dragon range AMD Ryzen 9 7945HX mobile CPU beats Intel Core i9-13980HX mobile CPU in multi-core despite much lower power consumption

[BiG StroOnZ]

Summary

Notebookcheck has the very first review of AMD Zen4 laptop CPU from 7045HX range. The new Ryzen 9 7945HX shows impressive multi-core performance in initial benchmarks. Intel's Raptor Lake Core i9-13980HX is beaten, even though the Ryzen 9 7945HX consumes much less power.  

 

 

 

 

 

Quotes

Quote

The multi-core performance is really impressive and the Ryzen 9 7945HX is sitting at the top of our comparison charts, even ahead of the latest Raptor Lake HX CPUs from intel. The advantage over the old Ryzen 9 6900HX in the previous Zephyrus Duo 16 (at 80W though) is also impressive.

 

The performance of the new Zen 4 processor is even more impressive when you look at the power consumption. All the Intel HX rivals usually consume much more than 150 Watts, but the Ryzen 9 7945HX tops out at 120 Watts. The two following pictures show the CPU data of the Ryzen 9 7945HX as well as the Core i9-13980HX in a single Cinebench R23 Multi run. We also limited the Core i9-13980HX to 120 Watts and got 28333 point sin CB R23 Multi compared to 34521 points for the Ryzen 9 7945HX. This means the new Zen 4 chip is roughly 22 % faster (in multi-core tests) at the same power consumption.

 

In addition to the default power levels, we also checked the performance of the new Ryzen 9 7945HX at lower TDP values and the results are extremely good. It is very obvious that Intel's best Raptor Lake mobile processor does not stand a chance against the new Zen 4 CPU at similar TDP levels. The sweet spot for the Ryzen 9 7945HX is definitely in the range between 80-100 Watts where we can only see a minor performance deficit.

 

The benchmark results above obviously do not give us a full picture yet, but the initial results are extremely positive. AMD managed to close the gap in terms of single-core performance with lower power consumption. The new Zen 4 chip is even faster in multi-core benchmarks and still consumes considerably less power at just 120 Watts, which means laptop manufacturer get more headroom for the cooling of the GPU, for example

 

My thoughts

This is a great showing by AMD and will help them in their attempts to win back some mobile market share over Intel. I think something to keep in mind about these multi-core results, is that Intel still takes the crown in single-threaded performance. Another caveat is gaming performance, where in nearly all games tested the Intel models are ahead in FPS. Regardless, the performance per watt of the 7945HX is highly impressive, and if this is a preview of Dragon Range; it seriously looks like AMD is headed in the right direction.

 

Sources

https://www.notebookcheck.net/AMD-strikes-back-Ryzen-9-7945HX-beats-Intel-Core-i9-13980HX-despite-much-lower-power-consumption.698349.0.html

https://videocardz.com/newz/amd-ryzen-9-7945hx-dragon-range-processor-has-better-multi-core-performance-at-less-power-than-intel-core-i9-13980hx

[TempestCatto]

Hopefully this pushes Intel to start making some advancements towards more power efficient chips. Having the fastest of something is great, but if it's too power hungry and hot, what's the point? Especially if being used in a more professional environment where that stuff actually matters more.

[starsmine]

1 minute ago, TempestCatto said:

Hopefully this pushes Intel to start making some advancements towards more power efficient chips. Having the fastest of something is great, but if it's too power hungry and hot, what's the point? Especially if being used in a more professional environment where that stuff actually matters more.

They have been. flagship they dont care that much about efficiency. 
The e cores are the most efficient x86 cores on the market, Raptor lake just used the same ones as Alder lake, Meteor and Arrow will have a significant boost there. 
The smaller Raptor lake dies will probably beat the cut down dragon range in efficiency by quite a bit. 

However, Dragon range isnt even peak efficiency for AMD this gen, Pheonix being monolithic require less joules to go between io and CCD/CCD and being on 4nm is going to be amazing for ultrabooks. 

[porina]

Dragon Range has a full node advantage over Raptor Lake so you'd hope it would do better.

 

Note these CPUs are going into high performance laptops, which will need to be plugged in when used at max power for any amount of time. And if you're plugged in, the power usage isn't as important as a more mobile type laptop that will be used unplugged more often than not. At worst it might mean a more heavy cooling system is needed and/or more cooling noise.

[starsmine]

Just now, porina said:

Dragon Range has a full node advantage over Raptor Lake so you'd hope it would do better.

Intel is also monolithic 
Chiplets significantly increase(like a magnitude) how many joules it takes to move data around from CCD to CCD/IO. 

so yes, intel 7 ultra (7+ if you will(or 10nm+++ (ice on 10+, alder on 10++))) is a node behind TSMC 5nm, but AMD also has to move data around to a 7nm IO die and the other CCD.

Quote

the curve itself has been improved, shifting prior-generation frequencies by around 200 MHz at ISO-voltage, or alternatively, reducing the voltage by over 50 mV at ISO-frequency.

vs normal intel 7. 

[porina]

27 minutes ago, starsmine said:

Intel is also monolithic 
Chiplets significantly increase(like a magnitude) how many joules it takes to move data around from CCD to CCD/IO. 

It is one of many design decisions they make. Gain a bit here, lose a bit there. I recall that with Zen 4 they changed the IF link (vs previous) to be half as wide but twice as fast, for same nominal bandwidth but using less power somehow. Intuitively dynamic losses should be higher with more clock, so maybe static losses are bigger than I think? Still, it would be interesting if there are any numbers put directly to IF.

[starsmine]

12 minutes ago, porina said:

It is one of many design decisions they make. Gain a bit here, lose a bit there. I recall that with Zen 4 they changed the IF link (vs previous) to be half as wide but twice as fast, for same nominal bandwidth but using less power somehow. Intuitively dynamic losses should be higher with more clock, so maybe static losses are bigger than I think? Still, it would be interesting if there are any numbers put directly to IF.

I know Dr. Ian cutress has published them at one point(for zen 3 I think, he asked about zen 4 and didnt get an answer IIRC), but I could not find them. 

[Arika]

why do we care about the efficiency at the highest tier of things? why can't people focus on the mid range? you know, the products people are more likely to buy

[Chiyawa]

O ho ho. Well, I mean I'm not surprise. Ryzen 7000 series desktops processors already show how power efficient they can be while outputting so much performance. So no surprise their mobile counterparts can pull it off.

 

But yeah, Intel should improve. I'm not sure if Performance Core and Efficient Core helps. For me, I rather have 8 Efficient Cores than having... 8 Efficient Cores and 2 Performance Core. Just make the Efficient Core run at 2.5GHz should be adequate.

[Brooksie359]

6 hours ago, porina said:

Dragon Range has a full node advantage over Raptor Lake so you'd hope it would do better.

 

Note these CPUs are going into high performance laptops, which will need to be plugged in when used at max power for any amount of time. And if you're plugged in, the power usage isn't as important as a more mobile type laptop that will be used unplugged more often than not. At worst it might mean a more heavy cooling system is needed and/or more cooling noise.

I would have to disagree. Even if the laptop has to be plugged into wall you would still want better energy efficiency for the sake of getting away with lighter cooling solutions and lighter power brick possibly. If the amd part can allow laptop manufacturers to create lighter laptops then It would be huge as that is honestly one of the main drawbacks of these top of the line laptops. 

[porina]

2 hours ago, Brooksie359 said:

I would have to disagree. Even if the laptop has to be plugged into wall you would still want better energy efficiency for the sake of getting away with lighter cooling solutions and lighter power brick possibly. If the amd part can allow laptop manufacturers to create lighter laptops then It would be huge as that is honestly one of the main drawbacks of these top of the line laptops. 

We'll have to agree to disagree. I'm not in the market for one of these monsters but I'd make similar arguments for gaming laptops. Making it thinner is not a major benefit. I'd rather have a better cooling solution as it would make more impact for actual use. Cheaper is always good though, both up front and running cost. More options doesn't hurt so people can find what suits them.

[Brooksie359]

2 hours ago, porina said:

We'll have to agree to disagree. I'm not in the market for one of these monsters but I'd make similar arguments for gaming laptops. Making it thinner is not a major benefit. I'd rather have a better cooling solution as it would make more impact for actual use. Cheaper is always good though, both up front and running cost. More options doesn't hurt so people can find what suits them.

I am not saying they would be putting in an insufficient cooling solution. I am saying that because it uses less power it produces less heat and can be cooled with smaller cooling systems making the laptop smaller. I have had many gaming laptops and you would be surprised how much a few extra pounds can really make or break how easy it is to bring around. If you can get top of the line performance in a smaller package why wouldn't you?

[porina]

24 minutes ago, Brooksie359 said:

I am not saying they would be putting in an insufficient cooling solution. I am saying that because it uses less power it produces less heat and can be cooled with smaller cooling systems making the laptop smaller. I have had many gaming laptops and you would be surprised how much a few extra pounds can really make or break how easy it is to bring around. If you can get top of the line performance in a smaller package why wouldn't you?

I have what I'd consider an upper mid range gaming laptop in an unremarkable 15" form factor. It is ball park a few kg in weight with power brick, or about 6 pounds. You're not going much lower than that without making big sacrifices. It has what I estimate to be perhaps 180W cooling solution given that's how much CPU + GPU combined can sustain on it in gaming.

 

Using the test data linked in OP as reference:

• If we take the scenario you're only doing CPU intensive workloads, the AMD offering would be unlimited by that cooling. The Intel offering may be slightly limiting. Peaks can exceed 200W but this may be allowed for short term. If placed in a bigger form factor laptop e.g. 17", it probably wont limit. In other words, if you're running pure CPU workloads, cooling these CPUs aren't a big deal.
• If we take a mixed CPU and GPU scenario, the gaming results show the Intel CPU is on top. We made need more data points on that since other differences in the laptop may contribute to this. It would be nice to see other mixed workloads too but the GPU contributes more.

For sure, AMD having a power efficiency is an advantage in some scenarios. I just don't feel for this application it is that big.

[leadeater]

7 minutes ago, porina said:

• If we take a mixed CPU and GPU scenario, the gaming results show the Intel CPU is on top. We made need more data points on that since other differences in the laptop may contribute to this. It would be nice to see other mixed workloads too but the GPU contributes more.

For sure, AMD having a power efficiency is an advantage in some scenarios. I just don't feel for this application it is that big.

Would also be helpful to track Wh used while running an in-game benchmark and comparing. Intel CPUs may be cable of having very large peak power but games aren't going to fully trigger that sort of power usage. The mid range power usage between an equivalent Intel and AMD laptop CPU could be a lot closer than in all core maximum load.

[mr moose]

9 hours ago, Arika S said:

why do we care about the efficiency at the highest tier of things? why can't people focus on the mid range? you know, the products people are more likely to buy

IKR, you buy a performance machine for maxing out work/games/whatever, who cares if it uses 200W more.  That's like two old school light bulbs.   

[Brooksie359]

6 hours ago, porina said:

I have what I'd consider an upper mid range gaming laptop in an unremarkable 15" form factor. It is ball park a few kg in weight with power brick, or about 6 pounds. You're not going much lower than that without making big sacrifices. It has what I estimate to be perhaps 180W cooling solution given that's how much CPU + GPU combined can sustain on it in gaming.

 

Using the test data linked in OP as reference:

• If we take the scenario you're only doing CPU intensive workloads, the AMD offering would be unlimited by that cooling. The Intel offering may be slightly limiting. Peaks can exceed 200W but this may be allowed for short term. If placed in a bigger form factor laptop e.g. 17", it probably wont limit. In other words, if you're running pure CPU workloads, cooling these CPUs aren't a big deal.
• If we take a mixed CPU and GPU scenario, the gaming results show the Intel CPU is on top. We made need more data points on that since other differences in the laptop may contribute to this. It would be nice to see other mixed workloads too but the GPU contributes more.

For sure, AMD having a power efficiency is an advantage in some scenarios. I just don't feel for this application it is that big.

I mean I guess I would just have to disagree. Let me give you a scenario. You are someone who travels alot and need a laptop that can game and do productivity work on the go. Are you going to choose the much heavier msi titan laptop just get a couple more frames in gaming? Ir would you opt for something like the asus dual 16 that was used in this review that is significantly lighter? 

[porina]

1 hour ago, Brooksie359 said:

I mean I guess I would just have to disagree. Let me give you a scenario. You are someone who travels alot and need a laptop that can game and do productivity work on the go. Are you going to choose the much heavier msi titan laptop just get a couple more frames in gaming? Ir would you opt for something like the asus dual 16 that was used in this review that is significantly lighter? 

Just for context, I looked up the weights of the two models:

 

Titan GT77HX 13VI - 3.3kg https://www.msi.com/Laptop/Titan-GT77-HX-13VX/Specification

ROG Zephyrus Duo 16 (2023) GX650PY - 2.67kg https://rog.asus.com/laptops/rog-zephyrus/rog-zephyrus-duo-16-2023-series/spec/

 

But you're comparing a 16" and a 17" laptop. Let's level the playing field a bit and find a 16" 13980HX laptop. Oh, Asus do one too.

ROG Strix SCAR 16 (2023) G634 - 2.50kg https://rog.asus.com/laptops/rog-strix/rog-strix-scar-16-2023-series/spec/

 

I haven't gone through the whole spec list for what other differences there may be between them, nor do I know their absolute performance. Suffice to say, anyone buying a laptop should get what suits them. Some might prefer a bigger screen from a 17" and that extra weight is worth it for them. The AMD CPU may have an advantage, or it might be irrelevant with all the other factors that go into a laptop. This is not a slam dunk for AMD.

[HenrySalayne]

14 hours ago, porina said:

The AMD CPU may have an advantage, or it might be irrelevant with all the other factors that go into a laptop. This is not a slam dunk for AMD.

It is. AMD's Zen4 processors are scaling much less with power than Intel's Alder Lake. Which results in less overclocking potential on AMD's side but it also helps to keep the clocks up with lower voltages. So what we see in laptops is pretty much this effect.

Der8auer showed a good graph about this some time ago, but I cannot find it.

 

https://www.anandtech.com/show/17641/lighter-touch-cpu-power-scaling-13900k-7950x/4

 

 

[leadeater]

48 minutes ago, HenrySalayne said:

It is. AMD's Zen4 processors are scaling much less with power than Intel's Alder Lake. Which results in less overclocking potential on AMD's side but it also helps to keep the clocks up with lower voltages. So what we see in laptops is pretty much this effect.

Der8auer showed a good graph about this some time ago, but I cannot find it.

 

https://www.anandtech.com/show/17641/lighter-touch-cpu-power-scaling-13900k-7950x/4

Problem is that's still only testing different maximum power profiles against a full all core 100% load, that's actually rare for most people.

 

The 13900K is more power efficient for a single thread 100% load, that still leaves the problem of what is the power usage for say 60% cores utilized at around 20%-40%. Which is more power efficient for that? Because that's way more representative of real world customer usage than CB or Blender is.

 

Spoiler

 

 

TechPowerUp fortunately has some data on that which proves useful towards this question, now keep in mind when reading these for laptop parts with more restrictive power limits the relative difference between them will be lower than a 13900K is with unlimited boost time and as much power limit as basically possible. Desktop K series 12th and 13th Gen parts are the outlier in Intel power efficiency to all other products that Intel offers (non-K desktop, laptop and server).

 

 

 

 

 

 

 

 

So while it is certainly true that all Zen 4 products are in general more power efficient than Intel Alder Lake/Raptor Lake the differences aren't always that pronounced or what the OEM/ODM has done in terms of product power configuration etc is a larger factor. It is often a really bad idea to assume the same CPU in a different laptop will perform and consume the same amount of power, it's more often not the situation than it is, particularly with Intel.

 

[HenrySalayne]

2 hours ago, leadeater said:

The 13900K is more power efficient for a single thread 100% load, that still leaves the problem of what is the power usage for say 60% cores utilized at around 20%-40%.

It depends on the environment. The 13900K is only more efficient because it is scaling much better with voltage. The 7950X really needs to go full-bore to get the SC clocks up to a comparable level. If we drop the clock speeds and voltages on both CPUs, the 7950X will pretty soon overtake the 13900K.

If we talk about laptops, limited cooling solutions and limited power envelopes, AMD's has the edge in SC and MC performance with their Zen4 CPUs.

2 hours ago, leadeater said:

So while it is certainly true that all Zen 4 products are in general more power efficient than Intel Alder Lake/Raptor Lake the differences aren't always that pronounced or what the OEM/ODM has done in terms of product power configuration etc is a larger factor.

I honestly wouldn't call Zen4 more power efficient across the board.  Zen4 will keep the crown up to the medium-high power levels, but eventually, Intel will pull ahead. 6+ GHz are necessary, Intel will be more efficient.

[leadeater]

1 hour ago, HenrySalayne said:

I honestly wouldn't call Zen4 more power efficient across the board.  Zen4 will keep the crown up to the medium-high power levels, but eventually, Intel will pull ahead. 6+ GHz are necessary, Intel will be more efficient.

It is across the board, let side line impossible clocks and configurations. Even then Intel 7 will still exhibit exponential power usage with clocks just like everything else. So if there ever is a cross over point where this could be true it's above the possibilities that Zen 4 with TSMC 5nm can do.

 

Even in these more worst case vs worst case situations (desktop K/X) were seeing around the 30% mark difference in power usage. Once those P cores down clock after maximum boost duration like they do in laptops the difference is going to be less than 30%.

 

I don't think there is any point along any voltage/frequency curve where Intel P-Cores on Intel 7 are more efficient than Zen 4 on TSMC 5nm. The problem is that is not the full picture, you must include total package power and performance.

 

It is perfectly possible that an Intel laptop CPU with 2 P cores and 6/8 E cores can offer the same real world performance at better power efficiency than an AMD 8 Zen 4 core laptop CPU (using IOD and CCD). I don't think this is actually the case but I'm sure it's a lot closer than many realize.

 

And Intel will always have the advantage with SC performance regardless of cooling solution, AMD CPUs outright use more power so require more cooling while offering less performance. Only Zen 5 could change that and it won't live in a vacuum devoid of an Intel 14th Gen market option. Prior comments restricted to IOD + CCD products.

 

I very much like Zen 4, and Zen 3 for that matter, but if you really look at some of the data out there and do good objective analysis AMD isn't that far ahead in terms of power efficiency. If we only look at 13x00K/KS etc then yes that is the case but that's only a very small tiny section of the market.

 

 

1 hour ago, HenrySalayne said:

The 13900K is only more efficient because it is scaling much better with voltage.

Scaling doesn't matter, it's outright in absolute terms using less power than Ryzen 7000/Zen 4 for full package consumption. It's also using less power while offering higher performance. At no point can you get a single Zen 4 core to the same performance of a Raptor Lake P core using less package power, it cannot be done on a chiplet based product currently.

 

If the starting point is lower then lowering the load thus the power usage this will scale down the same style of non-linear curve as would a Zen 4 core. So basically if it's even possible that a Ryzen 7000/ Zen 4 single core at a non-zero percent workload is offering the same equivalent performance at the same or less power the margin of difference is going to be very small that it will not matter enough.

 

Zen 4 cores within the CCD use less power than a Intel P core, but once you add in package power and overhead of things like IOD + IF that's how AMD loses out in single core efficiency then very quickly regain that metric above a single core. This is the trade off currently with the chiplet approach, it scales up extremely well however it does not scale down, like at all.

[HenrySalayne]

On 3/4/2023 at 4:52 PM, leadeater said:

Scaling doesn't matter, it's outright in absolute terms using less power than Ryzen 7000/Zen 4 for full package consumption. It's also using less power while offering higher performance. At no point can you get a single Zen 4 core to the same performance of a Raptor Lake P core using less package power, it cannot be done on a chiplet based product currently.

 

If the starting point is lower then lowering the load thus the power usage this will scale down the same style of non-linear curve as would a Zen 4 core. So basically if it's even possible that a Ryzen 7000/ Zen 4 single core at a non-zero percent workload is offering the same equivalent performance at the same or less power the margin of difference is going to be very small that it will not matter enough.

Across the entire frequency/voltage range? I was under the impression that AMD would eventually pull ahead if you lower the power target, even for SC workloads. But maybe I'm mistaken. Sadly I couldn't find any SC testing at lower power targets.

 

[leadeater]

29 minutes ago, HenrySalayne said:

Across the entire frequency/voltage range? I was under the impression that AMD would eventually pull ahead if you lower the power target, even for SC workloads. But maybe I'm mistaken. Sadly I couldn't find any SC testing at lower power targets.

Depends what you look at, the power usage by the CPU core or the entire CPU for the SC load. AMD is more efficient at the core, just not for the entire CPU.

 

Sadly Anandtech doesn't have the per core power in their review of the Ryzen 7000 like they did for Ryzen 5000. You can look at the older review, the Zen 3 cores used way less power so if you just look at the core only there is a point where the performance would be greater at the same given power between the cores being compared.

 

Ryzen 5000 a single core would use about 20W of power and the IOD would use about 27W of power. The amount of power of the IOD doesn't change a lot based on number of cores utilized. Without the same figures for Ryzen 7000 it's hard to be exact but I know the situation is still the same since total power for a 13900K is lower than for a 7950X for a single core load but higher for two 2 cores and above. The only way that is possible given what we know about the architectures and chiplets is that the IOD is using the same or more power than a single core, this is why loading more cores increases the performance efficiency of Ryzen 7000 and how it can be both more and less efficient than the 13900K based on the number of cores utilized.

 

If it were just about how much power the CPU core used then if one core for a 13900K used for example 30W and one core for Ryzen 7000 was 40W then for two cores Intel would be 60W and AMD 80W, but it's not like that.

 

Take that Zen 4 core and put it in a monlithic package and then the SC total package power will be lower. It's just that underlying compromise of chiplets at play.

[AlTech]

On 3/3/2023 at 2:10 AM, Arika S said:

why do we care about the efficiency at the highest tier of things?

Because AMD would like to put Intel mostly or completely out of business.

 

And some people like high end laptops?

On 3/3/2023 at 2:10 AM, Arika S said:

why can't people focus on the mid range?

Who? Reviewers?

On 3/3/2023 at 2:10 AM, Arika S said:

You know, the products people are more likely to buy

Sure, but if AMD's competing in the same spaces as Intel don't you think they deserve equal press coverage in product segments that almost nobody will buy?

[leadeater]

8 minutes ago, AluminiumTech said:

Because AMD would like to put Intel mostly or completely out of business.

So you think AMD wants to be struck the ire of Anti-Trust laws? You know many business are perfectly happy with making good technology and making sufficient revenue and profits right?