Although Intel disclosed performance figures for its Arc Alchemist-based ACM-G10 and ACM-G11 graphics processing units (GPUs) in their formal introduction on March 30, it didn't disclose transistor counts and die sizes. As it turns out, Intel's lower-performance ACM-G11 has a rather ample transistor budget, whereas the higher-performing ACM-G10 is relatively conservative, which is comparable to that of Nvidia's G104.
Intel's smaller ACM-G11 that powers entry-level A350M/A370M mobile GPUs is apparently 157 mm^2 and packs 7.2 billion transistors, whereas the higher-end ACM-G10 that will be used both for mobile and desktop PCs is 406 mm^2 large and integrates 21.7 billion transistors, according to Hardware Unboxed (via Andreas Schilling). Since Intel uses TSMC's N6 fabrication process for its Arc Alchemist discrete GPUs, we cannot directly compare their die sizes to AMD's RDNA 2 or Nvidia's Ampere GPUs as they are made using different nodes. But we still have some interesting points to discuss here.
Die size is an important metric for a processor as it allows one to estimate its approximate cost and make some guesses about its positioning. On the other hand, the transistor count allows us to assess its planned performance level and feature set.
The ACM-G10: High Transistor Count, Ordinary Performance
With its ACM-G10 GPU that measures 406 mm^2, Intel clearly did not want to position it for enthusiasts that seek every bit of performance possible. Contemporary enthusiast-grade GPUs from AMD and Nvidia measure 519.8 mm^2 (Navi 21) and 628.4 mm^2 (GA102), respectively, so Intel's ACM-G10 is hardly designed to compete with these two. Meanwhile, with a 406 mm^2 die size, Intel should have some flexibility regarding pricing for the GPU. In any case, if Intel's high-end GPU has proper drivers, cards based on the ACM-G10 have a reasonable chance to join our list of the best gaming graphics cards around.
| Row 0 - Cell 0 | Intel ACM-G10 | AMD Navi 21 | AMD Navi 22 | Nvidia GA103 | Nvidia GA104 |
| Row 1 - Cell 0 | Arc A770M | Radeon RX 6900 XT | Radeon RX 6850M | GeForce RTX 3080 Ti Laptop | GeForce RTX 3080 Laptop |
| Performance (FP32) | 13.5 TFLOPS | 18.7 TFLOPS | 13.2 TFLOPS | 16.7 TFLOPS | 13.6 TFLOPS |
| Transistor Count | 21.7 billion | 26.8 billion | 17.2 billion | ? | 17.4 billion |
| Die Size | 406 mm^2 | 519.8 mm^2 | 335 mm^2 | 496 mm^2 | 392 mm^2 |
| Node | TSMC N6 | TSMC N7 | TSMC N7 | SF N8 | SF N8 |
* TFLOPS based off base clocks
Intel's higher-end ACM-G10 in its mobile A770M incarnation shows that the GPU offers similar FP32 compute performance to AMD's Radeon RX 6850M and Nvidia's GeForce RTX 3080 — keep in mind that FP32 performance alone does not always reflect performance in actual games, however. The ACM-G10 is 17.5% larger than AMD's Navi 22 and packs 20.7% more transistors. Compared to Nvidia's G104, it has a similar die size but features 20% more transistors.
We can speculate that Intel's extra transistors were spent on the company's advanced media encoding/decoding capabilities and certain exclusive features (e.g., XMX matrix engines for AI compute) that require fixed-function hardware, but Intel has not officially confirmed this. Keeping in mind that Intel has historically been successful with various video streaming service platforms, investing the transistor budget in advanced media encoding and decoding capabilities makes quite a lot of sense for the company as it can sell boatloads of such GPUs to such clients.
For now, we can only wonder how the ACM-G10 is going to compete against AMD's next-generation RDNA 3 and Nvidia's Ada Lovelace GPUs that are coming this fall, just several months after Intel's desktop Arc Alchemist offerings are expected to land this summer.
The ACM-G11: Full of Features
According to Intel's materials, Intel's ACM-G11 and mobile products do not really offer impressive compute performance and fall behind both AMD's Navi 24 and Nvidia's GA102.
| Row 0 - Cell 0 | Intel ACM-G11 | AMD Navi 24 | Nvidia GA107 |
| Row 1 - Cell 0 | Arc A370M | Radeon RX 6500M | GeForce RTX 3050 Ti Laptop |
| Performance (FP32) | 3 TFLOPS | 4.48 TFLOPS | 3.76 TFLOPS |
| Transistor Count | 7.2 billion | 5.4 billion | ? |
| Die Size | 157 mm^2 | 107 mm^2 | ? |
| Node | TSMC N6 | TSMC N6 | SF N8 |
The chip's die size is 157 mm^2 and it packs about 7.2 billion transistors, so it's about 50 mm^2 bigger and 33% larger in terms of transistor count than AMD's Navi 24 that's made using the same TSMC N6 manufacturing technology. Unlike AMD's low-end offering, Intel's entry-level GPU has a wider 96-bit memory bus, four display pipelines, and fully-fledged media decoding/encoding engines. It even has a PCIe 4.0 x8 interface, which explains the larger die size and higher transistor count.
While we do not know whether Intel's A370M will be faster than AMD's Radeon RX 6500M in games — on paper, it looks like it will fall behind — it might be a better inexpensive discrete GPU for desktops and laptops due to its better display support and advanced media decoding and encoding capabilities.
Since AMD commands a fairly low market share, the real rival for Intel's ACM-G11, and the associated Arc A370M and A350M, will be Nvidia's GA107 GPU that powers such solutions as GeForce RTX 3050, RTX 3050 Ti, and MX550, three high-volume products. Nvidia traditionally does not disclose transistor count and die sizes of its entry-level notebook GPUs. Meanwhile, we can guesstimate that GA107 is larger and has more transistors than Intel's new GPU. We can only wonder which one is cheaper to make as, in many cases, price attracts PC makers to standalone low-end GPUs.
Nvidia's GA107 has premium media capabilities (sans AV1 encoding in hardware) and a 128-bit memory interface, so perhaps it will be a better entry-level solution for gaming and media playback. Yet, Intel's advanced media engines will be a huge benefit for its ACM-G11 since the GPU could be used not only by PCs, but also by video streaming platforms, just like its DG1 predecessor. It looks like Intel has made the right choice packing its ACM-G11 with features and not trying to save every penny as AMD did with its Navi 24.
Summary
When it comes to die size and transistor count, Intel clearly did not want to build a big-fat GPU with its ACM-G10. While the GPU may have some extra features, it will likely barely touch the performance of desktop parts like AMD's Radeon RX 6900 XT or Nvidia's GeForce RTX 3090 Ti. Yet, it promises to be a pretty good GPU for desktops and laptops, so perhaps this is what Intel always planned to build.
With its smaller ACM-G11 GPU, Intel decided to be slightly less conservative in terms of die size and transistor count. The GPU seems to be designed primarily with notebooks in mind, and Intel did not cut down its media processing capabilities, features, memory interface, or PCIe lane count. If it all pans out, the chip won't be too small nor too big, but will be just right for budget Goldilocks shoppers.
