Update – AMD asked us to remove this piece of content due to their NDA related to the Hawaii GPU. The article and its content will be published as soon as the NDA ends or the product is officially launched and announced by AMD.
The block diagram of the next generation AMD Radeon R9 290 series ‘Hawaii GPU’ has been created. Just recently we had the opportunity to detail the full technical specifications of the Radeon R9 290 series. Unfortunately we had to remove the item due to our NDA with AMD. Now we can look at the changes to the integral architecture that the Hawaii chip features have compared to its predecessor, Tahiti.
AMD Radeon R9 290 ‘Hawaii’ GPU block diagram detailed
The Radeon R9 290X and Radeon R9 290 both have the Hawaii GPU architecture. The Hawaii chip measures 438mm2, which is 24% larger than the Tahiti chip, which measures 389mm2. AMD spent a lot of time tweaking and organizing their Hawaii chip as it was the only chip they have actively developed since the last HD 7000 series cards were introduced a long time ago. Therefore we see a complete departure from the older GCN 1.0 design in the GCN 2.0 Hawaii architecture.
The Hawaii block diagram shows that the core is made up of eight Asynchronous Compute Engines (ACE) modules, which is twice as many as in Tahiti. In the middle of these modules is the main graphics command processor, which enables global data exchange. The Hawaii GPU consists of four shader engines, each with its own parallel linked geometry processors and rasterizers. Each shader engine contains 11 processing units with 64 stream processors, the complete Hawaii GPU Radeon R9 290X with 44 processing units and a shader number of 2816 streaming processors, 176 texture mapping units and 64 raster operator units.
All shader engines share the same pool of 16 64K L2 cache with an L2 / L1 bandwidth of up to 1 TB / s. The Hawaii GPU has eight 64-bit MCs or storage controllers that lead to a 512 -Bit bus interface with an abundant 4 GB GDDR5 memory.
In addition to the most important GEs, multimedia processors are offered, including VCE, UVD, DSPs for TrueAudio, XDMA, Eyefinity Controller and the DMA engine. Finally, there is a small PCI-e 3.0 bus interface that enables support with the latest PCI-e 3.0-capable motherboards.
A slide on the efficiency of the GCN 2.0 architecture shows the key improvements between Tahiti and Hawaii. We can see that the processing rate of the geometry with 4 billion primitives per second on Hawaii has increased 1.9 times compared to 2.1 billion on the predecessor. Similarly, texture and pixel fill rates have been increased by 30 and 90 percent, respectively, while bandwidth is 320 GB / s compared to 264 GB / s for the HD 7970. Ultimately, users on the calculator would get 12.8 peak GLops / mm2 compared to 12.2 in Tahiti. The Hawaii Radeon R9 290 series has a larger chip size of 438 mm2, so we expect up to 5.6 TFlops of computing power. AMD’s Hawaii Radeon R9 290 series launches next week. More information is available here.