Smarter Than Your Average Chip
Explaining further, Papermaster said 'we have implemented a self-training algorithm (machine learning) that assigns a set of weights based on branch outcome, and we have also increased our branch history table. This improves the accuracy of Ryzen.' In short, the chip pre-loads instructions based on recent behaviour and optimises their flow through the processor. We shan't know how well it works until a sample is in the labs, of course.
The back-end of the pipe, feeding the data, is Smart Prefetch. 'We have also invested in the algorithms and made key improvements in how we identify data streams, strides and regionality of data, so improving prefetch accuracy.' Put simply, the combination of smart prefetch and much-improved branch prediction should mean the fewest stalls and misses - the bane of any modern processor - for any AMD chip to date.
High Hopes For Performance And Power Efficiency
Talking up the virtues and engineering prowess behind Ryzen is well and good, you might say, but how does it compare against the beastlier Intel chips? AMD has already promised a 40 per cent improvement in clock cycle throughput at the same energy level as the best AMD processor of today, leading many to think of Ryzen as a genuine competitor to the Intel performance hegemony.
John Taylor sought to put some meat on the performance bones by walking through a video comparison of an eight-core, 16-thread Ryzen processor - the best of the Summit Ridge line, we imagine - up against an Intel Core i7-6900K. Hamstringing the AMD chip somewhat, it was running without any form of Precision Boost, meaning all cores and threads were operating at 3.4GHz. The Intel processor, meanwhile, ran at stock speeds, so anywhere between 3.2GHz and 3.7GHz, depending upon load.
Keen to stress that AMD really does have more in the tank due to optimisations that will be imbued before launch in Q1 2017, Taylor showed the Ryzen processor beating the also-8C/16T Intel chip in the taxing, multi-core HandBrake video-encoding tests by a few seconds and matching it in the stressful Blender benchmark.
Let's pause on that for a second. AMD is saying that it can comfortably match the performance of an $1,100 Intel chip with a non-optimised Ryzen, with the suggestion that it will beat it before launch, once Precision Boost comes online. And stressing the relative energy consumption, Lisa Su, AMD CEO, came on stage to show that Ryzen is not only putting up a fair fight against Intel's second-best consumer CPU, but also consuming a little less platform-wide power when doing so. 'We have the performance and the power,' she said.
Such a demonstration fills us with much-needed confidence that AMD really is on track to once again challenge Intel in the performance desktop space. Ryzen appears to have almost identical IPC-per-core as Intel's Broadwell architecture, and knowing that upcoming Kaby Lake doesn't improve upon performance in any meaningful way, the immediate future looks bright for AMD.
We're obviously keen to see how Ryzen benchmarks in our labs and, perhaps more importantly, just how well it scales frequency in the more cost-conscious, large-volume 4C/8T environment, one that is dominated by Intel at present. Interesting times.
Rounding off the supporting cast, we know that a new platform, AM4, will be home to Ryzen. On first glance, it features up-to-date technology found on competing Intel chipsets and, together with the chip, should offer the well-heeled enthusiast and entry-level workstation user serious food for thought on their next power purchase.
We have a much better idea on performance, a good grip on availability (within four months), but no inkling on price. Just how aggressive AMD will be remains to be seen, but we do hope that it takes Intel's seemingly exorbitant pricing to task. Anyone contemplating an AMD Ryzen/Vega combo early next year should be in for a treat.
