AMD Launches “the fastest CPU in the world”

Your move, Intel...
30 August 2022

The ongoing CPU war between Intel and AMD stepped up a gear on August 30th, when AMD announced its new generation Zen 4 Core processor range.

Dr Lisa Su, Chair and CEO of AMD said the announcement of the Zen 4 Core range was the spearhead of a rapid period of innovation that will introduce four new architectures across more than a dozen products from AMD over the next handful of quarters.

Launching on September 27th, 2022, all eyes are initially on the Zen 4 Core CPU, (also known as the 7000 series) which comes in four iterations – the top-of-the-line 16-core, 32-thread Ryzen 9 7950X, the high-mid-range 12-core 7900X, the 8-core 7700X, and the 6-core 7600X, ranging in price from the entry-level 7600X at just under $300 up to the $700 7950X.

The Need For Speed

Dr Su left attendees at the launch with no illusions – the 7000 series was designed specifically to be the fastest CPU in the world, and to leave “the competition” (a politically neutral way of saying “Intel”) behind in the dust of its clock speed.

Crucially, the 7000 series will be the first 5nm CPUs available for standard desktop PCs. It will feature PCle 5.0, DDR5, and an all new AM5 platform – which will involve getting a new motherboard, but for both gamers and creators, it’s probably going to be worth it, because as Dr Su explained, “it allows creators to carry on creating, rather than waiting around for their images to render.”

The fundamental idea of developing this evolution of the Zen 3 CPU was to deliver “extreme performance with extreme power efficiency,” said Dr Su, before going on to throw out some impressive statistics.

“We estimated Zen 4 would deliver an 8-10% uplift in IPC, compared to Zen 3. As we optimized the product for release though, we now see a 13% uplift in IPC in desktop applications.”

In less intense times, creators might not whoop and holler at the idea of a 13% uplift in IPC, but these are not such laid-back times, and every percentage point of additional IPC will count with the core audience for the 7950X.

In terms of boosting that clock speed, there have been front end design innovations that push the frequency at the top of the stack to 5.7 GHz in the 7950X – which is a healthy, if not world-expanding 800 MHz than the 5950X ever managed. In a straight comparison, total single-thread performance is up by a more impressive 29% in the 7950X too, which is perhaps just as well as the price on the 5950X is dropping to a point where it becomes easy to choose – unless there’s something manifestly better to choose.

If you’re looking for extra reasons to equip yourself – and potentially, your developers – with the 7950X, it has 80 MB of L2+L3 cache and a new 170W TDP envelope. That means for creators, the combination of the Zen 4 core and the AM5 platform delivers on average 40% more compute than the 5950X. And if you’re comparing it to the Intel alternative, the 7950X delivers up to 62% more compute performance, as Dr Su demonstrated with a gaming benchmark software at the launch.

Never Mind The Speed Boost, Here’s The Energy Efficiency

What is perhaps a headline in and of itself though, the 7950X delivers that 62% performance bump on the Intel equivalent with 47% better energy efficiency.

Better performance, better speed, save the planet – it’s a bold three-pronged attack on the market by AMD, which has tended to dominate the CPU battle in the last few years. When Intel releases full details of its next-generation Raptor Lake CPUs, probably in late September, it may struggle to stand up to the challenge – though loading its CPUs with efficiency cores might help it appeal to segments of the market.

Dr Su also spent some time focusing on the ‘entry-level’ 7600X CPU, which will likely take the lion’s share of standard ‘domestic’ desktop sales. The 6-core 7000 series CPU was shown to deliver consistently higher frames-per-second rates (as well as exceptional smoothness), finishing a benchmarking test some 11% faster than the Core i9 12900K, which is – for all of a month, at least – the best and fastest in the Intel stable.

The pitting of the entry-level AMD against the most powerful Intel offering was perhaps a little disingenuous, since everyone’s aware that Intel will soon be hitting back with its own next generation of CPUs, but the point that the entry-level AMD can handle both speed and computational weight was more than succinctly made.

Neural Nets and Front End Redesigns

CTO and EVP, Technology and Engineering, Mark Papermaster joined the launch to explain the redesign of the Zen 3 CPU into the Zen 4 in more detail. He told the crowd in Austin, Texas, that the ethos behind the Zen 4 was largely to speed up the way the CPU received instructions, and to deliver some front end design improvements that would build towards that 13% overall IPC uplift.

There are extra tricks up the sleeve of the 7000 series, he added. Having long supported Advanced Vector Extensions in x86, the 7000 series adds support for AVX512, which delivers up to a 1.3X improvement of 32-bit precision floating-bit inferencing operations compared to the Zen 3. “That lets the Zen 4 target heavy-lifting workloads like molecular simulation, ray tracing, and physics simulators,” he said.

“We’ve also added support for convolutional neural nets, with the Vector Neural Net Instruction (VNNI). That gives us a huge speed up for things like image processing and Natural Language Processing (NLP).”

The August announcement also teased further releases into 2023, including the RDNA3 gaming graphics architecture, the Zen 4 C Core – optimized for cloud-native computing – and the next-generation XDNA architecture, which will be available next year.

There’s no disputing that the Zen 4 Core range of CPUs will generate interest at all levels, from standard desktop users at the 7600X level, through to developer and creator levels with all the benefits of the 7950X, including its enhanced speed, massively improved TDP and energy efficiency, and its upward step in terms of neural net instructions.

Can Intel deliver anything even close in its Raptor Lake CPUs?

We’ll tell you next month.