Intel Corp (INTC)

[This Causes an Error]

Is Kaby Lake ahead of Ryzen based on what you have seen so far?

It depends on your workload. VideoCardz recently found some leaked Ryzen benchmarks in the 3D Mark database:





3D Mark physics can scale to many cores (it's a physics engine), so you see the top Ryzen tested (8 core/16 thread) beating out the 7700K (Kaby Lake-S, quad core/eight thread).

But observe the per core score numbers and something else becomes obvious -- Kaby Lake has much higher performance per core through a combination of higher performance/clock and faster clock speed.

Indeed, per core, the 7700K @ 4.5GHz scores 3552 points in this test, implying ~789 points/core/GHz. The Ryzen supposedly at 4GHz gets 2531 points per core, or 633 points/core/GHz.

Per core/per clock, based on this test, Kaby Lake is a hair under 25% faster per clock than Zen, core-for-core, clock-for-clock.

Does this pass a basic sanity check? AMD claims that Zen delivers 40% more performance/clock than its prior generation Excavator architecture. Here's a Geekbench 4 result for Excavator @ 4.2GHz (single core turbo):

https://browser.primatelabs.com/v4/cpu/1798979

Single core Integer: 2746
Single core Floating point: 2467.

On my Kaby Lake 7700K running at out of the box speeds (4.5GHz single core turbo), I get an integer score of 5317 and a floating point score of 5352.

So if you take my score, and normalize to 4.2GHz, you get: 4962 for integer and 4995 for floating point.

Now take the Excavator scores I showed you and just multiply by 1.4X, the 40% perf/clock improvement that AMD said it would deliver. What do you get for the single core int and floating point? 3844 and 3454, respectively.

Using those calculations, Kabylake should be ahead about 29% per core/per clock in integer and 44.6% in floating point. Now Zen does not have all the AVX2 machinery that Skylake does so in the GB4 subtests that use AVX2 will be much faster than they run on Zen (it's the GEMM and SFFT workloads, and Kabylake has a >2x performance advantage in those over Excavator, as expected) -- skewing the floating point average in Intel's favor.

I'm not going to do the recalculation, but I think if you stripped those subtests out you'd get something closer to the ~29% advantage that Kabylake seems to have in integer per core/per clock.

But, anyway, the leaked figures are close enough to the "theoretical" calculations that they pass the sanity test.

Bottom line, Kaby Lake-S will have a performance/clock and clock speed advantage per core, but with these Ryzen chips, AMD is throwing more cores at the problem. Intel also has a separate line of high-end desktop processors with more cores (the high-end desktop chips have between 6 and 10 cores currently), and that product line is expected to be refreshed in August.