News AMD Claims Ryzen AI Max+ 395 “Strix Halo” APU With Radeon 8060S Up To 68% Faster Than RTX 4070 mobile In 1080p Gaming Benchmarks

Kaching999

Kaching999

Galvanizer
AMD has shared more gaming benchmarks of its Ryzen AI Max+ 395 "Strix Halo" APU and its integrated Radeon 8060S GPU. AMD did share a few benchmarks on the AMD Ryzen AI Max+ 395 during its CES showcase, but now they have some more numbers available that compare the chip to the NVIDIA GeForce RTX 4070 laptop GPU. The numbers for the RTX 4070 were obtained on the ASUS ROG Flow Z13 laptop, and several games were used at 1080p High Settings. The numbers are native performance, so you aren't looking at upscaling or frame generation involved in these.

In all games tested (17 in total), the AMD Ryzen AI Max+ 395 was able to outperform the NVIDIA GeForce RTX 4070 laptop GPU. Now, it is not known what power configurations were used to test both systems, but here we can see the Ryzen AI Max+ 395 and its Radeon 8060 iGPU offering over 50% better performance and going as high as 68% in Borderlands 3. On average, this is a 23.2% improvement versus the RTX 4070, which is quite decent as it shows that AMD's integrated chips can now compete with high-end/mainstream laptop GPUs.

1737991423176.png

1737991465280.png


Original article: https://wccftech.com/amd-ryzen-ai-m...faster-than-rtx-4070-1080p-gaming-benchmarks/
News Source: @uzzi38
 
  • Like
Reactions: meizul, Crazy_Eddy and StygianClaw
The benchmarks have been released and honestly this looks pretty good, not quite 4070 laptop gaming performance but on par and sometimes better than the 4060 laptop gpu and this is all on an iGPU where the whole system draws less than 70 watts mostly.

Blow Your Mind Wow GIF by Product Hunt



Finally something good after the GPU launch madness.
 
  • Like
  • Wow
Reactions: meizul, Mann, nitin_g3 and 4 others
Just when I thought of getting Ryzen AI 9 HX 370 with 890m, AMD dropped a better one.

This keeps getting better, even Intel's lunar lake one with Ultra 7 258v is very good with power efficiency.
 
  • Like
Reactions: Kaching999
Skoomahed said:
I wonder how these perform for AI tasks like running LLMs locally. Being able to allocate more VRAM sounds like a nifty feature for these tasks especially
Click to expand...
Check out the hardware canucks video they go over it in that , it can easily run a 14b model with about 25 tokens/s, not anywhere close to the M4 due to it running on the proprietary MLX, but will only get better with adoption of ROCm in different workloads. It can do even better with the text generation models than 25tk/s but the ASUS z13 only comes with a maximum of 32gb of ram so the cpu kinda starts to starve for resources, after allocating anything less than 8gb of ram.
 
  • Wow
  • Like
Reactions: meizul and Skoomahed
Just imagine this beast in a gaming handheld, but again it consumes so much power if we want the full performance (120W against 30-35W) that battery life would be a nightmare (which is already the case with most handhelds).
 
  • Like
Reactions: Kaching999
N4GR1K said:
Just imagine this beast in a gaming handheld, but again it consumes so much power if we want the full performance (120W against 30-35W) that battery life would be a nightmare (which is already the case with most handhelds).
Click to expand...
There is also the 380 version with 8 cores and 16 threads with a 8050s gpu, that should have much better efficiency due to the lesser number of cores, should be very interesting to see where that one lands, it will a hundred percent outpace the 890m at 15-25 watts.
 
  • Wow
  • Love
Reactions: meizul and N4GR1K
djo1093 said:
Will the ram be used as VRAM If so that seems great
Click to expand...
The way it is priced, it seems being able to use the slower unified RAM as VRAM is probably the only benefit. For its TDP and price, it doesn't differentiate itself from a discrete laptop GPU in any way.
Kaching999 said:
There is also the 380 version with 8 cores and 16 threads with a 8050s gpu, that should have much better efficiency due to the lesser number of cores, should be very interesting to see where that one lands, it will a hundred percent outpace the 890m at 15-25 watts.
Click to expand...
The 8060s/8050s are RDNA 3.5 parts with more compute units. Architecturally, it is the same as the 890m, only difference is that they packed more CUs in a chip. Larger silicon is always going to be costlier, makes zero sense to pack an extremely expensive chip and then hobble it in a handheld.
 
  • Like
Reactions: N4GR1K
t3chg33k said:
TDP and price, it doesn't differentiate itself from a discrete laptop GPU in any way.
Click to expand...
how so? the tdp here means the entire power being consumed by the CPU+iGPU since it is integrated with the CPU and it consumes a maximum of 80-90 watts sustained, when you see more than that in the test it is actually only for short periods of time. It performs better than a 4060 while consuming nowhere close to the power consumed by it, the 4060 alone can consume about a 100 watts to provide most of its performance.

Also on the price part, it is more expensive now as it just came out and is only available on a sort of experimental product by ASUS which was already expensive when it came with discrete GPUs before this so expect the price to be relatively much lower in another model by some other manufacturer like HP or Acer, it might even come in mini pcs. Also this is the maxed out 395 version the 380 will be much cheaper than this and will have the same feature set of being able to allocate the VRAM.
t3chg33k said:
The 8060s/8050s are RDNA 3.5 parts with more compute units. Architecturally, it is the same as the 890m, only difference is that they packed more CUs in a chip.
Click to expand...
While that is true, it is also true that the more compute units don't really scale linearly in terms of power draw to achieve the same and actually better performance. Yes it will be more expensive than something with a 890m, but it will also have a much higher headroom in terms of performance and what they can do. Consider these to be like the 90/80 class of the iGPUs where as the 890/880m can be thought of as the 70/60 class, they can perform decently for lighter loads but lack the headroom to scale up.
 
Top Bottom