AMD AI max+ pro 395

[Corona Scurrae]

Seems like AMD might be planning to release this SOC for the enthusiast market

[Bad 💀 Motha]

Yes good, for an APU... but again, it's still just an APU.
Radeon 6600 performance is still just 1080p Gaming; nothing more.

My real major point of saying "APU crap" was mostly directed towards this "possible planning to release to enthusiast market"

The enthusiast market like higher end CPUs and GPUs, which we already have.

Now sure, maybe such advancements could lead to a more powerful CPU (or GPU) down the road that doesn't require as much power, that's still a long ways off though.

APU/NPU do have their uses, but in the enthusiast market where these users are more then capable of paying $800+ for a CPU and $1000+ for a GPU... sure... OK... so with those users, where does an APU have it's place among those users, well they really don't have a place there since that's not what they are for.

[xSOSxHawkens]

Originally posted by Bad 💀 Motha:

Yes good, for an APU... but again, it's still just an APU.
Radeon 6600 performance is still just 1080p Gaming; nothing more.

My real major point of saying "APU crap" was mostly directed towards this "possible planning to release to enthusiast market"

The enthusiast market like higher end CPUs and GPUs, which we already have.

Now sure, maybe such advancements could lead to a more powerful CPU (or GPU) down the road that doesn't require as much power, that's still a long ways off though.

APU/NPU do have their uses, but in the enthusiast market where these users are more then capable of paying $800+ for a CPU and $1000+ for a GPU... sure... OK... so with those users, where does an APU have it's place among those users, well they really don't have a place there since that's not what they are for.

not all enthusiasts want the top end performance, some want other priorites. Thats what the OP is trying to get through to you. Small Form Factor builds will be insane with this chip. Having a *small* nuc style (not like the Vega equiped Intel+AMD hybrid that once was, but actual tiny bricks) that are fully cappable of AAA gaming *and* Ai models is a reality with this chip. Its the closest thing to the new Mac chips on the x86 side of the spectrum, and has all the same amazing benefits that it can bring to the desktop PC platform.

Its frankly an amazing chip. I am excited to see if Intel can answer back, because they are *starting* to get their iGPU's into shape too. Would be great if the two got into an arms race in the APU market.

[Tiberius]

Originally posted by xSOSxHawkens:

Originally posted by A&A:

I don't know where you take this off. It is UP TO 50TFlops (FT8? Nothing said about that). A 5080 can do 56TFlops FT16 and 32 so and 8.

https://wccftech.com/amd-ryzen-ai-max-395-strix-halo-apu-over-3x-faster-rtx-5080-in-deepseek-benchmarks/

TFlops dont matter without the memory capacity to match when the Ai workloads need the information in memory to push through the TFlops...

For clarity the 5080 is still ahead as long as the Ai memory load is 16gb or less, but the moment it goes past that the APU wipes the floor with it. Same reason Apples new SOC is such an Ai power house. Unified memory pools up to 128GB.

Damn very impressive stuff. I cant wait to see what they can do with medusa halo and UDNA

[HypersleepyNaputunia]

Would like to know hiw many watts it uses

[xSOSxHawkens]

Originally posted by 我很高兴认识你:

Would like to know hiw many watts it uses

About 55w in most laptop and similar deployments.

Though I think on high end mobile or desktop deployments it has a configurable TDP up to 140w

[Bad 💀 Motha]

Originally posted by xSOSxHawkens:

Originally posted by Bad 💀 Motha:

Yes good, for an APU... but again, it's still just an APU.
Radeon 6600 performance is still just 1080p Gaming; nothing more.

My real major point of saying "APU crap" was mostly directed towards this "possible planning to release to enthusiast market"

The enthusiast market like higher end CPUs and GPUs, which we already have.

Now sure, maybe such advancements could lead to a more powerful CPU (or GPU) down the road that doesn't require as much power, that's still a long ways off though.

APU/NPU do have their uses, but in the enthusiast market where these users are more then capable of paying $800+ for a CPU and $1000+ for a GPU... sure... OK... so with those users, where does an APU have it's place among those users, well they really don't have a place there since that's not what they are for.

not all enthusiasts want the top end performance, some want other priorites. Thats what the OP is trying to get through to you. Small Form Factor builds will be insane with this chip. Having a *small* nuc style (not like the Vega equiped Intel+AMD hybrid that once was, but actual tiny bricks) that are fully cappable of AAA gaming *and* Ai models is a reality with this chip. Its the closest thing to the new Mac chips on the x86 side of the spectrum, and has all the same amazing benefits that it can bring to the desktop PC platform.

Its frankly an amazing chip. I am excited to see if Intel can answer back, because they are *starting* to get their iGPU's into shape too. Would be great if the two got into an arms race in the APU market.

LOL that's the definition of an enthusiast. But ok whatever you say Mr Delusional

[xSOSxHawkens]

Originally posted by Bad 💀 Motha:

Originally posted by xSOSxHawkens:

not all enthusiasts want the top end performance, some want other priorites. Thats what the OP is trying to get through to you. Small Form Factor builds will be insane with this chip. Having a *small* nuc style (not like the Vega equiped Intel+AMD hybrid that once was, but actual tiny bricks) that are fully cappable of AAA gaming *and* Ai models is a reality with this chip. Its the closest thing to the new Mac chips on the x86 side of the spectrum, and has all the same amazing benefits that it can bring to the desktop PC platform.

Its frankly an amazing chip. I am excited to see if Intel can answer back, because they are *starting* to get their iGPU's into shape too. Would be great if the two got into an arms race in the APU market.

LOL that's the definition of an enthusiast. But ok whatever you say Mr Delusional

Shh everyone... Dont tell this guy that there is an entire enthusiast market who wants the most bang for the buck possible in a small package. I guess he missed the memo all the way back when Intel and AMD both understood this for the first major gaming capable NUC lol...

I'm sure things like high end ITX means nothing to you :D

[xSOSxHawkens]

Originally posted by xSOSxHawkens:

Originally posted by 我很高兴认识你:

Would like to know hiw many watts it uses

About 55w in most laptop and similar deployments.

Though I think on high end mobile or desktop deployments it has a configurable TDP up to 140w

Though if you read the fine print in the results slide on the linked site, the 3x 5080 performance was at 55w TDP settings which is stupid impressive.

[A&A]

Originally posted by xSOSxHawkens:

https://wccftech.com/amd-ryzen-ai-max-395-strix-halo-apu-over-3x-faster-rtx-5080-in-deepseek-benchmarks/

TFlops dont matter without the memory capacity to match when the Ai workloads need the information in memory to push through the TFlops...

For clarity the 5080 is still ahead as long as the Ai memory load is 16gb or less, but the moment it goes past that the APU wipes the floor with it. Same reason Apples new SOC is such an Ai power house. Unified memory pools up to 128GB.

tl:dr
Single Nvidia consumer grade GPU: "Take a chassis by Ferrari, put a engine by Trabant."

Basicaly what I said.


Meanwhile their PRO version has 32GB, which clearly for single GPU is not enough... again. Unless they stack a few of them.

[xSOSxHawkens]

Originally posted by A&A:

Originally posted by xSOSxHawkens:

https://wccftech.com/amd-ryzen-ai-max-395-strix-halo-apu-over-3x-faster-rtx-5080-in-deepseek-benchmarks/

TFlops dont matter without the memory capacity to match when the Ai workloads need the information in memory to push through the TFlops...

For clarity the 5080 is still ahead as long as the Ai memory load is 16gb or less, but the moment it goes past that the APU wipes the floor with it. Same reason Apples new SOC is such an Ai power house. Unified memory pools up to 128GB.

tl:dr
Single Nvidia consumer grade GPU: "Take a chassis by Ferrari, put a engine by Trabant."

Basicaly what I said.


Meanwhile their PRO version has 32GB, which clearly for single GPU is not enough... again. Unless they stack a few of them.

Yeh, NV is, and sadly has been, penny pinching memory for generations now : /

[Corona Scurrae]

Originally posted by 我很高兴认识你:

Would like to know hiw many watts it uses

55W, 120W cTDP-up, 45W cTDP-down

[C1REX]

I wonder if these AI CPUs could do frame gen or upscaling. So it could take the performance hit from the GPU. DLSS4 and FSR4 look great but they are heavy.

It would be nice if a strong AI APU could help a dedicated GPU to upscale from 1080p to 4K without any hit to performance.

I also wonder if soldered into a motherboard faster RAM is the future on PC.

[¤☣wing☢zeяo☣¤™]

These things look great, I hope someone does a high end hand held with the 385, bit of a push though with 45w min TDP.

[Komarimaru]

Looks good on paper, until compare it to workstation cards with similar memory, then gets left in the dust under real work loads.

Memory, thankfully, isn't everything. Now if they could show it out performing an ada rtx 6000 or a100? Which still has less memory, I'd say it's more than just a novelty gimmick.

[xSOSxHawkens]

Originally posted by Komarimaru:

Looks good on paper, until compare it to workstation cards with similar memory, then gets left in the dust under real work loads.

Memory, thankfully, isn't everything. Now if they could show it out performing an ada rtx 6000 or a100? Which still has less memory, I'd say it's more than just a novelty gimmick.

I mean you are making a pretty stupid apples or oranges comparison. One is a few thousand for the whole machine and uses 55w of power. The other currently costs just under $8K USD just for the card lmao.

Guess what, you could purchase 4-5 entire APU equipped systems and probably have higher token rate in total across all of them (though each running its own model slower) than on one RTx-6k.

So yeh, apples to oranges, and also one that for the right workloads can be just as potent for the same total cost.

Now its down to the end users and use-case. And for most all personal Ai adventures, and for many in the industry who want a light weight Ai option, either the Mac or the AMD APU silicon will be the choice.

For anyone wanting to do pro-sumer or startup Ai they will alos probably gravitate towards either APU or Mac for Ai due to intial low cost and scalability with added units.

Only top end players who have 10's of thousands of bucks will be worried about or caring about RTX-6k cards. In which case the will likely rarely buy one, and will do the same the small people do but with GPU's instead of NPU's and just buy multiples and build a cluster. $30 will buy you an 8x6k build right now.

It doesn't just look good on paper, it *is* good. Its the compute of the second fastest consumer Ai card in the world on 1/5th the power budget and 1/2 or less the cost. For those into Ai as a small scale thing, students, researchers, consumers wanting to toy with it, this is huge. More so because it wont require transition away from x86/PC ecosystems.