Trinity: The next AMD APU Evolution
The concept of AMD’s Fusion Technology does not change with Trinity. Just as AMD engineered the GPU on-die with Zacate and Llano, Trinity follows with the same concept. AMD’s APU’s (Accelerated Processing Units) cannot be compared on an apple to apple level (no pun intended) with Intel. This is because IvyBridge CPU’s simply dominate AMD on the processing side, but AMD’s 7660D graphics dominate Intel’s integrated 4000 graphics. This is the conundrum for the consumer. At lower price points, consumers want the best bang for the buck. With the popularity of video encoding, movies and YouTube at an all time high, AMD is betting that their graphics will be the ultimate winner. Better still, AMD’s lower pricing will also attract consumers. Let’s take a look at the Llano to Trinity change.
Here is Llano. One area where AMD separate themselves from Intel with both Llano and Trinity is they dedicate more die space for the GPU. As we see, Llano uses four 32nm x86 “Stars” cores. This was not far off AMD’s retired Phenom II processors in CPU performance. The x86 cores each have 1MB of L3 cache.
With Trinity, AMD still dedicates a large amount of the die to the on-die GPU, but the cores change to AMD’s new Piledriver architecture, which comes with 2MB L2 cache (Llano had 1MB L2 cache). AMD also carries on their 32nm fabrication, and each module has two integer cores & a floating point. A four core Trinity APU has 2 modules (4 integer cores and 2 floating points per core).
Earlier this year, AMD released Bulldozer and they hoped it would put them back on the performance map. The release had mixed reviews and the performance did not live up to most expectations. However, it was brand-new architecture designed from the ground up, and it still included some remarkable innovations. AMD used x86 technology in a unique way by putting 2 cores on one module, and all the cores were connected through pipe-lines. Some speculate that Bulldozer did not live up to all the hype because AMD did not have the instruction sets all figured out.
AMD’s main focus with Piledriver was to improve the instruction sets and frequencies, creating better performance and a more efficient APU. By no means is anyone expecting Intel IvyBridge performance here, but increased performance over Bulldozer cores would certainly be welcome. Another change with Trinity is the GPU. The GPU cores are based on AMD’s Northern Islands and their 7650/7550D plus have up to 384 cores. Just as Llano had crossfire abilities, so does Trinity with the same discrete Radeon 6660 or 6670 for some serious graphic performance. Here we see Piledrivers improvements:
With AMD’s transition to the APU (Accellerated Processing Unit), Intel still uses what they call a CPU. Intel has also migrated to an on-die GPU and the immediate difference is the size each dedicates to the on-die GPU. AMD uses more die space while Intel’s is rather small. From an outside perspective, AMD’s total package looks much more balanced than Intel’s. We all know Intel dominates on the CPU side but AMD offers very good performance and much better graphics as mentioned earlier. In the mainstream arena, consumers look for the best value, and a more balanced package fits that bill in our opinion. This keeps the AMD train rolling, and we love to see the innovations moving forward.
With YouTube, movies and encoding are ever more popular… the graphics part of computing has become at least as important to regular computing. The UVD (Unified Video Decoder) that AMD uses is based on an ATI video processor. This is incorporated into the same die of the GPU for hardware decoding videos, along with the Advanced Video Processor (AVP). UVD handles decoding of video codecs entirely in hardware. Moving back in time, we saw UVD+ in the Radeon 3000 days, UVD 2 in the Radeon 4000 days and UVD 2.2 in RV770 and RV730 series of GPUs. Today, AMD uses UVD 3. UVD3 (Universal Video Decoder 3) is a new version for smoother and better video.
Now let’s move forward and take a look at the new A85X chipset and FM2 Socket.