake die shot:

Source? coz it looks nothing like Apollo Lake from wikichip

https://en.wikichip.org/wiki/intel/microarchitectures/goldmont

 

[TheELF]

Source? coz it looks nothing like Apollo Lake from wikichip

https://en.wikichip.org/wiki/intel/microarchitectures/goldmont

Well the bottom half does look the same...
(upper part is iGPU maybe? )

 

[Charlie22911]

That actually looks like the same photo, but cropped.

 

[IntelUser2000]

That actually looks like the same photo, but cropped.

I agree, the blocks are identical. Computerbase got the wrong picture.