AMD fights back as they showcase Strix point with XDNA2 NPU and Hawk Point while giving Intel the deserved punch in the nose

[filpo]

Summary

AMD has officially introduced its latest Ryzen 8000 APUs codenamed Hawk Point, which refreshes the existing Ryzen 7000 "Phoenix" APUs while simultaneously retaliating to Intel in their table

 

Quotes

Strix Point

Quote

In a recent announcement, AMD introduced the new Hawk Point series under the name Ryzen 8040. This series, an evolution from Phoenix, already features an integrated XDNA core. The Al core within the Ryzen 7040 series provides approximately 10 TOPS of compute performance for generative Al. With the upcoming 8040 series, AMD aims to increase that speed to 16 TOPS.

 

Additionally, AMD has revealed that the next-generation XDNA2 NPU core, part of the Strix Point APU series, is expected to achieve a threefold increase in performance.

 

Quote

AMD has officially confirmed that Strix Point is powered by the Zen5 CPU core architecture and will be equipped with the RDNA3+ graphics, also known as Navi 3.5. Although AMD has yet to delve into details about Navi 3.5, the current emphasis is on Al performance, where the next-generation XDNA2 core is anticipated to deliver a threefold increase in performance. It's important to note that previous speculations hinted at a potential performance range of 45 to 50 TOPS, which aligns with the 3x performance claim and suggests an approximate performance level of 48 TOPS.

 

According to rumours, Strix Point is expected to offer two configurations: one with 12 cores and another with 16 cores, referred to as the Halo configuration. The latter is rumored to be launched in 2025, positioning as a competitor to Apple's premium M- series. Meanwhile, the Strix Point series is slated to succeed Phoenix as the high-end mobile APU series and to be offered alongside Hawk Point next year.

Ryzen Series Rumoured Overview 

Quote

Hawk Point

Quote

The Ryzen 8000 "Hawk Point" APUs are designed for client and consumer PCs, mainly targeting the laptop and notebook segment. Still, there will be additional Ryzen 8000 SKUs for laptops & high-end notebooks in the coming year in the form of Strix Point.

 

The line-up is split into three segments, we first have the high-end Ryzen 8045HS APUs which will be the cream of the crop with the highest clock speeds, we then have the more mainstream Ryzen 8040HS SKUs and finally, we have the Ryzen 8040U which are entry-level SKUs designed for power-optimized platforms.

 

Quote

To sum up the Ryzen 8000 "Hawk Point" family, you are looking at:

• 4nm Zen 4 CPU Architecture
• RDNA 3 GPU Architecture
• Standard & Hybrid SKUs
• Up To 8 Cores / 16 Threads
• Up To Radeon 780M iGPU (12 Compute Units)
• Up To 5.2 GHz Clock Speeds
• Upgraded XDNA NPU
• Up To 16 TOPs For AI Compute
• TDPs Ranging From 15-54W
• Q1 2024 Launch For PCs

 

Quote

 

Quote

AMD Ryzen 8045HS "Hawk Point" APUs - 35-54W SKUs

Quote

First up, we have the Ryzen 8045HS SKUs which include the Ryzen 9 8945HS, Ryzen 7 8845HS, & the Ryzen 5 8645HS. These three chips have the same specifications as their Ryzen 9 7940HS, Ryzen 7 7840HS & Ryzen 5 7640HS SKUs with up to 8/6 cores, Radeon 780M/760M iGPUs, up to 5.2 GHz clock speeds, 16 MB total L3 cache and 35-54W TDPs.

 

AMD Ryzen 8040HS "Hawk Point" APUs - 20-30W SKUs

Quote

Moving down to the AMD Ryzen 8040HS SKUs, we have the Ryzen 7 8840HS and the Ryzen 5 8640HS. These two SKUs are designed as low-power offerings with 20-30W targets. The Ryzen 7 8840HS APU features 8 cores, 16 threads, 16 MB cache, a Radeon 780M iGPU, and up to 5.1 GHz clock speeds while the Ryzen 5 8640HS APU features 6 cores, 12 threads, 16 MB L3 cache, the Radeon 760M iGPU and up to 4.9 GHz clocks. These chips will come with lower base clocks and feature additional thermal/power limits to meet the requirements of power-constrained PCs.

AMD Ryzen 8040U "Hawk Point" APUs - 15-30W SKUs

Quote

Lastly, we have the AMD Ryzen 8040U SKUs which are once again the same configurations as the Ryzen 7040U SKUs. There are four SKUs which include the Ryzen 7 8840U, Ryzen 5 8640U, Ryzen 5 8540U, and the Ryzen 3 8440U. The 8540U and 8440U lack an NPU since they are based on the smaller hybrid die which incorporates Zen 4 and Zen 4c cores. These low-power solutions don't necessitate the inclusion of a dedicated XDNA NPU hence it is not available. These SKUs also come with just Radeon 740M iGPUs which packs 4 compute units.

 

My thoughts

Loving the then offensive but now humorous retaliation to intel (due to their recent marketing against AMD), it might seem childish but it's always nice to see corporations fight back in an amusing way (at least to me)

Back to the chips, they look quite promising with the 5.2GHz boost on the 8945HS with 8 cores, it'll be a great gaming cpu with decent onboard graphics too (better than a desktop 1060 afaik)

 

Sources

AMD announces Ryzen 8045HS, 8040HS and 8040U "Hawk Point" series powered by Zen4, RDNA3 and XDNA - VideoCardz.com

AMD next-gen "Strix Point" series with XDNA2 NPU to offer 3x performance boost for generative AI - VideoCardz.com

AMD Ryzen 8000 Hawk Point APUs Official: Zen 4 CPU, RDNA 3 GPU, & Upgraded XDNA AI NPU With 16 TOPs (wccftech.com)

Edited December 6, 2023 by filpo

[Athan Immortal]

I wonder if Intel got an inside line that this was coming and did a pre-emptive strike. Because honestly it looked so slap dashed together and not considering how many of the actions they're accusing AMD of they themselves have done.

 

With that said, these points are great.

 

• All cores use the same INSTRUCTION SET
• All cores have the same IPC
• All cores can multithread SMT
• Doesn't require hardware OS SCHEDULER
• All cores are EFFICIENT
• All cores can improve GAMING

Every one of those points really highlights how complex Intel's chips have gotten. I kind of liked the idea of efficiency cores, but they really have added such a layer of complexity, where AMD seem to have followed the Keep it Simple philosophy.

[starsmine]

21 minutes ago, Athan Immortal said:

I wonder if Intel got an inside line that this was coming and did a pre-emptive strike. Because honestly it looked so slap dashed together and not considering how many of the actions they're accusing AMD of they themselves have done.

 

With that said, these points are great.

 

• All cores use the same INSTRUCTION SET
• All cores have the same IPC
• All cores can multithread SMT
• Doesn't require hardware OS SCHEDULER
• All cores are EFFICIENT
• All cores can improve GAMING

Every one of those points really highlights how complex Intel's chips have gotten. I kind of liked the idea of efficiency cores, but they really have added such a layer of complexity, where AMD seem to have followed the Keep it Simple philosophy.

Design-wise, Intels e cores are more simple than AMD c cores

The C cores required the design first then min-maxing the design for area which means moving the blocks around in ways you cant do earlier.
Intel e cores were developed in parallel without having to know about other areas of the die. 

Normally when designing a core you are assigned an area, and you need to solve your section and keep it in that area. What happens is that you have some "dead" space if you solve it in less then that area. Another problem is as you squish a core together its different sections become locked together in time (Metronomes on a table) so the clock skews internally can be pulled out of phase as inductive and capacitive loads yoink a signal. Honestly, doing what AMD did is impressive with how fast they did it as well. 

[porina]

An ok evolution but nothing that radical. There's still scope for things to go wrong with scheduling so hope that doesn't come back to bite them.

 

Of the 6 points in their slide, I'd argue 4 of them are not advantages, just differences. The other two, I don't have data to agree or not. We'll have to wait for tests to come in.

[tim0901]

Honestly, this feels just as bad if not worse than Intel's BS to me.

 

As while Intel's marketing is completely ridiculous, their underlying point is correct: AMD's naming scheme is indeed misleading bullshit, even more so than Intel's own.

 

This however just feels... petty? It's not trying to call out Intel for anything, it's just reactionary "E-cores bad" whining.

 

For example, notice how they say "All cores have the same IPC" not "All cores have the same PERFORMANCE". Because Zen4c cores, with their half-sized L3 cache, are expected to perform worse than Zen4 cores. IPC means fuck all in the real world - it's just another number. Real-world performance is what matters.

 

And "Doesn't require hardware OS SCHEDULER" ... is that meant to be a good thing? May I remind everyone of the years of repeated problems of trying to get the Windows scheduler to play nicely with Ryzen? Even now it still isn't great, especially when you're using mixed-CCD CPUs like the 7950X3D. I would be very surprised if the rumoured mixed Zen5/Zen5c chips don't see very similar scheduling issues if they eventually arrive.

 

But then again, why would I expect anything good to come out of AMD's marketing department. They prove themselves as being completely incompetent basically every time AMD try to release a new product...

[filpo]

10 minutes ago, tim0901 said:

And "Doesn't require hardware OS SCHEDULER" ... is that meant to be a good thing? May I remind everyone of the years of repeated problems of trying to get the Windows scheduler to play nicely with Ryzen? Even now it still isn't great, especially when you're using mixed-CCD CPUs like the 7950X3D. I would be very surprised if the rumoured mixed Zen5/Zen5c cores don't see very similar scheduling issues if they eventually arrive.

I think the slide comparing against Intel is supposed to be comprehensible (somewhat) to newbies (same with the intel marketing of 2 days ago

For example, it says that 'all cores are efficient' and 'all cores improve gaming' which is quite vague and doesn't explain why. But if I were someone who just came to the pc world and didn't know anything yet, this marketing would draw me more to AMD 

[starsmine]

31 minutes ago, tim0901 said:

Honestly, this feels just as bad if not worse than Intel's BS to me.

 

As while Intel's marketing is completely ridiculous, their underlying point is correct: AMD's naming scheme is indeed misleading bullshit, even more so than Intel's own.

 

This however just feels... petty? It's not trying to call out Intel for anything, it's just reactionary "E-cores bad" whining.

 

For example, notice how they say "All cores have the same IPC" not "All cores have the same PERFORMANCE". Because Zen4c cores, with their half-sized L3 cache, are expected to perform worse than Zen4 cores. IPC means fuck all in the real world - it's just another number. Real-world performance is what matters.

 

And "Doesn't require hardware OS SCHEDULER" ... is that meant to be a good thing? May I remind everyone of the years of repeated problems of trying to get the Windows scheduler to play nicely with Ryzen? Even now it still isn't great, especially when you're using mixed-CCD CPUs like the 7950X3D. I would be very surprised if the rumoured mixed Zen5/Zen5c chips don't see very similar scheduling issues if they eventually arrive.

 

But then again, why would I expect anything good to come out of AMD's marketing department. They prove themselves as being completely incompetent basically every time AMD try to release a new product...

AMDs naming scheme is NOT misleading. 
It tells you the computer is new, it tells you what cores its using, and it tells you where it is in the line up. 
I keep having to say this. all consumers cares about is how fast it is compared to price, its battery life, and if its new. Only nerds care if a car has a super charged GM 3600 or a detuned LS when both make 300hp and make the care move just as fast and burn the same amount of gas. And the nerds are told exactly what core is used.

that is why AMD calls it 7020 series, 7030 series, 7040 series. Zen 2, Zen 3, Zen 4. ALL 7000 chips are new as of 2023.
or 8020, 8030, 8040, 8050, Again, Zen 2, Zen 3, Zen 4, Zen 5. ALL 8000 chips are new as of 2024. 

Intel was the one who kept saying a core isn't a core even though ALL their CPUs have cores in varying ratios that are COMPLETLY different architectures but will just say the core count, like 10 core... when its really 2 P cores + 8 E cores. If intel wanted to complain that AMD's first number does not you which cores are used in a chip, then they cant be having their line up be a mess like that. 


Not requiring a brand new OS scheduler is a good thing as it leads to weird performance issues with software the OS does not know about as it shuttles things around to the wrong cores.  Yes this happens on AMD with threadripper NUMA, but you dont need to do it on AMD cores/c/x3d when on the same die. You need a brand new OS scheduler for ALL Intel mobile and desktop chips. For AMD only a select few benefit from them. Adding more E cores has minimal effect on games due to slow clocks, the scheduler, and lack of IPC. AMD c cores have the same IPC, no more then a couple hundred less clocks, and the scheduler doesn't have to figure things out as it shares cache when its on the same die. (chiplets are different as L3 are away from you)

[TetraSky]

Here I was getting excited, then realized it was just their AI focused chip and my interest fell through the floor.

Looking forward to the Zen5 chips releasing later in 2024 I guess.

[tim0901]

1 hour ago, starsmine said:

*snip*

Explain this for me then:

 

I, an uneducated consumer, purchased a laptop with an R3 5400U cpu in it nearly 3 years ago. It's done me well, but it's not the quickest anymore and I could do with an upgrade.

 

I head over to my local Best Buy, where I am greeted with a sight in front of me: the latest laptop from AMD, with an R5 7520U chip inside. Sure it has the same number of cores as my current laptop, but it's an R5, rather than an R3, which means it's more powerful right? It has more GHz too, and it's 2 generations newer! So surely this should be a suitable upgrade for me! And at such a great price too!

 

Except it isn't. Because it's a 7020 series CPU, so it's Zen2 based. While their old 5400U is Zen3 based. So the laptop our uneducated consumer has just purchased is a direct downgrade over their previous system, despite the name suggesting otherwise.

 

So please, good Sir, explain how that is not anti-consumer and misleading. Everyone knows that uneducated consumers work on the theory of bigger number = better, but you can no longer rely on that theory under AMD's new naming scheme. I don't care if "consumers only care if its new" - if you are selling a consumer a worse product, but the name suggests that it should be better unless you have significant knowledge about the product, then you are misleading them.

[starsmine]

16 hours ago, tim0901 said:

Explain this for me then:

 

I, an uneducated consumer, purchased a laptop with an R3 5400U cpu in it nearly 3 years ago. It's done me well, but it's not the quickest anymore and I could do with an upgrade.

 

I head over to my local Best Buy, where I am greeted with a sight in front of me: the latest laptop from AMD, with an R5 7520U chip inside. Sure it has the same number of cores as my current laptop, but it's an R5, rather than an R3, which means it's more powerful right? It has more GHz too, and it's 2 generations newer! So surely this should be a suitable upgrade for me! And at such a great price too!

 

Except it isn't. Because it's a 7020 series CPU, so it's Zen2 based. While their old 5400U is Zen3 based. So the laptop our uneducated consumer has just purchased is a direct downgrade over their previous system, despite the name suggesting otherwise.

 

So please, good Sir, explain how that is not anti-consumer and misleading. Everyone knows that uneducated consumers work on the theory of bigger number = better, but you can no longer rely on that theory under AMD's new naming scheme. I don't care if "consumers only care if its new" - if you are selling a consumer a worse product, but the name suggests that it should be better unless you have significant knowledge about the product, then you are misleading them.

anyone who says an r3 5400U no longer meets their needs so they are going to go buy a new 350USD laptop is taking the piss. The 5400U is still blazing fast and only one generation old. And to say they purchased it nearly 3 years ago isn't being honest. It takes a quarter from release to finding laptops with it. The chip came out in Q1 2021, the earliest you could realistically buy laptops with it was late Q2. 

But besides that, a 7520U is faster in some tasks. It uses DDR5 and RDNA 2 graphics with AV1 Decoding vs vega, and uses less power so you have better battery life, AND its quick enough (just like the r3 5400U is) To call it a direct downgrade is bullshiting when it is just a wash and is a side grade. You have a new warranty, you have whatever new IO you wanted you didnt have before. 
I do agree this specific chip should probably be better off being called a 7325U or 7420U, but its one chip in the entire line up that Im not fond of the name of. 



Really though anyone trying to upgrade 5400U needs to actually ask what they are doing, there is no need to. 

[LAwLz]

Sigh...

AMD once again playing to the cheap seats.

 

 

Why does it matter to me as a consumer if something "require hardware OS scheduler" or not?

The whole "all cores are efficient" is kind of bullshit too. "Efficient" is not a binary thing where something is or isn't efficient, and Zen 4 cores are far less efficient than Zen 4c cores for example. In fact, Zen 4c cores being more efficient than Zen 4 cores were a selling point back when AMD announced them.

 

 

I think this will just blow up in AMD's face as they start leaning more into the hybrid designs that they are already pursuing.

There is a reason why they list Zen4 and Zen4c cores separately and run them at different speeds. There is a reason why they AMD makes distinctions between the two. Because they are not the same cores. They might have the same instruction set. They might have the same IPC. But they are not the same. But does that actually matter? It doesn't to me, just like it doesn't matter that the E-cores and P-cores have different characteristics. What matters is performance, price, efficiency, features and so on.

 

If one CPU can transcode a video in 1 minute and the other does it in 2 minutes then I don't really care if one had E-cores or Zen4c cores, Zen4 cores or P-cores. 

The cores things are interesting technical aspects that are fun to read about, but I don't think they serve any place in marketing material where one thing gets presented as "better" because it does something differently. 

 

 

Also, I really don't get why some people seem to dislike E-cores so much. I feel like it has become some cult-like thing where people get indoctrinated into thinking that E-cores = bad without putting any thought into it more than that.

[SpaceGhostC2C]

"Our bar: TALL. Their bar: short."

[Forbidden Wafer]

5 hours ago, LAwLz said:

Also, I really don't get why some people seem to dislike E-cores so much. I feel like it has become some cult-like thing where people get indoctrinated into thinking that E-cores = bad without putting any thought into it more than that.

They are garbage. I was running some single core simulations and got awful performance on a 13900H with DDR5 5600MTs compared to my good old Ryzen 1700 at stock clocks (3.2Ghz) and DDR4 2666 (couldn't get higher clocks stable).  

 

The simulation is memory bound, so it made no sense. I only figured out the E-cores were the problem after pinning the simulator to the different cores.

 

And as @leadeater pointed out, it is due different micro architecture with smaller load-store units and less load ports, which I later confirmed with VTune.

[leadeater]

7 hours ago, SpaceGhostC2C said:

"Our bar: TALL. Their bar: short."

Plot twist, chart was on time and smaller is better 

[Quackers101]

punch? it seems more like a pissing match.
Also generative AI? lol.

[RejZoR]

On 12/7/2023 at 12:23 AM, starsmine said:

anyone who says an r3 5400U no longer meets their needs so they are going to go buy a new 350USD laptop is taking the piss. The 5400U is still blazing fast and only one generation old. And to say they purchased it nearly 3 years ago isn't being honest. It takes a quarter from release to finding laptops with it. The chip came out in Q1 2021, the earliest you could realistically buy laptops with it was late Q2. 

But besides that, a 7520U is faster in some tasks. It uses DDR5 and RDNA 2 graphics with AV1 Decoding vs vega, and uses less power so you have better battery life, AND its quick enough (just like the r3 5400U is) To call it a direct downgrade is bullshiting when it is just a wash and is a side grade. You have a new warranty, you have whatever new IO you wanted you didnt have before. 
I do agree this specific chip should probably be better off being called a 7325U or 7420U, but its one chip in the entire line up that Im not fond of the name of. 



Really though anyone trying to upgrade 5400U needs to actually ask what they are doing, there is no need to. 

I still use Ryzen 5 2500U laptop for multimedia and office use and it's snappy enough that I don't even consider buying replacement. And I have a 5800X3D desktop so I'd notice the speed difference pretty quickly. I am running Manjaro GNOME on it because of Microsoft's BS with arbitrary hardware requirements for Windows 11, but that doesn't change the fact that Firefox is just as snappy on it as it is on 5800X3D in the desktop. And that's where most of the use is anyway.

 

5400U is not only clocked higher, it also has much larger caches and is built on much newer architecture. Should be plenty faster despite same core count.