Discussion Intel current and future Lakes & Rapids thread

Page 123 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Yotsugi

Golden Member
Oct 16, 2017
1,029
487
106
On my SkyLake U 15W laptop I do run compilation and encoding tasks (I do limit CPU frequency though because my laptop's cooling is far from great and I'm afraid such tasks will kill it). Still don't know what you're talking about.
You're a weird one then.
15W laptops are for all intents and purposes not made for heavy-ish workloads.
 

birdie

Member
Jan 12, 2019
98
71
51
You're a weird one then.
15W laptops are for all intents and purposes not made for heavy-ish workloads.

I guess I have an entire company of weird people then 'cause most of us have mobile Intel U CPUs and almost everyone runs compilation and virtualization tasks. Some do rendering and video encoding.
 

BigDaveX

Senior member
Jun 12, 2014
440
216
116
I have an i5 8250U, another 15W chip, and it does very well when it comes to encoding. Granted, QuickSync probably helps, but it seems to manage pretty respectable clockspeeds under heavy load. 15W chips have come quite a way in the last few years.
 
  • Like
Reactions: birdie

birdie

Member
Jan 12, 2019
98
71
51
Ice Lake features a 18% IPC increase (!) vs. Sky Lake (which AMD has finally matched only with Zen 2.0) and people are flaying Intel like there's no tomorrow. What the hell?

Yes, they are very late with the 10nm node. Yes, it still doesn't work as Intel intended but they are on the right path and they are co-developing the 7nm node, so the future looks bright.

Why all the hate and criticism? I'm dumbfounded.
 
  • Like
Reactions: Zucker2k

ajc9988

Senior member
Apr 1, 2015
278
171
116
...................................................................................................................................................


Nothing exceptional, perf/watt wise it score 590pts in CB R15 at 15W while a Raven Ridge is at 490 and Picasso at 540 pts or so, that s what you call quite faster..?

GPU wise it s weaker than even Raven Ridge, it s only due to 3733MHz RAM (more than in DTs...) that it can compete with 2400 MHz RAM equipped RR, not sure that it will be competitive with Picasso + 2933 RAM.

On the IPC front Intel s 18% is BS, from PCWorld review we can deduct that in Geekbench 5% of the better "IPC" AVERAGE is brought by AES and AVX2.

In CB R15 ST it is 3% faster clock/clock than SKL and roughly 4% slower than Zen 2, i guess that to get their 18% they also included Quiksync as another IPC provider...
Were those run at the same frequency? It was my understanding they were not, meaning you were not comparing IPC. To compare that, you need ram at same speed and timings, you need CPU core frequency the same, try to control for other factors as much as possible, etc.

They didn't test IPC, they tested out of box performance, which is very different.
 
  • Like
Reactions: Zucker2k

Yotsugi

Golden Member
Oct 16, 2017
1,029
487
106
Ice Lake features a 18% IPC increase (!) vs. Sky Lake (which AMD has finally matched only with Zen 2.0) and people are flaying Intel like there's no tomorrow. What the hell?
IPC in vacuum matters nothing if you can't clock the chip anything reasonable.
You can make the widest core ever, but it's gonna clock ~200MHz and do nothing.
 

Bouowmx

Golden Member
Nov 13, 2016
1,138
550
146
Not hating on architecture, Sunny cove, but the implementation is bad (3.3/3.5 GHz all-core on i5, i7)
 

ajc9988

Senior member
Apr 1, 2015
278
171
116
Ice Lake features a 18% IPC increase (!) vs. Sky Lake (which AMD has finally matched only with Zen 2.0) and people are flaying Intel like there's no tomorrow. What the hell?

Yes, they are very late with the 10nm node. Yes, it still doesn't work as Intel intended but they are on the right path and they are co-developing the 7nm node, so the future looks bright.

Why all the hate and criticism? I'm dumbfounded.
Yes, let's ignore it being 5 years later, all iterations of skylake with little to no IPC change, and let's ignore the IPC of cannon lake, which is a token CPU and a black mark on Intel.

Let's also ignore the pricing differential and the years of gouging and incrementation in development.

Let's forget the marketing lies, etc.

Now, I do have criticism for AMD as well, but except for the APU comparison, I've critiqued it versus its own products.
 

ajc9988

Senior member
Apr 1, 2015
278
171
116
Not hating on architecture, Sunny cove, but the implementation is bad (3.3/3.5 GHz all-core on i5, i7)
You are taking issue with process, not uarch.

I take issue with marketing and incrementation and bleeding consumers
IPC in vacuum matters nothing if you can't clock the chip anything reasonable.
You can make the widest core ever, but it's gonna clock ~200MHz and do nothing.
I disagree. Double the IPC and cut the frequency to 40% and you still get more performance. It is a balancing act, and slower doesn't mean worse.

All processes are going to slow down due to heat density. Get used to that idea. It is physics.
 
Last edited:
  • Like
Reactions: birdie

Abwx

Lifer
Apr 2, 2011
10,847
3,297
136
Were those run at the same frequency? It was my understanding they were not, meaning you were not comparing IPC. To compare that, you need ram at same speed and timings, you need CPU core frequency the same, try to control for other factors as much as possible, etc.

They didn't test IPC, they tested out of box performance, which is very different.

Those are scores at 15W TDP, so everything is included, IPC and intrinsical efficency of the uarch at a given node...
 

Hitman928

Diamond Member
Apr 15, 2012
5,179
7,630
136
I guess I have an entire company of weird people then 'cause most of us have mobile Intel U CPUs and almost everyone runs compilation and virtualization tasks. Some do rendering and video encoding.

Sounds like your company's productivity is being severely limited by choice of hardware.
 
  • Like
Reactions: beginner99

ajc9988

Senior member
Apr 1, 2015
278
171
116
Those are scores at 15W TDP, so everything is included, IPC and intrinsical efficency of the uarch at a given node...

No, you don't understand IPC then. IPC is instructions per cycle. The cycle is predicated by the clock and frequency of the CPU. So you have to run them at the same frequency interval so that the same number of cycles occur in a given period to then compare the number of instructions it can complete per cycle.

You can try to back recreate this when running at different frequencies, but as some software doesn't scale linearly of give to this type of comparison, it is preferred to use the same clock speed.

Slower actually means worse when you previous stuff runs way faster at reasonable power.
No, it doesn't.

With a new process, you increase transistor count to theoretically increase IPC. This, due to smaller transistors, causes a higher density in heat. If the process cannot overcome that to allow frequency scaling, you are left trying to design higher IPC with the extra area at the smaller node, that way to keep performance moving forward even with slower clocks.

I would take an overall huge IPC to huge frequency so long as frequency was not cut as much as IPC.

People keep thinking that speed will increase forever. It won't.
 

birdie

Member
Jan 12, 2019
98
71
51
Sounds like your company's productivity is being severely limited by choice of hardware.

Or maybe my co-workers are content with lightweight laptops which don't feature loud blower fans. I really don't understand why you keep doubting and attacking 15W Intel parts as if everyone must run heavy laptops with 45W or higher TDP just because they are available.

Nothing is severely limited.
 

Yotsugi

Golden Member
Oct 16, 2017
1,029
487
106
No, it doesn't.
Yes it does, CML-U clocks are way up versus ICL-U.
With a new process, you increase transistor count to theoretically increase IPC. This, due to smaller transistors, causes a higher density in heat. If the process cannot overcome that to allow frequency scaling, you are left trying to design higher IPC with the extra area at the smaller node, that way to keep performance moving forward even with slower clocks.

I would take an overall huge IPC to huge frequency so long as frequency was not cut as much as IPC.

People keep thinking that speed will increase forever. It won't.
You're trying to make an obtuse explanation for something people already know.
 
  • Like
Reactions: Thunder 57

Abwx

Lifer
Apr 2, 2011
10,847
3,297
136
No, you don't understand IPC then. IPC is instructions per cycle. The cycle is predicated by the clock and frequency of the CPU. So you have to run them at the same frequency interval so that the same number of cycles occur in a given period to then compare the number of instructions it can complete per cycle.

You can try to back recreate this when running at different frequencies, but as some software doesn't scale linearly of give to this type of comparison, it is preferred to use the same clock speed.

You are in a circular reasonning where you think that you can harvest one thing several times..

If it s tested at 15W it means that the IPC advantage is accounted, by running them at same frequency you ll have a chip that consume more than the other, namely the one with higher IPC, you can then downclock the one with higher IPC such that it consume the same and it will still provide better throughput due to frequency being downclocked by a smaller amount than the IPC difference.

Once all is made you can compare the chips at equal power, wich is the only metric that matter, or should we compare them at different powers, that is at, same frequency.?.

Really, i wonder the lengths some people are going through once the numbers do not reflect their hopes..
 
  • Like
Reactions: Thunder 57

Hitman928

Diamond Member
Apr 15, 2012
5,179
7,630
136
Or maybe my co-workers are content with lightweight laptops which don't feature loud blower fans. I really don't understand why you keep doubting and attacking 15W Intel parts as if everyone must run heavy laptops with 45W or higher TDP just because they are available.

Nothing is severely limited.

There is no attacking. I just bought a 15W CPU laptop and am very happy with it, but I wouldn't try to run heavy compute tasks on it for my job either. If you're using a tool that takes you multiple times longer to complete a task than the proper tool, you are severely limiting your productivity. I wouldn't try to run tasks like that for work on any laptop except maybe a few choice DTR type systems.
 
  • Like
Reactions: ajc9988

ajc9988

Senior member
Apr 1, 2015
278
171
116
Yes it does, CML-U clocks are way up versus ICL-U.

You're trying to make an obtuse explanation for something people already know.

You are comparing two different process nodes, one of which is significantly denser than the other. What you may or may not know is Intel's 14nm process for Skylake was MORE dense than the 14nm++ process used by coffee lake and likely comet lake CPUs.

So, although the clocks are higher, it is using an older uarch, it is using a less dense process, etc. That doesn't make one better or worse, rather it makes it different. That is why you must not just look at frequency or IPC, rather holistic performance, although the others are not trivial either.

You are in a circular reasonning where you think that you can harvest one thing several times..

If it s tested at 15W it means that the IPC advantage is accounted, by running them at same frequency you ll have a chip that consume more than the other, namely the one with higher IPC, you can then downclock the one with higher IPC such that it consume the same and it will still provide better throughput due to frequency being downclocked by a smaller amount than the IPC difference.

Once all is made you can compare the chips at equal power, wich is the only metric that matter, or should we compare them at different powers, that is at, same frequency.?.

Really, i wonder the lengths some people are going through once the numbers do not reflect their hopes..
This shows a lack of understanding. TDP is NOT IPC. Limiting the amount of power, which IPC has a specific definition not related to the heat disipation and Intel made the term TDP worthless in regards to their products, that does NOT bake in the IPC advantage, nor does comparing equivalent TDP processors give any notion of the IPC of said processors. You test IPC in a SPECIFIC WAY. That means controlling the frequency. You can try to control memory for bandwidth and latency as well. You cannot control for the cache variance, so that gets baked into the IPC. You can try to control for the instructions sets used.

IPC isn't about power consumption. It never has been. It is what its name suggests: INSTRUCTIONS PER CYCLE. It has nothing to do with TDP or how much energy it takes to do the task.

You are talking about matching power consumption and comparing performance, which is NOT IPC! That is a different metric of performance per watt. Although you may prefer that metric, it is NOT IPC!

Also, did you read my prior posts attacking this chip in other ways?
 
Last edited:
  • Like
Reactions: Zucker2k

ajc9988

Senior member
Apr 1, 2015
278
171
116
There is no attacking. I just bought a 15W CPU laptop and am very happy with it, but I wouldn't try to run heavy compute tasks on it for my job either. If you're using a tool that takes you multiple times longer to complete a task than the proper tool, you are severely limiting your productivity. I wouldn't try to run tasks like that for work on any laptop except maybe a few choice DTR type systems.
I think another way around it is to do the work on low power laptops backed up by powerful servers that do the heavy lifting after you get the work ready. But, that is my opinion.
 
  • Like
Reactions: beginner99

birdie

Member
Jan 12, 2019
98
71
51
There is no attacking. I just bought a 15W CPU laptop and am very happy with it, but I wouldn't try to run heavy compute tasks on it for my job either. If you're using a tool that takes you multiple times longer to complete a task than the proper tool, you are severely limiting your productivity. I wouldn't try to run tasks like that for work on any laptop except maybe a few choice DTR type systems.

If the tasks we run can complete on our laptops in a moderate amount of time, then there's no real reason to use "full-featured" CPUs. Maybe you should stop thinking for us - we employ quite smart people who understand the choices they make and buy the systems they find appropriate for their work.
 
  • Like
Reactions: Drazick and ajc9988

BigDaveX

Senior member
Jun 12, 2014
440
216
116
My main home rig is still rocking an i7 5820K, which will still knock the socks off my laptop in any reasonably intensive task despite its age.

However, my laptop's i5 8250U does make for a very capable substitute in situations where I need to quickly edit and encode a video there and then. And to be honest, I actually hadn't bought it with the intention of doing that - I thought the CPU was 4C/4T when I bought it, but when I found out it was actually 4C/8T, I tried out some heavier stuff and was surprised just how well it did.

If anything, the main limiting factor with the laptop wasn't the CPU, but rather the godawful hard drive it came with, which I swapped out for an SSD pretty quickly. Well, that and the iGPU is probably completely hopeless when it comes to any game made after the Bush administration, but hey, that's what my Switch is for, right?
 

Hitman928

Diamond Member
Apr 15, 2012
5,179
7,630
136
If the tasks we run can complete on our laptops in a moderate amount of time, then there's no real reason to use "full-featured" CPUs. Maybe you should stop thinking for us - we employ quite smart people who understand the choices they make and buy the systems they find appropriate for their work.

I've seen a lot of purchasing managers make a lot of awful decisions. I'm not saying you can't run heavy compute tasks on a low powered laptop, just that the productivity will be severely limited. There's no way around that. Either what is being run isn't really compute heavy loads or you could get the work done a lot faster and complete many more tasks in the same amount of time using a more powerful system. If there's some way around this, please enlighten me.
 

ajc9988

Senior member
Apr 1, 2015
278
171
116
If the tasks we run can complete on our laptops in a moderate amount of time, then there's no real reason to use "full-featured" CPUs. Maybe you should stop thinking for us - we employ quite smart people who understand the choices they make and buy the systems they find appropriate for their work.
That is also a fair point. Sometimes sending the prep'ed packages to the server, then pulling them back from the server, unless doing like a virtual remote desktop type situation, will increase time, depending on the job.

There are also budget considerations as well, and the human factor on what the employees are willing to use and will enjoy using, etc.

It varies by company and I will not say either choice is wrong UNTIL I see the workloads for myself, which then I could make a better determination on what would be the best under the circumstances.

Also, I don't know your data package for internet, your internal systems, costs of building out, power consumption and space limitations, etc. A lot goes into the calculation on what is best use for the projects being done.

Although, I do agree with Hitman928 that purchasing managers can make some HORRIBLE decisions sometimes.
 

Abwx

Lifer
Apr 2, 2011
10,847
3,297
136
You are talking about matching power consumption and comparing performance, which is NOT IPC! That is a different metric of performance per watt. Although you may prefer that metric, it is NOT IPC!

Also, did you read my prior posts attacking this chip in other ways?

Lack of understanding you said...?..

What is the purpose of comparing at same frequencies other than to check the arch logical efficency without looking at the electrical efficency..?.

Because there s no use of a CPU hat would have 10% higher IPC at the expense of 20% more power.
You think that say 10% more throughput/Hz is free of power charge..?

So testing at the same power will give an indication of IPC and intrinsical efficency of the uarch, ultimately the only thing that mater is the perf/watt, exactly what i m stating, and in this respect one has to admit that 10% better Cinebench score at same 15W power than a 12nm based product is nothing to bragg about, looking at Intel vs Intel 10nm ICL has 20% better perf (and perf/watt) at isopower than the most refined 14nm SKL parts.