Monday, June 17th 2024

AMD Ryzen AI 9 300 Posts a 20% Performance Upgrade with Both Graphics and CPU Over Previous Gen

The top-spec AMD Ryzen AI 9 300 series "Strix Point" processor, the Ryzen AI 9 HX 370, is expected to post a 20% performance improvement over both the CPU and integrated graphics fronts, over its predecessor, the Ryzen 9 8945HS "Hawk Point," according to leak by Golden Pig Upgrade. On the CPU front, the HX 370 packs a 12-core/24-thread CPU based on a combination of four "Zen 5" and eight "Zen 5c" cores. The single-thread performance gains on the basis of the "Zen 5" microarchitecture's generational IPC increase, besides higher clock speeds; while the multithreaded performance increases on account on more cores. This performance increase isn't linearly scaling with the 50% increase in core-count.

On "Hawk Point," all eight cores are "Zen 4," capable of boosting to high frequencies, with two of them being marked as CPPC preferred cores, capable of boosting the highest. On "Strix Point," however, only four cores are based on the "Zen 5" architecture and capable of boosting to high frequency bands; while the other eight are "Zen 5c," which don't boost as high. While the IPC of "Zen 5c" is identical to "Zen 5," the fact that it doesn't boost as high, means that the generational multithreaded performance gain from the core-count increase is expected to be closer to 20%, with Golden Pig Upgrade talking about a Cinebench R23 nT score of over 20000 points, with "Hawk Point" scoring around 16000 points.
Things get interesting with graphics. The new RDNA 3.5 iGPU on "Strix Point" packs 16 compute units (CU), compared to 12 CU on the "Hawk Point." These 16 CU work out to 1,024 stream processors, a 33% increase over the 768 stream processors of "Hawk Point," and yet there are many other factors that decide graphics performance besides CU count, and so Golden Pig Upgrade expects a 20% graphics performance improvement, which should make the new iGPU beat the Intel Arc Xe-LPG graphics of Core Ultra "Meteor Lake" processors by at least 20%. AMD, in its product announcement slide, claimed a 36% graphics performance lead over the Arc Graphics iGPU of the Core Ultra 9 185H processor.

As for the NPU, AMD has already claimed an AI inferencing performance of 50 TOPS, which goes a fair bit above the 40 TOPS required to meet Microsoft's Copilot+ AI PC program. This figure of NPU performance is needed for Windows to run local sessions of Copilot, minimizing back-and-forth from the Cloud, and improving privacy.
Source: HotHardware
Add your own comment

42 Comments on AMD Ryzen AI 9 300 Posts a 20% Performance Upgrade with Both Graphics and CPU Over Previous Gen

#26
Kapone33
So handhelds like the Steam Deck are using the first gen chip. This is the 3rd generation. It could mean we have a full market for handhelds based on performance and higher res screens on coming generations of handhelds. What AMD have achieved with their APUs is actually truly astonishing and has created a market that they control. The launch of the MSI Claw and its $899 price where I live and only a staunch Intel would pay $200 more for that vs the Ally. The craziest is that the Steam Deck has been in the top 10 of Global sales on Steam since it launched. I hope the next one shoves the price down from $439 but a handheld with a replaceable AM5 socket would be a dream to have in this space.

During Covid one of the best selling products were 3060 based laptops as it was the least expensive way to buy a complete PC at the time. It even effected the Steam Charts. These handhelds are in a space where a 7800XT or 4070S compete with handhelds in price. Laptops are even worse as you can pay up to $7000 for a 4090 based Gaming laptop where I live.

If the performance is indeed 20% not us as consumers but the companies that are making handhelds will put even more pressure on APU prices. Right now a 5900X is cheaper than a 8700G.
GigaherzFunnily and confusingly enough "strix" is more known to be the brand of low end ROG boards sitting below the Formula, Hero, Extreme, Zenith and Apex Brand. So no real high end assiciation for me there.
Where I live the "lowend" Strix board that I bought for AM5 was $600. Even B650E Strix boards are more expensive than X670E from some Companies. Strix for me means uncomprimised performance. The B550XE Strix is the best B550 board you can buy period. That board has features that put some X570 boards to shame.
Posted on Reply
#27
Tek-Check
GigaherzFunnily and confusingly enough "strix" is more known to be the brand of low end ROG boards sitting below the Formula, Hero, Extreme, Zenith and Apex Brand. So no real high end assiciation for me there.
There is only one AMD Strix APU generation in the world. It's AMD, not Asus...
As I said, all you need is look out for newest Ryzen 9.
If that is not enough, ask for AMD 370.
There is literally nothing confusing, as there are only two SKUs.
Posted on Reply
#28
Caring1
Nothing confusing about naming schemes, at all.
You can get an X570, slap in a 5700x3d and a 5700xt and bob's ya uncle, simplez.
Posted on Reply
#29
Random_User
Good stuff. Need the real reviews. Hopefully, AMD won't delay the release, like they usually do with their mobile and APU series, due to AI monicker time around.

I couldn't care less about naming, if it was limited to NPU/AI only series, with "ordinary" APU/mobile lineups, continuing their older garbage naming. But sadly, "there can be only one" (AI).

P.S.: they might push AI 300 support to W11 not only because W10 being EOL soon. But due to a bit better hybrid-friendly scheduler, as intel did with their Alder lake.
AusWolfWith this stupid naming, you don't even know what the previous gen is anymore.
That's the whole point. The AI products should eclipse all previous "non-intelligent" products. Core Ultra is of the same marketing "bottling".
Posted on Reply
#30
Darmok N Jalad
The Quim ReaperThey should just name all their CPUs...Bob.

Slow Bob
Medium Bob
Fast Bob
Faster Bob
Bob
Bob Pro
Bob Max
Bob Ultra

Personally, I'm waiting for Clippy Ultra for my next upgrade.
Posted on Reply
#31
ADB1979
GigaherzI dont care if the cpu increases performance by 90000% and makes my breakfast. The naming scheme now has arrived at such a high level of BS where I simply refuse to buy until they are back at a comprehensive level.
:laugh:
Posted on Reply
#32
Random_User
Darmok N JaladBob
Bob Pro
Bob Max
Bob Ultra

Personally, I'm waiting for Clippy Ultra for my next upgrade.
it's rather this way:

Rob
Rob Pro
Rob Max
Rob Ultra

Fits better in the context of higher margins. :laugh:
Posted on Reply
#33
SL2
Here's the video with AMD-Donny, and why they changed the model numbers. 11:20 minutes.

(Maybe he's out of his element..)

Posted on Reply
#34
Tek-Check
DenverThey claimed it was 50% faster than a "ULV" meteor lake chip with 4CU/XEs iGPU (Half of the 155/185H model).
Interesting. It doesn't sound encouraging.

Is there any graph or slide they showed this on?
SL2Here's the video with AMD-Donny, and why they changed the model numbers. 11:20 minutes.
His impression was that the 'hate' was press driven... I actually liked the numbering system. It was logical in 99% of cases. Sure, one could always nitpick one or two models that wouldn't fit perfectly, but hey, the world is not perfect either...

The "HX" ferish both by Intel and AMD is nonsesne. Nobody needs those letters near i9 or R9. If i9 or R9 cannot defend their top tier on their own, then letters will not help.

If anything, they could indicate desktop halo chip adapted to laptops with the last number, sich as 375, five points higher than Strix 370. Get rid off those terrible letters HX, XTX, etc. It's waste of everyone's time, trying effort and pronunciation effort.
Posted on Reply
#35
Minus Infinity
I call BS, but we will know soon enough. No way 4 Zen 5 + 8 Zen 5c and 16 CU iGPU only score 20% higher than 8 Zen 4 + 12 CU iGPU which uses slower memory. It would basically mean IPC regression and or ridiculous thermal issues resulting in much lower clocks. 33% more cores across the more and only 20% more performance, they could have just launched Rembrandt++ with more cores and achieved same result.
Posted on Reply
#36
SL2
Minus InfinityI call BS, but we will know soon enough. No way 4 Zen 5 + 8 Zen 5c and 16 CU iGPU only score 20% higher than 8 Zen 4 + 12 CU iGPU which uses slower memory. It would basically mean IPC regression and or ridiculous thermal issues resulting in much lower clocks. 33% more cores across the more and only 20% more performance, they could have just launched Rembrandt++ with more cores and achieved same result.
Yeah it doesn't make sense. Why MSI would downplay the performance at Computex tho.. it can't be ES results either.

It's all BS. Not that anyone cares or notices, as everyone's so upset with two letters in a name. Yeah it's a bad name but come on, the amount of pent up feelings in this thread is hilarious.
Posted on Reply
#37
Crazybabe
I am waiting for my AI Thermal Pad...
Posted on Reply
#38
Unregistered
This timespy score is actually extremely close to what I'm getting on a Ryzen 3600 with an RX580 Radeon, but somehow I have doubts it would translate to equal performance in real life gaming. Could it be that they've really reached 580 performance on iGPUs? That would be impressive to me.
#39
qcmadness
Minus InfinityI call BS, but we will know soon enough. No way 4 Zen 5 + 8 Zen 5c and 16 CU iGPU only score 20% higher than 8 Zen 4 + 12 CU iGPU which uses slower memory. It would basically mean IPC regression and or ridiculous thermal issues resulting in much lower clocks. 33% more cores across the more and only 20% more performance, they could have just launched Rembrandt++ with more cores and achieved same result.
Not thermal problem.
The Zen 5c are clocked much lower.

And we don't know the power consumption yet.
Posted on Reply
#40
R0H1T
qcmadnessNot thermal problem.
The Zen 5c are clocked much lower.

And we don't know the power consumption yet.
Much lower compared to what? Do we know zen5 or zen5c nominal speeds?

As for the IGP results it really depends on a lot of factors including but not limited to the "NPU" & how much TDP is reserved for it.
Posted on Reply
#41
Denver
Tek-CheckInteresting. It doesn't sound encouraging.

Is there any graph or slide they showed this on?


His impression was that the 'hate' was press driven... I actually liked the numbering system. It was logical in 99% of cases. Sure, one could always nitpick one or two models that wouldn't fit perfectly, but hey, the world is not perfect either...

The "HX" ferish both by Intel and AMD is nonsesne. Nobody needs those letters near i9 or R9. If i9 or R9 cannot defend their top tier on their own, then letters will not help.

If anything, they could indicate desktop halo chip adapted to laptops with the last number, sich as 375, five points higher than Strix 370. Get rid off those terrible letters HX, XTX, etc. It's waste of everyone's time, trying effort and pronunciation effort.
Take a look at the footnotes. I doubt that Lunar Lake will be revolutionary, but perhaps it will achieve performance equivalent to the 7840U using less power because it is manufactured in a superior process. I also expect lower consumption at low load by having memory in a more efficient package.
Posted on Reply
#42
Tek-Check
DenverTake a look at the footnotes. I doubt that Lunar Lake will be revolutionary, but perhaps it will achieve performance equivalent to the 7840U using less power because it is manufactured in a superior process. I also expect lower consumption at low load by having memory in a more efficient package.
Thanks. If 8-core Xe2 iGPU claim ~50% uplift over 4-core Xe iGPU in synthetic TimeSpy, that suggests, at best, gaming performance around 760M iGPU and not 780M.
Posted on Reply
Add your own comment
Jul 21st, 2025 20:57 CDT change timezone

New Forum Posts

Popular Reviews

TPU on YouTube

Controversial News Posts