Intel Arc a350m & a370m: entry-level graphics Cards for Notebooks Presented
The cat is out of the bag, what takes a long time will finally be well and many more sayings are appropriate. Intel has finally officially introduced the two notebook graphics cards A350M & A370M. These are in the entry-level segment and larger models are to follow in the summer.
The Arc mobile graphics card family will consist of five different models. Intel has now officially presented the two entry-level models and roughly shown the other three.
The two models shown, Arc A350M and Arc A370M, are based on the ACM-G11 GPU, while the other three graphics cards will be based on the more powerful ACM-G10.
Here is a rough overview of the next five models:
| A350M | A370M | A550M | A730M | A770M | |
|---|---|---|---|---|---|
| GPU | ACM-G11 | ACM-G10 | |||
| Xe-Cores | 6 | 8 | 16 | 24 | 32 |
| FP32-ALUs | 768 | 1.024 | 2.048 | 3.072 | 4.096 |
| RT-Cores | 6 | 8 | 16 | 24 | 32 |
| Matrix-Cores | 96 | 128 | 256 | 384 | 512 |
| Graphics-Clock | 1.150 MHz | 1.550 MHz | 900 MHz | 1.100 MHz | 1.650 MHz |
| FP32-Performance | 1,8 TFLOPS | 3,2 TFLOPS | 3,7 TFLOPS | 6,8 TFLOPS | 13,5 TFLOPS |
| FP16-Performance | 3,6 TFLOPS | 6,4 TFLOPS | 7,4 TFLOPS | 13,6 TFLOPS | 27 TFLOPS |
| FP16-Performance over Tensor | 14,1 TFLOPS | 25,4 TFLOPS | 29,5 TFLOPS | 54,1 TFLOPS | 108,1 TFLOPS |
| Texture units | 48 | 64 | 128 | 192 | 256 |
| Storage | 4 GB GDDR6 | 8 GB GDDR6 | 12 GB GDDR6 | 16 GB GDDR6 | |
| Storage Interface | 64 Bit | 128 Bit | 192 Bit | 256 Bit | |
| L2-Cache | 4 MB | ? | ? | 16 MB | |
| TDP | 25 – 35 Watt | 35 – 50 Watt | 60 – 80 Watt | 80 -120 Watt | 120 -150 Watt |
<<Lenovo Legion 5 82NL000RGE: Real Good Mid-Tier Laptop>>
Intel Arc A350M
The smallest model gets six Xe graphics cores, which corresponds to about 768 FP32 ALUs. Each core also has a unit for ray tracing. The GPU clock is 1,150 MHz and is fast enough for beginners.
The graphics card has 4 GB of current GDDR6 memory and has a 64-bit memory interface. Intel names between 25 and 30 watts as the TDP, which indicates different strong configurations. Intel has not provided any information on memory bandwidth or memory clock.
IntelArc A370M
The Arc A370M is a slightly more powerful version of the A350M. But it offers a total of 8 Xe cores and thus comes to 1,024 ALUs, and the GPU clock is also slightly higher at 1,550 MHz. The VRAM remains the same with 4 GB and 64-bit memory interface. The card may consume between 35 and 50 watts.
Intel Arc A550M
Welcome to the middle class! The Arc A550M is based on the ACM-G10 chip and offers a full 16 Xe cores and 2,048 FP-32 ALUs. However, the memory clock is significantly lower at just 900 MHz. This slows down the performance of the card and was probably chosen so that it is not dangerous for the big cards.
With 8 GB, there is twice the amount of VRAM and even a 128-bit memory interface. Double the amount of memory is worthwhile, especially for games with high-resolution textures or for video editing. The power consumption here is already a proud 60 to 80 watts.
Arc A730M and Arc A770M
The Arc A730M gets 24 Xe cores and 3,072 FP32 ALUs. The VRAM also grows to 12 GB and the memory interface to 192 bits. The almost flagship can then consume between 80 and 120 watts in operation.
The notebook flagship gets the full expansion: 32 Xe cores and 4,096 ALUs, with a GPU clock of 1,650 MHz, generous 16 GB of VRAM with a 256-bit memory interface. The high-end model can consume between 120 and 150 watts.
More small details about the new architecture
In addition to specific details about the models, Intel also talked about the Alchemist architecture. The L1 cache is 192 KB per Xe core and is installed there at the same time. With the Arc A770M it is 6,144 KB with 32 cores.
The L2 cache is 4 MB in the models with ACM-G11 and 16 MB in the others. The two caches are thus significantly larger than those of the competition. It will be exciting to see how much games benefit from this.
The Xe cores can calculate 128 FP32 ALUs per clock, but also 128 INT32 ALUs at the same time. They also master the FP16 and INT8 format. So, during one clock, a Xe core can calculate FP32, 2x FP16 and 4x INT8 at the same time. That’s actually neat.
The so-called matrix units are responsible for machine learning at Intel. There are 16 such units per Xe core. So with the Arc A770M there are 512 units. During one cycle you can calculate 128x FP16, 256x INT8 or 512x INT4/INT2 at the same time.
How the individual graphics cards will clock is still unclear. The clock behavior is very important and is constantly being improved by AMD/Nvidia. Intel only speaks of “dynamic clock rates” here and differentiates between “Graphics Clock” and “Higher Clock”. Here, too, there is no precise explanation. But I’m assuming some kind of base clock and a turbo.
All Arc graphics cards have a media engine that can handle all current codes. This also includes H.264, H.265, VP9 and AV1. Intel specifically calls up to [email protected] 12-bit HDR when decoding and up to [email protected] 10-bit HDR when encoding.
Some benchmarks from Intel
Of course, Intel didn’t miss the opportunity to show some benchmark results from their new graphics cards. However, these should be treated with a pinch of caution, as manufacturers usually choose their tests very wisely. Nevertheless, they give a good overview of what to expect. Intel emphasizes that the Arc 3 series is intended for gaming in HD resolution and medium details.
Xe Super Sampling as a DLSS Killer
Of course, Intel also has a kind of AI upscaling capability with XSS, similar to Nvidia’s famous DLSS and AMD’s FSR 2.0. Intel lists Death Stranding, Grid Legends, Hitman 3, Shadow of the Tomb Raider and Ghostwire Tokyo as supported launch titles.
<<Acer Nitro RX321QU Pbmiipx: No frills>>
The function is not yet available when the small graphics cards appear and should also be submitted in the summer.
There are enough connections
Intel calls DisplayPort 1.4a and DisplayPort 2.0 ready for a graphics card. Currently there is no DP 2.0, because the final specifications are not yet clear. The function should therefore be able to be activated later with an update.
There is also HDMI 2.0b, which unfortunately isn’t quite as fresh as new. If the DP port is unlocked to DP 2.0, it should be able to do two times 8K60 HDR and four times 4K120 HDR. Which sounds pretty neat.