Nvidia’s flagship gaming GPU “GA102” gets a beautiful tie shot

The GA102 GPU is Nvidia’s largest gaming GPU for gaming and consumer segment. It is used on top of the GeForce RTX 3090 and GeForce RTX 3080 graphics cards, and many other products are planned that will be seen in the GeForce and Quattro segments in the coming months. So far, we have only seen the physical GPU, but we do not see the exact tie shot of the chip except for the renders provided by Nvidia.

Nvidia’s flagship Ampere Gaming GPU, The GA102, Gets First Die-Shot – Powers GeForce RTX3090, GeForce RTX3080 & Upcoming Quadro Line

The first tie shots are now provided by no one else Fritz Fritz Well versed in taking high resolution pictures of various CPUs and GPUs. Fritz’s most recent photo highlights the Gen 3 Dyce featured on the Risen 5000 CPUs.

Nvidia GeForce RTX 3080D HP OEM driver found – GA102 based issue with 20GB VRAM

To get these high resolution images, a GeForce RTX 3090 graphics card was used. Since the cooler was removed and modern GPUs did not have the same IHS as the old one, with the most exposed and specialized equipment, Fritz was able to get a detailed picture of the full amp GA102 GPU death.

Review the NVIDIA GA102 GPU Specifications

To revisit everything we know about the Nvidia GA102 Ampere GPU, the chip has yet to find a fully powered SKU (yet) on the market. The full GA102 GPU is made up of 7 graphics processing clusters with 12 SM units in each cluster. It produces 84 SM units for a total of 10752 cores in a 28.3 billion transistor set measuring 628.4 mm2. Includes 10MP L1, 6MP L2 cache, and many ROPs, DMUs, memory controllers and NVLng Highspeed I / O Hub.

Asus GeForce RTX 3090, RTX 3080, RTX 3070 Rock Strikes White Graphics Cards with Image, Beautiful and High Factory Overlock

The Ampere SM is divided into four processing modules, each with 32 FP 32 cores, 16 IND 32 cores, a tensioner core, a warp scheduler and a dispatch unit. This is possible with an updated database with a data path that provides 16 FP32 execution units, while the other provides 16 FP32 or 16 INT32 execution units. It includes 128 fp 32 cores, 64 INT 32 cores, 4 tensor, 4 wrap planners and 4 dispatch units in one ampere SM. Each module includes a new L0 instruction cache and 64 kb recording file for a total of 256 kb per SM.

Additional insights into specific areas of GPU death have been described by Twitter colleague Lokusa, who always shares insightful knowledge on his Twitter feed and has mapped out various GPU / CPU die shots in the past. Of Lokusa Below you can see the amazing work he did to map the Ampere GA102 GPU accurately mapped with every aspect and area of ​​the chip. His works can be seen in the image below:

Again, this is an impressive product from both Fritz and Locusa and we can’t wait to do the same with AMD’s Big Navi (Navi 21) GPU featured on the Radeon RX6900 and RX6800 series graphics cards.

Source

Leave a Comment