Discussion Intel Meteor, Arrow, Lunar & Panther Lakes Discussion Threads

[Tigerick]

As Hot Chips 34 starting this week, Intel will unveil technical information of upcoming Meteor Lake (MTL) and Arrow Lake (ARL), new generation platform after Raptor Lake. Both MTL and ARL represent new direction which Intel will move to multiple chiplets and combine as one SoC platform.

MTL also represents new compute tile that based on Intel 4 process which is based on EUV lithography, a first from Intel. Intel expects to ship MTL mobile SoC in 2023.

ARL will come after MTL so Intel should be shipping it in 2024, that is what Intel roadmap is telling us. ARL compute tile will be manufactured by Intel 20A process, a first from Intel to use GAA transistors called RibbonFET.





Intel Core Ultra 100 - Meteor Lake



As mentioned by Tomshardware, TSMC will manufacture the I/O, SoC, and GPU tiles. That means Intel will manufacture only the CPU and Foveros tiles. (Notably, Intel calls the I/O tile an 'I/O Expander,' hence the IOE moniker.)

 

[H433x0n]

You should quit tech discussions,...

In the first run it run at 30W and score 1930 pts, in the following runs it run at 24W and score 1800 pts, the same as the 13700H@45W.

Is it possible to be that dense..?...

Why are you so passionate about this particular topic? Intel has a less efficient CPU at the moment.. so what? It’s well known at this point.

I posed a simple hypothetical to another poster and you launched into a multi page tirade that was off topic from my original post.

 

[Geddagod]

It is more efficent the same way as at same throughput a 10C/20T is more efficient than a 8C/20T.

Except a 10 core cpu takes the same area as... 10 cores, and requires 10 ringstops.
A 6+8 requires the same area as 8 P cores, with 8 ringstops.
Totally different scenarios.

A 6 + 8 has 20 threads, so does a 10C/20T, you didnt notice this point..?..

It's almost as if the hybrid CPU has many threads that are dramatically weaker than the 10C/20T ones, despite thread count remaining the same...

Then do you have data which shows that the situation is different at low TDPs? Please show It.

Alder Lake’s Power Efficiency – A Complicated Picture

Reviews across the internet show Alder Lake getting very competitive performance with very high power consumption.

chipsandcheese.com

Esentially this entire article.
Gracemont is pushed way past where it should be. The only reason they are doing that is to ensure that the E-cores are overperforming for their area, so that they could save a couple mm2 in die space, esp for desktop.
Just going to quote the article:

Gracemont too is pushed past its sweet spot. To be blunt, Alder Lake’s E-Cores have no business going above 3.5 GHz. But Intel has decided to make them do exactly that, so they don’t go beyond being a performance per area play. Unsurprisingly, this doesn’t look good when reviewers expect E-Cores to boost energy efficiency:

The E-cores, when being pushed, are esentially mostly there for area efficiency.

There is no written rule that Intel has to keep the same die size.

No, but it is the most logical thing to base it on. Not just on die size, but also stuff like ring bus stops, since that can also quickly become a limiting factor.

If Intel wants to keep the die size then It would be 8P core CPU, If they want to keep the performance then It would be a 10P core CPU, but obviously It would be larger.

Wonder why Intel just didn't create a larger die then. It's almost as if the die has to be economical for it to ever be produced anyway...

What?

You said what's the point of this when Phoenix is not competing with an 8P core CPU, but an 6+8 one. Which I agree, compare 8 core Phoenix, to a 6+8 MTL one. 6+8 CPUs should be compared to 8 core CPUs.

Why are you so passionate about this particular topic? Intel has a less efficient CPU at the moment.. so what? It’s well known at this point.

Makes coping much easier. If a 6+8 CPU has the same perf/watt as an 8 core CPU, it's not because Intel might have reached parity, it's just because the real competitor to a 6+8 CPU is a 10 Core CPU.

 

[SiliconFly]

Also anyone else experiencing slow load times on Anandtech or something? I didn't participate in the recent discussions for a couple days cuz Anandtech was loading pretty slowly for me for some reason...

Me too. I think either the forum module was unstable and had to be fixed (or simply restarted). Or worse, the servers it was hosted on was under DDoS attack. Either way, it's back to normal now.

 

[SiliconFly]

I have been too. I was wondering if my ISP was blocking the website or something.

It's definitely not the ISP or backhaul. I checked it. It's the servers. Now fully functional.

 

[Abwx]

Except a 10 core cpu takes the same area as... 10 cores, and requires 10 ringstops.
A 6+8 requires the same area as 8 P cores, with 8 ringstops.
Totally different scenarios.

The point is not to discuss wich technical choice is the most efficient $ wise in area usage, but to compare the perfs under a consumer point of view scrutinity, we are not bean counters who rely as a (CB score)/(core area) metric.

Intel thought that craming 2 E cores instead of 1 Pcore is area more efficient, provide better manufacturing margins while being on par in perfs with 6 + 4 matching a 8C/16T from AMD.

At the end of the day it doesnt work, neither for the area efficency nor for perfs because of process lag, currently even a 6 + 8 with 20T is well behind a 8C/16T, if anything a 6 + 8 MTL wont close this gap.

 

[Geddagod]

The point is not to discuss wich technical choice is the most efficient $ wise in area usage, but to compare the perfs under a consumer point of view scrutinity, we are not bean counters who rely as a (CB score)/(core area) metric.

In which case comparing 8+16 vs 16 cores is fine, since those are the top skus in desktop.
But by the same metric, one should also compare a 6+8 sku vs a 8 core sku in mobile.
That's the consumer centric view.
In the "design" centric view, it would be a a 6+8 vs 8, or a 8+16 vs 12. Because that's the design decision

ntel thought that craming 2 E cores instead of 1 Pcore is area more efficient,

Intel thought that cramming 4 E-cores instead of 1 P core*
4 E-cores is the same area as 1 P-core. The same ringbus stop number too.

 

[SiliconFly]

...your intel bias, live with it...

...it s just that you are disgruntled by your favourite brand lagging AMD...

In your own words: With your AMD bias, I can tell you whats going to happen to your favourite brand starting December 14th.

Intel, for the last 5 years, have been releasing outdated products built on outdated nodes based on outdated and power-hungry core designs. Something that even I find a bit despicable. But in spite of all these severe shortcomings, they managed 88% desktop & laptop CPU market share consistently leaving AMD in dirt picking up scraps. And the credit goes to Intel marketing. They're one of the best in the industry.

Now Intel has given these sharks with an amazing product they've been waiting for all these years. A hyper power-efficient CPU thats far more advanced and a true engineering marvel compared to AMD's old and outdated chiplet design. Not only they're just going to simply outsell AMD like always, but this time it's going to a massacre and it's going to be very brutal.

AMD's time has come. Live with it. And I think it's time to start praying so that Intel has some decent competition in the future for our own good. 😧

 

[H433x0n]

In your own words: With your AMD bias, I can tell you whats going to happen to your favourite brand starting December 14th.

Intel, for the last 5 years, have been releasing outdated products built on outdated nodes based on outdated and power-hungry core designs. Something that even I find a bit despicable. But in spite of all these severe shortcomings, they managed 88% desktop & laptop CPU market share consistently leaving AMD in dirt picking up scraps. And the credit goes to Intel marketing. They're one of the best in the industry.

Now Intel has given these sharks with an amazing product they've been waiting for all these years. A hyper power-efficient CPU thats far more advanced and a true engineering marvel compared to AMD's old and outdated chiplet design. Not only they're just going to simply outsell AMD like always, but this time it's going to a massacre and it's going to be very brutal.

AMD's time has come. Live with it. And I think it's time to start praying so that Intel has some decent competition in the future for our own good. 😧

Don’t do this type of stuff… brand wars are dumb. It’s cool to prefer a certain brand (I do) but open partisan warfare is more for Wccftech comment sections.

 

[SiliconFly]

Don’t do this type of stuff… brand wars are dumb. It’s cool to prefer a certain brand (I do) but open partisan warfare is more for Wccftech comment sections.

Ok 👍

Looks like XMX confirmed in ARL.

If they add XMX, add more Xe cores and increase clocks a bit, it may come closer to a RTX 3050.

 

[Abwx]

In your own words: With your AMD bias, I can tell you whats going to happen to your favourite brand starting December 14th.

There s no bias when comparing actual TDP numbers, but since AMD make a better showing in this register some people who seems ferociously Intel centric will brand as AMD biaised whoever point those numbers...

Point is that the gap is such that MTL could well be below expectations in respect of the competition, seems that it s even more nerve hitting for some, at this point Intel released no hard numbers, that s even more indicative that it wont be the miracle awaited by the same people...

In which case comparing 8+16 vs 16 cores is fine, since those are the top skus in desktop.
But by the same metric, one should also compare a 6+8 sku vs a 8 core sku in mobile.
That's the consumer centric view.

That s what i did, and it appear that a current 6 + 8 require about 45W to score 1800 pts in CB R15 while a 8C/16T require only 25W, from here we can deduct that MTL will have to provide 70-80% better perf/watt to be on par perf/watt wise at 25W against a 7840U.

 

[TESKATLIPOKA]

Alder Lake’s Power Efficiency – A Complicated Picture

Reviews across the internet show Alder Lake getting very competitive performance with very high power consumption.

chipsandcheese.com

Esentially this entire article.
Gracemont is pushed way past where it should be. The only reason they are doing that is to ensure that the E-cores are overperforming for their area, so that they could save a couple mm2 in die space, esp for desktop.
Just going to quote the article:

The E-cores, when being pushed, are esentially mostly there for area efficiency.

I asked you for data, which shows that 6P+8E is not performing as 10P at lower TDPs.
This doesn't show that.
6P at 1GHz + 8E at 1.9GHz perform and consume the same as 10P at 1.51GHz in libx264.

Or If I set frequency to 1GHz, then 4x Golden coves are only 67.5% faster than 4x Gracemont in libx264.
So once more, saying 6P+8E is comparable to 8 Golden cove cores is not true.

No, but it is the most logical thing to base it on. Not just on die size, but also stuff like ring bus stops, since that can also quickly become a limiting factor.

Wonder why Intel just didn't create a larger die then. It's almost as if the die has to be economical for it to ever be produced anyway...

The logical thing to me would be to look at the performance.
You are looking at die size occupied by cores.
Why It should be more important than performance is something I don't understand.

You said what's the point of this when Phoenix is not competing with an 8P core CPU, but an 6+8 one. Which I agree, compare 8 core Phoenix, to a 6+8 MTL one. 6+8 CPUs should be compared to 8 core CPUs.

My point was that 6P+8E is comparable to 10 Golden Cove cores.
So an 8C Phoenix is fighting against Alder(Raptor) Lake 6P+8E, which is equal to 10 Golden Cove cores.

 

[SiliconFly]

That s what i did, and it appear that a current 6 + 8 require about 45W to score 1800 pts in CB R15 while a 8C/16T require only 25W, from here we can deduct that MTL will have to provide 70-80% better perf/watt to be on par perf/watt wise at 25W against a 7840U.

Thats not the way to measure MTL's power efficiency. It's efficiency is architectural and cannot be computed by how power efficient the CPU tile alone is or how much power it draws under certain workloads. Especially not with benchmarks like CB R15. MTL's power efficiency relies more on real world use cases. Like how a common office worker or a average home user uses the laptop.

Under these circumstances, the CPU tile & GPU tile will be OFF most of the time considering these people rely mostly on browsers, office apps and/or productivity apps mostly. And not benchmarks, gaming, video encoding, compression or other stress tests. MTL's efficiency drops significantly in stress tests or under heavy workloads. It's a known fact considering RWC is a bit power hungry.

MTL is designed to be power efficient for common/average users & under these circumstances, it's expected to almost double battery life for them according to Intel's claims. And it sounds very probable considering MTL runs with the CPU tile & GPU tile off most of the time for these people.

 

[Geddagod]

I asked you to show me data, which shows that 6P+8E is not performing as 10P at lower TDPs.
This doesn't show that.

Oh wait, so you want me to show you something that doesn't exist?
What that article clearly shows is that comparing E-cores when they are pushed to the max to perform similarly as the P-cores at high TDPs is unfair because they are being pushed clearly out of their perf/watt curve in order to maximize their perf/area.

The logical thing to me would be to look at the performance.

Extremely stupid smh (talking about that logic, not u lol)

Why It should be more important than performance the hybrid CPU provides is something I don't understand.

Because the design itself slots in 4 E-cores into one P-core. This isn't just die size, or I would easily be claiming some weird ratio of Zen 4C cores = 1 Zen 4 core (like 2 Zen 4 cores = 3 Zen 4C cores) for mobile skus. The way the 4 E-core clusters are designed is to easily slot into 1 P-core cluster.

My point was that 6P+8E is comparable to 10 Golden Cove cores.

When the boost of the E-cores are far out of their ideal range in order to maximize their perf/area, sure. That's highlighted in the chip and cheese article.

That s what i did,

I'm fine with that. Compare a 6+8 MTL with a 8 core Phoenix CPU, that's what it should be compared against, not a hypothetical 10 core Phoenix laptop CPU.

It's efficiency is architectural and cannot be computed by how power efficient the CPU tile alone is or how much power it draws under certain workloads

Perf/watt under load is an important metric to measure. Also, it's architecture causes it to get worse perf/watt under load vs Phoenix because it's not monolithic.

 

[Markfw]

While I will not argue anything about MTL until we have real benches from multiple sites, I do want to suggest to those saying "Intel will conquer the world on Dec 14th" That one CPU category does not make for the entire industry, including the most profitable sector, servers, not to mention HEDT and desktop. As was said above, why don't we not get crazy, and just wait and tone it down.

 

[Abwx]

Thats not the way to measure MTL's power efficiency. It's efficiency is architectural and cannot be computed by how power efficient the CPU tile alone is or how much power it draws under certain workloads. Especially not with benchmarks like CB R15. MTL's power efficiency relies more on real world use cases. Like how a common office worker or a average home user uses the laptop.

Under these circumstances, the CPU tile & GPU tile will be OFF most of the time considering these people rely mostly on browsers, office apps and/or productivity apps mostly. And not benchmarks, gaming, video encoding, compression or other stress tests. MTL's efficiency drops significantly in stress tests or under heavy workloads. It's a known fact considering RWC is a bit power hungry.

MTL is designed to be power efficient for common/average users & under these circumstances, it's expected to almost double battery life for them according to Intel's claims. And it sounds very probable considering MTL runs with the CPU tile & GPU tile off most of the time for these people.

Dunno what are those "certains workloads" but the behaviour at 25-45W will be reproduced at 10-15W, in low usage environments it take 13W for a 13700H to match a 7.5W 7840U, that s the same 13/7.5 = 1.73x perf/watt improvement required for a 6 + 8 MTL to match the 7840U.

 

[Thunder 57]

In your own words: With your AMD bias, I can tell you whats going to happen to your favourite brand starting December 14th.

Intel, for the last 5 years, have been releasing outdated products built on outdated nodes based on outdated and power-hungry core designs. Something that even I find a bit despicable. But in spite of all these severe shortcomings, they managed 88% desktop & laptop CPU market share consistently leaving AMD in dirt picking up scraps. And the credit goes to Intel marketing. They're one of the best in the industry.

Now Intel has given these sharks with an amazing product they've been waiting for all these years. A hyper power-efficient CPU thats far more advanced and a true engineering marvel compared to AMD's old and outdated chiplet design. Not only they're just going to simply outsell AMD like always, but this time it's going to a massacre and it's going to be very brutal.

AMD's time has come. Live with it. And I think it's time to start praying so that Intel has some decent competition in the future for our own good. 😧

At least he's not calling people bozos.

Also you really should reserve judgement until a prodcut is launched or you may end up looking silly. It's so awesome that Intel has yet to give us any performance numbers? It's also so awesome they aren't bringing it to desktop? I doubt it beats RPL in raw performance. Come on you must be smarter than that.

That said, I do think it will be quite efficient. Far more than RPL. I look forward to leaks and launch.

 

[SiliconFly]

At least he's not calling people bozos.

Also you really should reserve judgement until a prodcut is launched or you may end up looking silly. It's so awesome that Intel has yet to give us any performance numbers? It's also so awesome they aren't bringing it to desktop? I doubt it beats RPL in raw performance. Come on you must be smarter than that.

That said, I do think it will be quite efficient. Far more than RPL. I look forward to leaks and launch.

Sorry about that. But when I keep seeing consistent posts stating AMD and 7840U are superior at everything this and that, I had to make a point to put a stop to that tunnel vision. I do know MTL's performance is not expected to be better than RPL and MTL's power efficiency under full load will be no different than RPL.

But MTL isn't targeted at that. Its expected to have very good power efficiency for average users in average use cases mentioned previously. Squarely focusing on weaknesses like CB R15 is a waste of time imho as it doesn't really matter. Thats was the point i was trying to make. And apologies for being a bit too blunt. 🫢

Dunno what are those "certains workloads" but the behaviour at 25-45W will be reproduced at 10-15W, in low usage environments it take 13W for a 13700H to match a 7.5W 7840U, that s the same 13/7.5 = 1.73x perf/watt improvement required for a 6 + 8 MTL to match the 7840U.

You're still missing the point. Forget 13700H for a while. It's obsolete. Take MTL. It's designed to run with the CPU tile & GPU tile off most of the time. Should be able to comfortably beat competition in power efficiency in average/everyday workloads for average consumers like office workers or common home users. Seeing the demos, it's pretty decent in gaming performance too. Although I do think the drop in power efficiency will be quite significant during gaming as all the tiles will be active.

Look at it this way, for majority of the people, MTL will be using only the SoC tile & the I/O tile most of the time which is expected to significantly increase power efficiency. But once the user starts something heavy, efficiency will drop significantly. Thats why I keep using the term architectural power efficiency cos the CPU tile alone is actually a power hog (cos RWC is RPL+ or 13700H+ if you like).

And since the CPU/GPU tiles are expected to be "off" most of the time under average use conditions, power efficiency benchmarks that stress the CPU/GPU tiles will make no sense in case of MTL. Only performance benchmarks will actually produce meaningful results. And MTL isn't expected to beat previous gen in performance under full load I think. I maybe wrong.

 

[Glo.]

Sorry about that. But when I keep seeing consistent posts stating AMD and 7840U are superior at everything this and that, I had to make a point to put a stop to that tunnel vision. I do know MTL's performance is not expected to be better than RPL and MTL's power efficiency under full load will be no different than RPL.

But MTL isn't targeted at that. Its expected to have very good power efficiency for average users in average use cases mentioned previously. Squarely focusing on weaknesses like CB R15 is a waste of time imho as it doesn't really matter. Thats was the point i was trying to make. And apologies for being a bit too blunt. 🫢

If a certain workload uses 5W of power on RPL, then how much more efficient 5W of power used by MTL will be?

 

[SiliconFly]

If a certain workload uses 5W of power on RPL, then how much more efficient 5W of power used by MTL will be?

Looks more like a trick question. 5W on RPL & 5W on MTL should be the same as 5W = 5W irrespective of the platform or architecture if you ask me. Or maybe I didn't get the question right.

But if you're asking about performance with very light work loads like browsing, RPL like 13700H will be using some of the E cores. 5W to 10 W sounds just about right.

And in MTL for the same work load, the LP E cores in the SoC tile should be able to handle it most of the time bringing the power usage well below 5W as the other tiles will be off mostly. Just a guess.

 

[Glo.]

Looks more like a trick question. 5W on RPL & 5W on MTL should be the same as 5W = 5W irrespective of the platform or architecture if you ask me.

Its not a trick question. 5W of power is 5W of power regardless of the platform. If certain workload: web browsing, watching youtube, editing documents in editors uses 5W of power, regardless of platform it is: AMD, Intel, ARM, it will have equal efficiency.

You claimed that in the workloads that people actually care about MTL will run circles around RPL. Before you claim this: you have to ask a question how much performance you can squeeze out of certain thermal envelope.

If we are talking very low power thermal envelopes - it will be difficult for Intel to achieve that goal. To vastly outperform previous gens of CPUs.

Secondly, Efficiency Will NOT sell this. Ultimate performance is what sells computers in PC world.

 

[Geddagod]

Its not a trick question. 5W of power is 5W of power regardless of the platform. If certain workload: web browsing, watching youtube, editing documents in editors uses 5W of power, regardless of platform it is: AMD, Intel, ARM, it will have equal efficiency.

You claimed that in the workloads that people actually care about MTL will run circles around RPL. Before you claim this: you have to ask a question how much performance you can squeeze out of certain thermal envelope.

If we are talking very low power thermal envelopes - it will be difficult for Intel to achieve that goal. To vastly outperform previous gens of CPUs.

Secondly, Efficiency Will NOT sell this. Ultimate performance is what sells computers in PC world.

Eh. Normally I would agree, esp on desktop. Battery life sells tho, for mobile. Stuff like this:

 

[SiliconFly]

Just an observation. I see many people use the assumption that in Intel CPUs, 1 P core equals 4 E cores. It's not the case. I used a ADL die shot to measure the area. That is, after deducting the L3 cache area for both, a E core complex is roughly 1.43x bigger than a P core. That is, 1 P core is only around 2.8 E cores.

Including the L3 cache will throw the calculations off as L3 cache sizes can vary which isn't an accurate way to measure. But if included, in this case, the E core complex with L3 is roughly 1.3x bigger than a P core with L3. That is, with L3 (for both), 1 P core is around 3 E cores. But I think we shouldn't include L3 as it's not the right way imho.

 

[Geddagod]

Just an observation. I see many people use the assumption that in Intel CPUs, 1 P core equals 4 E cores. It's not the case. I used a ADL die shot to measure the area. That is, after deducting the L3 cache area for both, a E core complex is roughly 1.43x bigger than a P core. That is, 1 P core is only around 2.8 E cores.

Including the L3 cache will throw the calculations off as L3 cache sizes can vary which isn't an accurate way to measure. But if included, in this case, the E core complex with L3 is roughly 1.3x bigger than a P core with L3. That is, with L3 (for both), 1 P core is around 3 E cores. But I think we shouldn't include L3 as it's not the right way imho.

It's not perfect, but it's a bit obvious the way a E-core cluster spots into slightly more area than an P-core that 1:4 is the ratio. You can't do a 1:3 ratio because there are no "3 E-core clusters" shaped that way. Same reason for AMD mobile- sure one Zen 4C core is smaller than 1 Zen 4 core. But you don't see people saying 2 Zen 4 cores = 3 Zen 4C cores, they aren't shaped that way for it to work out like that, in that case it's just a 1:1 ratio (though this changes with Bergamo due to L3 cuts).
Also, with MTL, the E-core cluster ends up being ~10% larger than a P-core.
If you look at this die shot:

The setup of the cluster is pretty interesting. Even with a RPL die shot, there appears to be a good amount of space left where an extra bit of L3 can be extended. If this ratio holds, who knows, maybe that slightly better area utilization might happen with ARL or PTL. Would be pretty interesting.
Also, idk abt ur math, but Locuza found one E-core cluster to be ~25% larger than an E-core cluster, not ~45% for ADL. And I might be capping but I'm pretty sure that ratio shrunk for RPL vs ADL as well (meaning the P : E core area ratio was closer to 1:1).

 

[Hulk]

In your own words: With your AMD bias, I can tell you whats going to happen to your favourite brand starting December 14th.

Intel, for the last 5 years, have been releasing outdated products built on outdated nodes based on outdated and power-hungry core designs. Something that even I find a bit despicable. But in spite of all these severe shortcomings, they managed 88% desktop & laptop CPU market share consistently leaving AMD in dirt picking up scraps. And the credit goes to Intel marketing. They're one of the best in the industry.

Now Intel has given these sharks with an amazing product they've been waiting for all these years. A hyper power-efficient CPU thats far more advanced and a true engineering marvel compared to AMD's old and outdated chiplet design. Not only they're just going to simply outsell AMD like always, but this time it's going to a massacre and it's going to be very brutal.

AMD's time has come. Live with it. And I think it's time to start praying so that Intel has some decent competition in the future for our own good. 😧

One thing here I am noticing is that Intel DOES have pretty great marketing. In fact, sometimes I wonder of the cart drives the horse at times? What I mean is instead of engineering telling marketing, "This is what is going to be great about our next release, start working on the propaganda," it's more like marketing telling engineering, "Look we need one BIG thing to push, it may even be a corner usage case, but as long as it shows well in simple dummy bar graphs and easy slogans we can sell it." Give us crazy long battery life in some usage case!

Engineering says, "Okay, I guess we can figure a way to shut almost everything down and maximize for video playback and so we could probably put some distance on the pack with that metric."

It's like "teaching to the test." If it's a bad test you're teaching to then it's a waste of time. Teaching for a good/useful test can be beneficial.

As always when new releases are coming I'm getting hyped for MTL and Zen 5..

 

[tamz_msc]

You got your data with/without Intel DTT?

Sure, the only way to "disable" DTT(without uninstalling the driver, which is not recommended by Intel) is to use ThrottleStop. But you cannot have ThrottleStop and HWinfo ON at the same time, so you have to apply the settings and close its window. Anyway, here's the results -

ThrottleStop OFF

Ultra Performance (PL2 = 51 W, PL1 = 30 W)



Optimized (PL2 = 51 W, PL1 = 25 W (variable))



Cool (PL2 = 51 W, PL1 = 15 W)



ThrottleStop ON

Ultra Performance



Cool