AMD To Refresh Instinct MI300 Series With MI350 AI Accelerator Using 4nm Node This Year
It looks like AMD is preparing a 4nm refresh of its MI300 AI accelerator known as MI350 which is planned for later this year.
AMD MI350 AI Accelerator to Feature Refreshed 4nm Architecture, Planned to Launch Later This Year
AMD's current MI300 lineup consists of the AI-optimized MI300X and compute-optimized MI300A accelerators, but it looks like the company is planning to expand its portfolio. We recently saw the appearance of the MI388X as a possible export variant for China, but AMD said it was held back from shipping. The MI388X was supposed to be another CDNA 3 offering using 5nm and 6nm process technology, but it looks like AMD has a proper refresh planned for its Instinct family later this year.
According to a report from Trend force, it looks like AMD might be launching a new part called the Instinct MI350 that will use the latest CDNA 3 architecture using TSMC's 4nm process node. Although details of the Instinct MI350 are thin, AMD itself recently teased that it A future update of the Instinct MI300 series offering advanced HBM3E capabilities. So a great architecture on the 4nm node combined with high HBM capabilities can lead to some decent benefits.
Additionally, TrendForce notes that the expansion of export controls now includes not only previously restricted AI chips from NVIDIA and AMD, such as NVIDIA A100/H100, AMD MI250/300 series, NVIDIA A800, H800, L40, L409, L400S. but also their next-generation successors such as NVIDIA's H200, B100, B200, GB200, and AMD's MI350 series. In response, HPC manufacturers have increasingly developed products that comply with the new TPP and PD standards, such as NVIDIA's adjusted H20/L20/L2, which are export-eligible.
Video cards AMD was also able to find a Singapore listing that confirms the Instinct MI350 accelerator lineup. The product has already been introduced for silicon manufacturing and optimization.
Remember that AMD will compete with both NVIDIA and Intel in the AI space. gave Blackwell B100 GPUs are in production and the B100/B200 It will be available to users soon. Meanwhile, so does Intel Announced its Gaudi 3 Accelerator. which offer 50% faster AI compute than NVIDIA H100 GPUs. So the place is heating up. I Recent MLPerf benchmarksNVIDIA and Intel were the only ones to submit their AI performance benchmarks, while AMD missed out on the spotlight as it didn't submit any numbers.
TrendForce has also shared a complete list of products that are affected by the latest version of US export controls against China. These include several current and upcoming GPUs, including AMD's Instinct MI388X and MI350 series.
US Export Controlled Products (restricted to China / until March 29):
| the seller | Product | Processing technology | Release date |
|---|---|---|---|
| NVIDIA | GB 200 | 4nm (TSMC) | 2 AH 2024 |
| NVIDIA | B200 | 4nm (TSMC) | 2 AH 2024 |
| NVIDIA | B 100 | 4nm (TSMC) | 2 AH 2024 |
| NVIDIA | H200 | 4nm (TSMC) | 11/2023 |
| NVIDIA | H100 | 4nm (TSMC) | 03/2022 |
| NVIDIA | H800 | 4nm (TSMC) | 03/2022 |
| NVIDIA | L40/L40S | 5nm (TSMC) | 10/2022 |
| NVIDIA | RTX 4090 | 5nm (TSMC) | 10/2022 |
| NVIDIA | A100 | 7nm (TSMC) | 05/2020 |
| NVIDIA | A800 | 7nm (TSMC) | 05/2020 |
| AMD | MI250 | 6nm (TSMC) | 11/2021 |
| AMD | MI250X | 6nm (TSMC) | 11/2021 |
| AMD | MI300/MI309 | 5nm (TSMC) | 11/2021 |
| AMD | MI300X/MI388X | 5nm/6nm (TSMC) | 12/2023 |
| AMD | MI350 | 4nm (TSMC) | 2 AH 2024 |

There is also AMD has confirmed its next-generation MI400 AI accelerator, which should be released in 2025. and feature a more capable architecture built for the AI-era. AMD is also working on it. ROCm has made the software suite and some blocks open source. To fine-tune its performance for AI workloads.
AMD Radeon Instinct Accelerators
| Name of the accelerator | AMD Instinct MI400 | AMD Instinct MI350X | AMD Instinct MI300X | AMD Instinct MI300A | AMD Instinct MI250X | AMD Instinct MI250 | AMD Instinct MI210 | AMD Instinct MI100 | AMD Radeon Instinct MI60 | AMD Radeon Instinct MI50 | AMD Radeon Instinct MI25 | AMD Radeon Instinct MI8 | AMD Radeon Instinct MI6 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| CPU architecture | Zen 5 (Exascale APU) | N/A | N/A | Zen 4 (Exascale APU) | N/A | N/A | N/A | N/A | N/A | N/A | N/A | N/A | N/A |
| GPU architecture | cDNA 4 | cDNA 3+? | Aqua Vingram (CDNA 3) | Aqua Vingram (CDNA 3) | Aldebaran (CDNA 2) | Aldebaran (CDNA 2) | Aldebaran (CDNA 2) | ARCHITECTURE (CDNA 1) | Vega 20 | Vega 20 | Vega 10 | Fiji Xt | Polaris 10 |
| GPU process node | 4nm | 4nm | 5nm+6nm | 5nm+6nm | 6 nm | 6 nm | 6 nm | 7nm FinFET | 7nm FinFET | 7nm FinFET | 14nm FinFET | 28nm | 14nm FinFET |
| GPU chiplet | TBD | TBD | 8 (MCM) | 8 (MCM) | 2 (MCM) 1 (per die) |
2 (MCM) 1 (per die) |
2 (MCM) 1 (per die) |
1 (monolith) | 1 (monolith) | 1 (monolith) | 1 (monolith) | 1 (monolith) | 1 (monolith) |
| GPU core | TBD | TBD | 19,456 | 14,592 | 14,080 | 13,312 | 6656 | 7680 | 4096 | 3840 | 4096 | 4096 | 2304 |
| GPU clock speed | TBD | TBD | 2100 MHz | 2100 MHz | 1700 MHz | 1700 MHz | 1700 MHz | 1500 MHz | 1800 MHz | 1725 MHz | 1500 MHz | 1000 MHz | 1237 MHz |
| INT8 Compute | TBD | TBD | 2614 Tops | 1961 Tops | 383 tops | 362 tops | 181 tops | 92.3 tops | N/A | N/A | N/A | N/A | N/A |
| FP16 Comput | TBD | TBD | 1.3 PFLOPs | 980.6 TFLOPs | 383 TFLOPs | 362 TFLOPs | 181 TFLOPs | 185 TFLOPs | 29.5 TFLOPs | 26.5 TFLOPs | 24.6 TFLOPs | 8.2 TFLOPs | 5.7 TFLOPs |
| FP32 Comput | TBD | TBD | 163.4 TFLOPs | 122.6 TFLOPs | 95.7 TFLOPs | 90.5 TFLOPs | 45.3 TFLOPs | 23.1 TFLOPs | 14.7 TFLOPs | 13.3 TFLOPs | 12.3 TFLOPs | 8.2 TFLOPs | 5.7 TFLOPs |
| FP64 Compute | TBD | TBD | 81.7 TFLOPs | 61.3 TFLOPs | 47.9 TFLOPs | 45.3 TFLOPs | 22.6 TFLOPs | 11.5 TFLOPs | 7.4 TFLOPs | 6.6 TFLOPs | 768 GFLOPs | 512 GFLOPs | 384 GFLOPs |
| VRAM | TBD | HBM3e | 192 GB HBM3 | 128GB HBM3 | 128 GB HBM2e | 128 GB HBM2e | 64 GB HBM2e | 32 GB HBM2 | 32 GB HBM2 | 16 GB HBM2 | 16 GB HBM2 | 4 GB HBM1 | 16 GB GDDR5 |
| Infinity cache | TBD | TBD | 256 MB | 256 MB | N/A | N/A | N/A | N/A | N/A | N/A | N/A | N/A | N/A |
| Memory clock | TBD | TBD | 5.2 Gbps | 5.2 Gbps | 3.2 Gbps | 3.2 Gbps | 3.2 Gbps | 1200 MHz | 1000 MHz | 1000 MHz | 945 MHz | 500 MHz | 1750 MHz |
| Memory bus | TBD | TBD | 8192 bit | 8192 bit | 8192 bit | 8192 bit | 4096 bit | 4096 bit bus | 4096 bit bus | 4096 bit bus | 2048 bit bus | 4096 bit bus | 256-bit bus |
| Memory bandwidth | TBD | TBD | 5.3 TB/s | 5.3 TB/s | 3.2 TB/s | 3.2 TB/s | 1.6 TB/s | 1.23 TB/s | 1 TB/s | 1 TB/s | 484 GB/s | 512 GB/s | 224 GB/s |
| Form factor | TBD | TBD | OAM | APU SH5 socket | OAM | OAM | Dual slot card | Dual slot, full length | Dual slot, full length | Dual slot, full length | Dual slot, full length | Double slot, half length | Single slot, full length |
| Cooling | TBD | TBD | Passive cooling | Passive cooling | Passive cooling | Passive cooling | Passive cooling | Passive cooling | Passive cooling | Passive cooling | Passive cooling | Passive cooling | Passive cooling |
| TDP (Maximum) | TBD | TBD | 750W | 760W | 560W | 500W | 300W | 300W | 300W | 300W | 300W | 175W | 150W |
