Discussion Intel current and future Lakes & Rapids thread

[eek2121]

You forgot heat and fan noise , a laptop that you cant use on your lap isn't great .
What im really interested to see with TGL is power usage with low thread count thats the bit where i think TGL will be behind Renior, it will beat it in ST but Renior doesnt even get close to 15W using a single thread, AMD choose to leave performance on the table here.

I should have included thermals, as well, definitely. Would’t want to get 3rd degree burns from a laptop. 😉

 

[eek2121]

4700u turbos to 4.1 and 4800 turbos to 4.2. Not sure how much more single thread performance they can wring out of TSMC 7nm within a 15 or even 28 watt power budget.

Don’t confuse TDP with power. Zen 2 can hit 4.75ghz using 11W, not sure about Renior.

 

[itsmydamnation]

4700u turbos to 4.1 and 4800 turbos to 4.2. Not sure how much more single thread performance they can wring out of TSMC 7nm within a 15 or even 28 watt power budget.

lots, 4.2 on my 4700u in AVX256 (y-cruncher) T1 consumes 11watts Peak SOC total power.

 

[jpiniero]

https://twitter.com/x/status/1301220574463750146

Ian says he was told the shipping Tiger Lake stepping does actually support LPDDR5. So it's not a validation issue, it's an OEM's don't want it issue.

 

[A///]

@jpiniero I know Samsung had working prototypes of LPDDR5 like 2 years ago. Are there early production units available on the market for purchase?

 

[A///]

That's pretty surprising. I was expecting products shipping soon, not 2 months from now. Especially since it seems like they are blowing most of their 10 nm capacity on this.

It should make for good holiday sales. Realistically, they're better off putting their resources into mobile if the margins are great for them. They'll have a good 4Q, too. It's perfect for the holidays period where people will be buying new laptops for themselves, a friend, family member, significant other, etc.

 

[Exist50]

https://twitter.com/x/status/1301220574463750146

Ian says he was told the shipping Tiger Lake stepping does actually support LPDDR5. So it's not a validation issue, it's an OEM's don't want it issue.

I've been hearing the opposite, and the "architectural support" wording seems to suggest it's not currently supported in silicon.

@jpiniero I know Samsung had working prototypes of LPDDR5 like 2 years ago. Are there early production units available on the market for purchase?

Phones already support and ship with it.

 

[A///]

Phones already support and ship with it.

Oh, right. I don't follow phone news at all.

 

[IntelUser2000]

Upon what known iGPU performance metrics are you basing this on?

Also, they list 2.8GHz as the base frequency which would also indicate cTDP of 28W, would it not?

The Acer device uses Intel's Dynaming Tuning, which uses "AI" to steadily reduce performance from higher PL levels rather than drop off the cliff as with predecessors. First introduced with Icelake.

Test Intel Tiger Lake - sprawdzamy wydajność laptopa z Core i7-1165G7 z grafiką Iris Xe

www.komputerswiat.pl

The 28W PL1 configuration reduces to 17W over a ~10 min period. Some vendors treat PL1 values similar to PL2 values.

 

[Hitman928]

The Acer device uses Intel's Dynaming Tuning, which uses "AI" to steadily reduce performance from higher PL levels rather than drop off the cliff as with predecessors. First introduced with Icelake.

Test Intel Tiger Lake - sprawdzamy wydajność laptopa z Core i7-1165G7 z grafiką Iris Xe

www.komputerswiat.pl

The 28W PL1 configuration reduces to 17W over a ~10 min period. Some vendors treat PL1 values similar to PL2 values.

So it is configured to 28W but throttles to 17W over long sustained loads, it just tries to do so in a softer manner than previous models.

 

[IntelUser2000]

So it is configured to 28W but throttles to 17W over long sustained loads, it just tries to do so in a softer manner than previous models.

Yes, that explains the weird behavior, since the reviewer does not understand how it works. So he would keep running the tests one after another and subsequent tests would show lower scores. The other site with the Acer is getting 2100 points, while the site we are discussing is getting 2500.

17W is PL1, Acer just changes the name a bit.

 

[ondma]

Yes, that explains the weird behavior, since the reviewer does not understand how it works. So he would keep running the tests one after another and subsequent tests would show lower scores. The other site with the Acer is getting 2100 points, while the site we are discussing is getting 2500.

17W is PL1, Acer just changes the name a bit.

That gradual power decrease (machine learning) actually seems counterproductive to me, depending on the workload. Seems like you would get the most work done by using up all the thermal budget before throttling down. For sure, if the task could be completed within the 18 seconds, you would be better off with the slide one scenario.

 

[Hitman928]

That gradual power decrease (machine learning) actually seems counterproductive to me, depending on the workload. Seems like you would get the most work done by using up all the thermal budget before throttling down. For sure, if the task could be completed within the 18 seconds, you would be better off with the slide one scenario.

I assume that's why it's using the workload prediction algorithm. If the algorithm determines that the load will finish in 15-20 seconds, then it will still look like slide 1. If the algorithm determines the workload will last 10 minutes, it will look like slide 2. How well this works in real life, I guess we'll see. My gut reaction is that it won't really make a difference or may even have a negative effect on many workloads but that also probably just depends on how much more performance you are getting for that increase in power.

 

[Hitman928]

I guess all of this doesn't matter because according to Intel, "TDP isn't useful"

https://twitter.com/x/status/1301225968372195329

 

[coercitiv]

I guess all of this doesn't matter because according to Intel, "TDP isn't useful"

That I find insulting.

 

[IntelUser2000]

That I find insulting.

They have engineers and people who know the reality, but they also have those that mask it. Looks like lot of their marketing is like the latter.

 

[Nereus77]

I guess all of this doesn't matter because according to Intel, "TDP isn't useful"

https://twitter.com/x/status/1301225968372195329

I find that answer vague and unconvincing.

 

[uzzi38]

Well they're right. TDP isn't useful.

Especially now that Intel's base clocks appear to be rated at 28W and 9W (guesstimate) respectively. The one reason why TDP was useful was to give a baseline for base clocks. Now it can't even do that.

 

[Gideon]

While AMD is somewhat better in this regard, It's really long overdue they created some new standard metric that is actually useful.

For instance replace TDP with 3 different metrics:
Watts for all-core-load @ base clocks
Watts for single-core peak turbo (only really needed in mobile)
Watts for worst case Vector workload

 

[uzzi38]

While AMD is somewhat better in this regard, It's really long overdue they created some new standard metric that is actually useful.

For instance replace TDP with 3 different metrics:
Watts for all-core-load @ base clocks
Watts for single-core peak turbo (only really needed in mobile)
Watts for worst case Vector workload

Honestly I'd even just take the first two. The latter isn't all that useful for mobile at least (as both companies would fly way over sane-looking power values).

 

[IntelUser2000]

While AMD is somewhat better in this regard, It's really long overdue they created some new standard metric that is actually useful.

To be perfectly frank, no we don't.

People here know what the numbers are.

You are buying a laptop yes? Your decisions will be many, and one of them will be the CPU. The difference between 25W and even 45W CPUs aren't huge nowadays with designs overlapping each other.

If you are not buying a laptop and making this argument then you are just talking about it like we do lots here. Whether its Intel vs. AMD or pure curiosity, the result is the same.

So if the price is right for you, and the portability is ok too, then it doesn't matter what the TDP is as long as the system can cool it enough that long term performance is sustained. The final CPU might end up being a 15W one or a 45W one.

And TDP metric will never go away. Enough data will always exist for us enthusiasts/fanboys/speculators/fanfiction people. If they hide that data then Intel/AMD/Nvidia/etc will be obfuscating the data and we should complain.

So what are we arguing about in the first place?

Honestly I'd even just take the first two. The latter isn't all that useful for mobile at least (as both companies would fly way over sane-looking power values).

I also agree with this. The third isn't needed. Yes some will try but know if you are buying an ultrabook then you won't get everything.

Scratch that. I'm not sure why we need the single core watts either. Possibly the only time it matters is when you are running sustained load on battery. Otherwise it'll very likely be a bursty workload and go to sleep using mW. And single core TDP will be less demanding to thermals than multi-thread workload anyway.

 

[Cardyak]

While AMD is somewhat better in this regard, It's really long overdue they created some new standard metric that is actually useful.

For instance replace TDP with 3 different metrics:
Watts for all-core-load @ base clocks
Watts for single-core peak turbo (only really needed in mobile)
Watts for worst case Vector workload

I agree, and as @uzzi38 already stated, the first two are the most important.

Honestly just be transparent with users and inform them of the maximum power draw when

A) Boosting on one core
B) Boosting on all cores

No need for PL1, PL2, and all the marketing conflation.

Just tell me the maximum power that will be drawn in boost scenarios and set that as your TDP.

For years now consumers have been deceived by ranges such as Coffee Lake and Comet Lake, where a CPU is identified as "35W" but that actually only applies to the base clock, and in reality when the CPU is boosting across all cores you end up with something closer to 90W. Just look at the 9900T, it's ostensibly rated as a 35W low power CPU, but when it's boosting across all 8 cores it draws a level of power comparable to a top-end desktop chip. It's ludicrous marketing that borders on outright duplicity.

 

[teejee]

While AMD is somewhat better in this regard, It's really long overdue they created some new standard metric that is actually useful.

For instance replace TDP with 3 different metrics:
Watts for all-core-load @ base clocks
Watts for single-core peak turbo (only really needed in mobile)
Watts for worst case Vector workload

But TPD (Thermal Design Power) is intended to be used to design a proper cooling for a system. What you propose is a definition how to specify actual power draw in three different usage scenarios.
It would be great if this is specified for all CPU's but it shouldn't be called TPD.
It should rather be called something including "Power draw..."

 

[IntelUser2000]

@Cardyak For desktops you make total sense, and I think this is where the same thought comes from.

But on laptops it doesn't really matter. There are good designs and bad ones, even with similar price range, so designs will very much overlap each other. Ultimately they are limited by thermals so they are not allowed to really exceed PL1 TDP like they do on desktops. If they lie or, the CPU is set to have infinite PL1 duration like on desktop chips the laptop will thermal/power throttle and the performance ends up being no better.

\It would be great if this is specified for all CPU's but it shouldn't be called TPD.
It should rather be called something including "Power draw..."

I don't know why Intel said that nonsense because they are well aware themselves that TDP(specifically PL1 long load) is what determines how the cooling should be. It's just marketing being crap again like earlier today with the presentation.

What does it mean by Power Draw for laptops? Battery life? Clearly except under heavy load(which doesn't last more than 2 hours anyway) in battery life the CPU contributes very little because it goes from peak to sleep so often. That's how you get 10+ hours of battery life on modern devices.

 

[uzzi38]

What does it mean by Power Draw for laptops? Battery life? Clearly except under heavy load(which doesn't last more than 2 hours anyway) in battery life the CPU contributes very little because it goes from peak to sleep so often. That's how you get 10+ hours of battery life on modern devices.

The CPU can contribute in how quickly it can ramp up and down alongside how quickly it can complete the short tasks waking the chip up from sleep, but otherwise agreed.