Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)

[DisEnchantment]

Speculate at will

 

[DrMrLordX]

What I meant is that Krackans MSRP is the same as Hawk Points MSRP when it was new, has nothing to do with what Hawk costs now. It's a 1:1 price tier replacement, nothing more, nothing less.

We can't make that assumption off one questionable offering, though if that's going to be the case for all of Kraken then that's a mistake. Kraken was supposed to occupy the new price point for Hawk Point, not the launch price point. Low end Strix Point was supposed to slide in where high(er) end Phoenix/Hawk Point launched.

 

[moinmoin]

Btw. anybody knows what's the meaning behind "Krackan"? Searching for it outside of AMD makes it looks like little more than a misspelling.

 

[leoneazzurro]

XMG (Tongfang) launches a barebone gaming laptop line with the 9955HX as a base option and 9955X3D available!

XMG NEO 16 (A25) - Gaming laptop with RTX 50 and AMD Ryzen

The latest generation of the XMG NEO 16 is ready to defend its title as the world's fastest laptop for gaming and content creation.

www.xmg.gg

And the version with 9955HX (not 3D) costs less than the equivalent version with a Intel Ultra 9 275HX, while the one with X3D around 170€ more!

No reason at all to go Intel for this category of laptops this year, I'd say.

 

[eek2121]

I hate this NPU stuff so much

You really shouldn’t. Ignore the AI LLM hype. NPUs have legitimate uses and can accelerate many workloads. Saying you hate NPUs is like someone saying they hate FPUs.

Even for gaming, NPUs open up a whole bunch of cool opportunities.

I dislike the AI hype with a burning passion, but you won’t find me hating on local AI/machine learning at all.

 

[Hitman928]

Even for gaming, NPUs open up a whole bunch of cool opportunities.

I’m curious what opportunities you see for NPUs in gaming.

 

[Kepler_L2]

Btw. anybody knows what's the meaning behind "Krackan"? Searching for it outside of AMD makes it looks like little more than a misspelling.

Legal-friendly spelling of Kraken

 

[Philste]

We can't make that assumption off one questionable offering, though if that's going to be the case for all of Kraken then that's a mistake. Kraken was supposed to occupy the new price point for Hawk Point, not the launch price point.

Cheapest Yoga Lenovo presented at MWC (Slim 7) is also STARTING at 999€, with AI 7 350 being the top config. Also Krackan is said to be at least the same size if not bigger than Hawk Point, why would they sell it cheaper?

Btw, AMD deleted all Die size measurements from their spec sheets when Strix Halo and Krackan Spec sheets went online.

 

[MS_AT]

I’m curious what opportunities you see for NPUs in gaming.

One that M$ attempted with Qualcomm was DLSS-like functionality what could be a fallback option if the GPU is unable to handle this itself, probably it's not suitable if you have physically separate memory pools (like with dGPUs) but devices like Strix Halo, could care less since the frame-buffer sits anyway in unified memory.
Maybe offloading texture decompression.
Finally a more controversial use case could be voicing background NPCs (the chit-chats could be generated for each sessions keeping them fresh and since it's a background it would not need to be anything meaningful and the voice wouldn't need to be exactly of high quality).

But might be I am looking for a problems that this solution could solve😉

 

[Glo.]

I’m curious what opportunities you see for NPUs in gaming.

For example you can have properly acting like "real" players of your team in games like eFootball, EA Sports FC.

 

[adroc_thurston]

For example you can have properly acting like "real" players of your team in games like eFootball, EA Sports FC.

uh, no?

I’m curious what opportunities you see for NPUs in gaming.

there are ~none~.

 

[eek2121]

uh, no?

there are ~none~.

Incorrect, unless you mean currently.

A few articles to get you digging:

Introducing Neural Processor Unit (NPU) support in DirectML (developer preview) - DirectX Developer Blog

With the release of DirectML 1.13.1 and the ONNX Runtime 1.17, we are excited to announce developer preview support for NPU acceleration in DirectML, the machine learning platform API for Windows. This developer preview enables support for a subset of models on new Windows 11 devices with Intel®...

devblogs.microsoft.com

Gaming with Windows ML - DirectX Developer Blog

Neural Networks Will Revolutionize Gaming Earlier this month, Microsoft announced the availability of Windows Machine Learning. We mentioned the wide-ranging applications of WinML on areas as diverse as security, productivity, and the internet of things. We even showed how WinML can be used to...

devblogs.microsoft.com

unfortunately I have a busy day, so don’t have time to post more, but Google has plenty of results.

 

[adroc_thurston]

Incorrect, unless you mean currently.

A few articles to get you digging:

DirectML != D3D. Has nothing to do with vidya.
NPUs are a tumor.

 

[eek2121]

DirectML != D3D. Has nothing to do with vidya.
NPUs are a tumor.

Go read the links. Microsoft is investing heavily in this area. I did a bit of reading on this a while ago. Enhanced AI, accelerated level generation, etc. there are many uses that have been named. The software is still very much WIP. If I have time later, I may post more. For now, Zoom call.

 

[biostud]

I’m curious what opportunities you see for NPUs in gaming.

Maybe something crazy like real AI for the NPC and enemies in a game...

But I guess it's more awesome to have a bunch of improved Clippy assistants in Windows. 😛

 

[adroc_thurston]

Go read the links. Microsoft is investing heavily in this area. I did a bit of reading on this a while ago. Enhanced AI, accelerated level generation, etc. there are many uses that have been named. The software is still very much WIP. If I have time later, I may post more. For now, Zoom call.

That's a lot of words to say "it's useless".

 

[Hitman928]

Go read the links. Microsoft is investing heavily in this area. I did a bit of reading on this a while ago. Enhanced AI, accelerated level generation, etc. there are many uses that have been named. The software is still very much WIP. If I have time later, I may post more. For now, Zoom call.

I didn't really see anything of note outside of things that GPUs are already doing in gaming, or could do better than an NPU. Since we're talking gaming, the GPU being there is already a given.

 

[Timorous]

I’m curious what opportunities you see for NPUs in gaming.

Only thing I can think of are to replace some of the canned NPC responses in games. A bit like those Skyrim mods that hooked up NPCs to chatGPT so they could more dynamically respond to player prompts with voice.

So something like Elder Scrolls 6 shipping with a game appropriate LLM that can be used by the ancillary NPCs to enhance immersion.

Aside from that edge case, I don't see a use for it.

 

[Philste]

You really shouldn’t. Ignore the AI LLM hype. NPUs have legitimate uses and can accelerate many workloads. Saying you hate NPUs is like someone saying they hate FPUs.

Even for gaming, NPUs open up a whole bunch of cool opportunities.

That might be completely true, but I don't see it being useful for me as a normal basic customer in the next few Years. If you work with it and actually know you can use it it is possibly great.

Only thing I see is that I bought a Copilot+ PC in October and still have no Copilot+ functions despite MS and Asus still claiming available in November (2024).

 

[DrMrLordX]

Cheapest Yoga Lenovo presented at MWC (Slim 7) is also STARTING at 999€, with AI 7 350 being the top config.

So long as Intel keeps cutting margins on their mobile processors, AMD will need to do the same. There are some nice Hawk Point lappies for sale right now in the $800 range in the US market.

Also Krackan is said to be at least the same size if not bigger than Hawk Point, why would they sell it cheaper?

To compete with Intel dumping product at reduced margin. Strix Point is taking over for Hawk Point/Phoenix at the high end, so Kraken needs to slide in underneath it.

 

[HurleyBird]

there are ~none~.

None right now, but there are obviously hypotheticals. There's a scenario where NPUs make sense to free up GPU resources when games start to incorporate local AI models while *also* wanting to look pretty. Even if the GPU has enough compute to spare, memory capacity is tricky.

Now, whether on the system level it would just make sense to have more a powerful GPU with gobs more memory, or move towards gigantic APUs with shared memory, is an open question. But from the perspective of selling CPUs that "boost your frames" (or free up enough VRAM so that your frames don't tank), then sure.

 

[Glo.]

None right now, but there are obviously hypotheticals. There's a scenario where NPUs make sense to free up GPU resources when games start to incorporate local AI models while *also* wanting to look pretty. Even if the GPU has enough compute to spare, memory capacity is tricky.

Now, whether on the system level it would just make sense to have more a powerful GPU with gobs more memory, or move towards gigantic APUs with shared memory, is an open question. But from the perspective of selling CPUs that "boost your frames" (or free up enough VRAM so that your frames don't tank), then sure.

I wonder what will be said when PS6 will ship with an NPU and devs will use it "heavily" in next gen games...

P.S. NPUs will be extremely useful in APUs like Strix/Medusa Halo.

 

[LightningZ71]

The big, unanswered question for me with APUs equipped with nontrivial NPUs is about memory bandwidth. Most APUs out there are living in a cost controlled environment that heavily constrains memory bandwidth. Strix Point has a decently beefy iGPU and is saddled with lower spec "dual" channel DDR5. Similar goes for Arrow Lake H. Even Strix Halo with it's 256 but memory interface is still getting saturated with a 40cu iGPU. Where's the extra memory bandwidth going to come from to keep the NPU properly fed? When the NPU gets used for anything but the most trivial tasks, does the frame rate collapse because it's fighting with both the iGPU and the CPU cores for what scraps of bandwidth are available?

This is one area where I see Apple having a slight advantage as they tend to provide a bit more memory bandwidth compared to the capabilities of their processor specs. Performance should fall off a bit more gracefully there as all three subsystems get loaded.

 

[darkswordsman17]

For the person talking about Microsoft, uh, yeah have fun with that. Microsoft is more interested in using AI to replace game developers so they can get even more control over the gaming market than they are for meaningful things that would be fun. Oh look at that, Cortana will order you some Mt Dew and Flamin Hot Cheetos so you can keep...doing whatever it is that Halo is these days. Extreme! Cool, now if they'll just hurry up with the robots, we'll get Cartman's mom from that South Park episode. Lit! Leet gamerz all want that! Wait, they can use AI to combine Call of Duty with Flight Simulator, so you can play as one of the passengers on the 9/11 planes! Press X to pay respect! Or maybe they'll let you see what flying on Epstein's plane was like? Or better yet, they'll let you change your voice, so all the pedophiles and white supremacists can groom kids on Xbox Live!

This talk about how they are gonna do all this stuff reminds me of the crypto evangelists that swore it was gonna save the planet, by solving climate change (cause crypto is super duper efficient!), economic inequality, and not even joking they claimed it would solve world hunger. I was going to say that the "NPUs are gonna be so useful for gaming!" crowd as not being as delusional, but I'd guess they are, mostly because its a much lower stakes argument. So its gonna fix dialogue in games, uh, ok I guess if you want a bunch of pointless dialogue added to show off their chatbot, I don't. It'll be like having a game full of...whatever the fairie from Ocarina of Time "Hey, hey, listen!!!" But then I also am not impressed by chatbots in general. I couldn't care less about how "human sounding" they are when they still can't simply do the things I actually want, like auto-cancelling my previous car insurance plan after I have to manually redo it because that's the only way to actually check prices properly and save because they deliberately designed it that way to frustrate people into just keep rolling with their current while they increase price. If you think corporations are actually gonna let you use AI to save money or do anything but simplify things into a "Don't think! Buy NOW!" for everything, you're delusional. I don't care if it can do multi-step tasks when its things I'd specifically want to be a part of. Like I keep seeing stuff like them being shown booking a flight or finding a restaurant, when its simply doing a simple Google Maps search of basic restaurant info, no comparison, no factoring in price or other. And you know that's gonna be heavily adjusted based on things like restaurants playing along with Google and the like. Likewise, I couldn't care less about it booking a vacation when its not doing anything like comparing prices and features and the like, because that's not going to be what it does, its not there to actually help you, its gonna be there to make everything impulse purchases so you just buy buy buy without thinking. And then its gonna hound you endlessly when you don't buy. I can't wait for DoorDash simulator on my Meta Quest 5, it'll be so realistic, I can even do it while actually DoorDashing, I can make virtual money while being exploited in real life!

The magical thinking that somehow, despite them slopping everything else up with AI, and them slopping up gaming already, they're not gonna slop up gaming with AI slop, is just more of a fantasyland than the Mushroom Kingdom. We're ending up with this bizarro thing where AI liars are promising the magical "plug in and learn anything and everything, instantly!" meta Matrix (aka, the movie, where they could just plug in and have this grand story and blah blah blah, you get all that while not having to actually live in that horrible world), while they're leading us to the world the Matrix takes place in. The more I think, the real AI race is if we can convince enough of these delusional people to Web 3.0 themselves into comas before they do that to us. I guess what I'm saying is, we need crypto, AI, and gaming to all unite! Please JHH you're our only hope, make the 6969Td, complete with a new I/O in the Geforce Link Oscillating Relaxer Yummy Human Oral Likeness Experience.

Using NPU to make decent NPC and bots in games sounds good, heck, I've argued that before even, but knowing how all the other AI is going, its not gonna be what people want.

None right now, but there are obviously hypotheticals. There's a scenario where NPUs make sense to free up GPU resources when games start to incorporate local AI models while *also* wanting to look pretty. Even if the GPU has enough compute to spare, memory capacity is tricky.

Now, whether on the system level it would just make sense to have more a powerful GPU with gobs more memory, or move towards gigantic APUs with shared memory, is an open question. But from the perspective of selling CPUs that "boost your frames" (or free up enough VRAM so that your frames don't tank), then sure.

Hypotheticals that will end up like all the other ones these liars keep spouting, not living up to their claims and instead just being annoying and forced on you. Although maybe Sega should consider a Seaman remaster. Cause, using the NPU is gonna harm performance, especially since they can't even seem to do much AI without hitting up servers which means that you'll be stuttering simply waiting for their AI model to hit up their servers. If you have enough model to run it locally its gonna be taking resources from the CPU and/or GPU. Outside of maybe making ElderScrolls conversations less awkward, but that's a huge part of the charm of those games, in my opinion, so its not even something I want to fix. Then again, a chatbot hallucinating could be fun, but chances are it'll be more annoying because you know its not a canned response, and it'll just be a less funny version. But then you'll just end up with a bunch of people falling in love with video game characters. Imagine the DraftKings enabled Fleshlight where if you do well in Madden you get real money and pleasured. If not, it strips the skin off it and drains your bank account. And dudes will be like "so realistic, that's exactly how women are!"

 

[darkswordsman17]

The big, unanswered question for me with APUs equipped with nontrivial NPUs is about memory bandwidth. Most APUs out there are living in a cost controlled environment that heavily constrains memory bandwidth. Strix Point has a decently beefy iGPU and is saddled with lower spec "dual" channel DDR5. Similar goes for Arrow Lake H. Even Strix Halo with it's 256 but memory interface is still getting saturated with a 40cu iGPU. Where's the extra memory bandwidth going to come from to keep the NPU properly fed? When the NPU gets used for anything but the most trivial tasks, does the frame rate collapse because it's fighting with both the iGPU and the CPU cores for what scraps of bandwidth are available?

This is one area where I see Apple having a slight advantage as they tend to provide a bit more memory bandwidth compared to the capabilities of their processor specs. Performance should fall off a bit more gracefully there as all three subsystems get loaded.

Exactly, using the NPU is gonna take resources from other parts. But hey, just sign up for Xbox ALive so you can keep your chatbot smart and your gameplay uninterrupted. I wonder what streamer they'll sign as their first AI Xbox voice so you can hear them scream slurs at you while their character teabags you.

 

[fastandfurious6]

0. across society and industries "creation" is plateauing, most new movies games music etc is hot trash

1. games of recent times are increasingly terrible, even AAA. Art and meaning are getting lost

2. using AI for ingame content generation (stages, voice, objects, w/e) exacerbates this, games will feel even more fake

3. using AI for "performance increase" is a pipe dream pushed by those not understanding how performance works

did I miss anything