AMD Teases Ryzen AI Max+ 395 "Strix Halo" APU 1080p Gaming Performance, Claims 68% Faster than RTX 4070M

[Cheeseball]

I'd say if it was "as fast" as an RTX 4070M, that's goddamn impressive. 68% faster? If these things land near $1200 it's a buy for sure. Replace the wifey's RTX 3060 laptop.

LOL (not laughing at you), these are nowhere near landing around $1,200.



The Z13-ACRNM took me a couple of timed drops before I could get one in 2023 and even then it released at US $2,500. Definitely not worth it for the performance alone. My X13 (also 2023, RTX 4070M, 60W TGP though) now cost the same but at least has way better battery life due to the 7940HS and being able to manually disable (park, perhaps?) cores in Armory Crate.

 

[Sound_Card]

Official name is actually 4070 laptop GPU, which is a bit of a mouthful.

You do realize that it's 120w for the SoC (meaning about half the wattage is divided to the 16 core CPU).

Completely fair game. The 4070m just got molested.

 

[hsew]

Would have been much more apples-to-apples if AMD compared it against ASUS’s Zephyrus or ProART models which come equipped with an HX370 and RTX 4060/4070. Would have probably been able to maintain the same power envelope too. Because we all know the 13900H is a thirsty girl

 

[Cheeseball]

Would have been much more apples-to-apples if AMD compared it against ASUS’s Zephyrus or ProART models which come equipped with an HX370 and RTX 4060/4070. Would have probably been able to maintain the same power envelope too. Because we all know the 13900H is a thirsty girl

The new ProArt models would've been a fair comparison since the PX13 is most likely the successor of my Flow X13 (but with a half-good OLED but 60Hz without VRR panel), unless they decide to make an updated X13 again.

The Zephyrus G14s and G16s are bigger (one inch diagonal) laptops so they have better cooling, but they're really aimed at being just gaming laptops, not 2-in-1 or Windows tablets.

 

[andrehide]

I would rather wait for the Ryzen AI Max+ Pro Ultra Turbo Ti XT 396 "Strix Halo" APU.

 

[oxrufiioxo]

I really like my 2022 X13 hopefully this is awesome. I don't use mine for gaming though.

 

[mkppo]

AMD has just published its "How to Sell" Lie Ryzen AI MAX series guide

120W APU vs a tablet. Does AMD think that little about their customers?

How about a scenario where no one is lying? Has that thought even come across your mind or did you skim the article, half understand it and had an insatible hunger to comment immediately? I see no lies here, except you accusing someone else of lying when they haven't.

Yes, but nowhere does it say that’s where the performance numbers are from. They say “similar” multiple times. What is similar? What are the power limits? What’s the airflow? What’s the runtime?

It does say, read the article. The chassis is a Z13, there are reviews of this "gaming laptop in a tablet form". The 120W TDP for the CPU + GPU is the same for both systems. One has the 395 for CPU/GPU while the other has a 13900H + 4070.

Oh and the airflow comes from centrifugal fans and runtime when plugged in is indefinite till it dies.

For some context, this is being tested Vs. a 50 W (65 W dynamic boost) 4609 core, 128 bit, 8 GB variant of the RTX "4070" laptop GPU, inside a tablet. Not the 200 W 12 GB 5888 core desktop variant, or the 115 W (140 W boost) laptop variant.

At these wattages a boosting 4060 Ti would perform similarly.

It's also the peak gain from a specific game, Borderlands 3, well known for favoring AMD GPUs, the performance gains ranged from +1% to this value.

Calling it a tablet and stopping there isn't enough tbh and needs more context. It's pretty thick and is designed to dissipate 120W, so it doesn't really matter if it's a "tablet". ASUS are also marketing it a gaming ready. If it was a 13" laptop, the result wouldn't really be much different. Also, both systems were tested at 120W because that's fair. If you're comparing against a 115W 4070, that's way more than 120W for CPU+GPU and won't even fit in the same chassis.

Essentially what AMD are trying to show is the 395 is faster than competing Intel CPU+ Nvidia GPU combos at less than 120W. Obviously comparing to higher TDP variants isn't fair because the 395 tops out at 120W. I guess it'd be closer if that 13900H was replaced with a faster Ryzen 7945hx3d or something, but the Z13 doesn't come with that so they showed what they comes with.

 

[TumbleGeorge]

Also remember the Strix Halo SoC is both a GPU and a CPU.

Yes literally SoC mean... this+RAM/VRAM+SSD or other permanent memory in the chip matrix. In this case with strix halo dynamic memory is soldered close to the chip but isn't part of it's matrix. SSD is optional in m.2 slot.

A system on a chip is an integrated circuit that compresses all of a system’s required components onto one piece of silicon.

 

[Visible Noise]

It does say, read the article. The chassis is a Z13, there are reviews of this "gaming laptop in a tablet form"

JFC, you’re telling me to read the article when you didn’t?

According to the document, AMD's reference system

The Z13 is the comparison system, not the AMD system.

My god, unbelievable.

 

[Apocalypsee]

It's sad that this chip is put on very expensive laptops, it should be put on more mainstream platform and not being under ROG tax.

 

[mkppo]

JFC, you’re telling me to read the article when you didn’t?

The Z13 is the comparison system, not the AMD system.

My god, unbelievable.

How do you even know I didn't read the article. I thought you were questioning the Intel/Nvidia system because that question at least made sense to me. Your actual question doesn't.

AMD's reference system is a reference system. It doesn't matter jack $ if it's in a tablet, desktop, server, fridge or whatever form factor. As long as it's not overclocked, it will show how the 395 performs. It's probably this chassis because it's faster than the old intel/nvidia combo and is the same chassis but can also be a generic reference laptop, you know the stuff that exists for decades everytime a new mobile CPU comes out. Stating 'reference system' is common practice. It shows how the CPU should perform when set up optimally.

Still trying to find AMD's lie here. Maybe you can chime in with some details?

 

[eldon_magi]

This, three NVME plus a way to add in a 10GbE connector, preferably via pcie slot.

it's called a mini pc with oculink.

 

[JustBenching]

That's still crazy powerful for an APU!

It is fast for sure but you cannot compare it to a mobile chip powerlimited unless you know the scaling of that chip. For example my 6700S drops like a rock in performance at 35w (it scales negatively), while at 65w it's twice as fast (vs 35).

 

[AusWolf]

AMD has just published its "How to Sell" Lie Ryzen AI MAX series guide

120W APU vs a tablet. Does AMD think that little about their customers?

1. The ROG Flow isn't a tablet. It's a laptop with the hardware built under the screen instead of under the keyboard. It's 13.4".
2. The Ryzen AI Plus Plus something (I can't be asked to remember such a long name) is a 50 W APU with a cTDP of 120 W. Do you know what a cTDP is?

I'd say if it was "as fast" as an RTX 4070M, that's goddamn impressive. 68% faster? If these things land near $1200 it's a buy for sure. Replace the wifey's RTX 3060 laptop.

Now I understand why AMD didn't spare any time for RDNA 4 at CES. This chip is probably a much bigger hit than any desktop GPU could be.

 

[TumbleGeorge]

This chip is probably a much bigger hit than any desktop GPU could be.

Oh, AMD undoubtedly wants to convince us of this in every way to make us want to buy it, since the company's and investors' profits depend on it. But in the end, for people who are not overly enthusiasts, not rich and not zombie customers who buy everything that is advertised to them, for normal customers with low incomes the acquisition price will be decisive. If there is an unreasonable premium surcharge, purchases from normal customers who think with their heads, not with their spinal cord, will not occur.

 

[hsew]

Guessing AMD isn’t gonna be interested in laptop dGPUs much longer. As in I’d be shocked to see an RX 90M series, unless those absurdly overpriced 7800M OCULink docking stations are selling better than anyone’s wildest dreams. Oh well. They haven’t been competitive in that segment for over a decade at this point. Does AMD even have 1% of the laptop dGPU market at this point? Even Intel saw the writing on the well and dipped after a single generation.

 

[AusWolf]

Oh, AMD undoubtedly wants to convince us of this in every way to make us want to buy it, since the company's and investors' profits depend on it. But in the end, for people who are not overly enthusiasts, not rich and not zombie customers who buy everything that is advertised to them, for normal customers with low incomes the acquisition price will be decisive. If there is an unreasonable premium surcharge, purchases from normal customers who think with their heads, not with their spinal cord, will not occur.

Well, I'm definitely not the target audience - the only mobile thing I was really interested in buying in the last 10 years was the Steam Deck. I've had a couple of phone swaps, but only because the battery died, and I also bought a netbook because I needed something to download stuff onto while I'm away, but that's it. I'm an oldschool "gaming = desktop PC" kind of guy otherwise.

I still admit that for people interested in mobile products, a 50 W APU that is 68% faster in games than a 50 W mobile GPU is really extraordinary (assuming that these are actual numbers, not just marketing).

Guessing AMD isn’t gonna be interested in laptop dGPUs much longer.

With APUs like this, they won't have to be.

 

[LittleBro]

Compared with individual CPU TDP and individual GPU TDP combination, this is actually much better aproach. In situations where CPU power draw is low (CPU not required so much), TDP headroom goes to iGPU, thus not limitting performance of GPU, thus yielding more fps.

Having RTX 4060 Ti performance in APU is actually impressive. Good for any light gaming machine, suitable even for 240 Hz e-sports (CS2). I expect the price to be astronomical.

 

[progste]

Just give me this chip in a reasonably compact laptop and I'm in.

 

[LittleBro]

Why laptop? It'd do great in any SFF.

 

[Durvelle27]

If they made this available in desktops I’d buy for sure. I could build 2 great SFF rigs for the kids who still use a GTX 670

 

[halcyon]

I mean, they did do a comparison vs the 4090 for LLM inference (on a model that was larger than the 4090's vram), and they do have an specific page just for LLM stuff (3rd pic in the OP), so I believe they're aware of this market space.

Yeah, this is all they posted (that I can find):

from here: https://www.amd.com/content/dam/amd...-max-series-how-to-sell-guide-competitive.pdf

With small print: "Testing as of Dec 2024 using Llama 70b 3.1 Nemotron Q4 K M quantization running through llama.cpp and LM Studio. Input prompt length 100 token prompt. System configuration for Ryzen AI Max+ 395: AMD reference board, 55W TDP, Radeon™ 8060S graphics, 128GB RAM, 1TB SSD, using Llama 3.1. Configuration for Nvidia RTX 4090: ASUS ProArt X670E-CREATOR WIFI motherboard, AMD Ryzen 9 7900X processor, 32GB system RAM, 40GB GPU memory, 1TB SSD, Windows 11. (https://blogs.nvidia.com/blog/ai-decoded-lm-studio/). Manufactures may vary configurations yielding different results. SHO-14"

Yeah, that's a lame attempt to try and sell it, but people buying it for that purpose, have the money, but are no technical idiots. They need more than that.

Guess we'll have to wait for shipping hardware, nVidia DIGITS release and actual reviews.

I doubt that many people will be buying Strix Halo $2000+ laptops/mini-PCs to run games at 1080P. Maxxed out Strix Halo 395 PCs basically a mini-workstation level configuration (cores, max ram, ram bandwidth, speed), and they are missing the opportunity in marketing it as such. AI crazed people have money burning in their hands waiting for somebody to sell them more unified memory for bigger models, gamers - not so much.

 

[Daven]

Oh, AMD undoubtedly wants to convince us of this in every way to make us want to buy it, since the company's and investors' profits depend on it. But in the end, for people who are not overly enthusiasts, not rich and not zombie customers who buy everything that is advertised to them, for normal customers with low incomes the acquisition price will be decisive. If there is an unreasonable premium surcharge, purchases from normal customers who think with their heads, not with their spinal cord, will not occur.

You are 100% describing Nvidia and Nvidia customers with this comment (zombies, give into advertising, company/investors profits over anything else, etc). This AMD SoC will be cheaper than laptops with Intel's fastest mobile CPU and an Nvidia x070 laptop chip and therefore applies to 'the rest of us' and 'many of us' have been asking for an SoC with a powerful GPU for a long time, ever since the first powerful AMD SoC in the PlayStation and Xbox.

I mean for god's sake, people are camped out for the 5090 which is only a marginally better product than before:

People already camping out for GeForce RTX 5090 ahead of launch - VideoCardz.com

I mean how many products has Lisa Su signed and put her bitmoji in presentations. I'm sorry but I can't abide by comments knocking non-Nvidia companies for zombie buying or giving into advertising on a tech site where people LITERALLY worship Nvidia.

 

[TumbleGeorge]

This AMD SoC will be cheaper than laptops with Intel's fastest mobile CPU and an Nvidia x070 laptop chip

From your mouth(keyboard), to God's ears(PC monitor)!

 

[JustBenching]

From your mouth(keyboard), to God's ears(PC monitor)!

It will also be slower but you know, zombie buyers believe the marketing

I mean there is a reason they chose a tablet to compare it against. If im not mistaken a dekstop 4070 is ~50% faster than a mobile 4070. AMD claims their APU is 68% faster than the 4070m, which makes it faster than a desktop 4070. Aight