Cinebench R15 IPC Comparison Graphs

Instruction throughput is one of several methods to determine a microprocessor's performance. It is more commonly written as IPC, or instructions-per-clock. In order for a comparison of instruction throughput, all processors being compared must be operating at identical frequencies, with all turbo and low-power features disabled. A true instructions-per-clock comparison can't be made if these criteria aren't met.

The graphs below depict IPC improvements for architectures from both AMD and Intel over the past seven years.

Cinebench R15 — Out-of-the-Box Single-Threaded Performance

Intel's high-clocked Skylake-X processor leads the way with single-threaded performance. AMD's Zen architecture gains an incredible lead over its predecessors, albeit, we must remember that Cinebench is a floating-point benchmark; by design, AMD's construction core architectures contained half as many floating-point units as they did integer units, which is why they have always struggled in Cinebench. Happily however, Zen returns to a more balanced design, and that allows it to keep pace with Intel. AMD aimed to match Intel's Broadwell-based Core i7-6900K in single-thread, and that is exactly what we can see here.

Also worth noting is that both AMD Piledriver-based chips achieve identical scores, however the A-Series variant lacks a Level 3 cache and must therefore depend on its 200 MHz higher core frequency to obtain the FX's score.

Out-of-the-box single-threaded performance in Cinebench R15
Out-of-the-box single-threaded performance in Cinebench R15.

Cinebench R15 — Normalized to 3.00 GHz Single-Threaded Performance

In order to initiate an instruction throughput comparison, we must first normalize the processors to one fixed frequency. All models will comfortably operate at 3.00 GHz, and in the case that we might want to add older architectures to the graphs at a later date, the lower frequency also ensures that we don't need to start from scratch with what we have already. All architectures dating back to AMD's K8 and Intel's NetBurst are capable of reaching 3.00 GHz. Perfect.

At 3.00 GHz, the past three generations of Intel microprocessors suddenly find themselves very close to each other, while AMD's Zen architecture squeezes between Broadwell and Skylake. The latter two architectures both show that they're incremental updates at best.

Unsurprisingly, all AMD construction core architectures demonstrate poor performance here.

Normalized 3.00 GHz single-threaded performance in Cinebench R15
Normalized 3.00 GHz single-threaded performance in Cinebench R15.

The Final Instructions-per-Clock Uplift Results

Our final graph compares instruction throughput improvements, generation over generation, using AMD's Bulldozer architecture as our 100% baseline.

Note: If you're looking to compare IPC uplift between two specific architectures, you should divide rather than subtract. For example, the IPC difference between AMD's Zen and Intel's Skylake architectures is 5.664% (207.937 ÷ 196.825), and not 11.112% (207.937 − 196.825).

Generational instructions-per-clock uplift, relative to AMD Bulldozer
Generational instructions-per-clock uplift, relative to AMD Bulldozer.

'Coffee Lake' Update (October 7, 2017)

The graphs have now been updated to include the recently launched Intel Core i7-8700K. Furthermore, for the sake of obtaining more accurate data for results against Coffee Lake's predecessors, we decided to include all mainstream tier Intel Core i7 products dating back to Sandy Bridge in 2011.

The Core i7-8700K now dethrones the previous single-thread champion, the Core i7-7700K, predominantly with sheer core frequency improvements. An underclock to 3.00 GHz reveals very similar numbers to the chip it replaces; the only variable being more last-level cache for that single-threaded performance.

'Zen+' Ryzen 7 Update (April 20, 2018)

The newly launched AMD Ryzen 7 2700X has now been added to the graphs.

Paired with an X470 motherboard and system memory running at 2,933 MT/s, the second-generation Ryzen 7 processor shows an adequate clock-for-clock improvement of 4% over the initial Zen architecture from last year. However, the new iterations of AMD's Precision Boost and Extended Frequency Range technologies, which now apply to all threads of the chip, are sure to bring some much larger gains in multi-threaded workloads. Zen's unique architectural design means that you'll see even greater numbers still with faster system memory than what was tested, and this applies to both generations.

'Zen+' Ryzen Threadripper Update (August 14, 2018)

The graphs have been updated, following the release of AMD's second-generation Ryzen Threadripper processors.

Again, system memory is running at 2,933 MT/s. Out of the box, the new Threadripper models appear to have no problem matching Intel's Skylake and Kaby Lake offerings. With AMD Precision Boost Overdrive technology working its magic, the 16-core 2950X shows off some seriously strong single-threaded performance. Its 32-core sibling demonstrates equally impressive numbers, albeit a little lower due to the frequency deficit.

As with the second-generation AM4-based Ryzen processors, AMD Precision Boost 2.0 technology can actually apply to all cores if given the conditions to do so. That's some serious horsepower for a content creation machine.

'Zen 2' Ryzen 7 and Ryzen 9 Update (July 12, 2019)

AMD's brand-new third-generation Ryzen processors have now been released, which means it's time to update these graphs once again.

What happens when you take an already-strong floating-point unit design, and make it even stronger? Here's your answer. Even with DDR4-3200 system memory, it's a very strong showing for AMD. While Intel may still have the single-threaded performance advantage here, thanks to those very high frequencies, a combination of faster system memory and future Windows patches from Microsoft will certainly help to close that gap.

Zen 2 introduces a much stronger memory controller, now situated within the I/O chiplet of the package. AMD suggests to keep the memory controller in a 1:1 ratio between it and the Infinity Fabric; the underlying interconnect by which the processor as a whole communicates to other parts of the package. A 1:1 ratio is maintained with system memory up to DDR4-3733, before it falls back to the older 2:1 ratio of previous generations. For latency-sensitive workloads such as gaming, you'll definitely want to keep that ratio at 1:1, but other workloads may prefer the additional memory bandwidth offered by faster system memory.

On a final note, we conducted some game stream testing using OBS at 1080p (1920 × 1080) resolution and found that when games were forced to occupy threads only within the same core complex, minimum frame rates were considerably better for both processors, beating out even the Intel Core i9-9900K by almost 25% on average, and sitting comfortably above 60 frames per second on a more consistent basis. That may be important for those of you looking to simultaneously play and stream your games.