Zen 2 relative performance vs Skylake

[porina]

Ok, it's only been 6 weeks since Zen 2 came out, and I only just got around to doing half the testing. Here's the SMT/HT on comparison of Zen 2 (3600) vs a Skylake (6700k) when both are fixed at running 2 cores at 3.0 GHz (configured as 1+1 on the 3600). The idea of limiting the cores and clocks, beyond making them the same, is to limit the influence of other parts of the system such as ram. Although L3 size hopefully doesn't make much difference, it is possible had I used a 9900k and did same, it might perform slightly better in some situations. No HT/SMT scaling results for now.

 

3DPM I don't have much to say about. I only run this as it gives interesting (higher than anything else) scaling with HT/SMT. At best it is a highly niche compute scenario.

 

Cinebench needs no introduction. R15 sees more of an advantage than R20.

 

Y-cruncher calculates digits of Pi and is generally optimised to scale with available hardware, although for now it does not have specific Zen2 consideration so there is some potential for further improvement there. 25m is pretty small, but 1b runs to minutes, and will involve more significant ram access.

 

Prime95 isn't a just stress test, but is indicative of much software used in the search of prime numbers. Two scenarios were used with the built in benchmark running one task per core. 64k FFT is small and fits well within the CPU's cache. Note in absolute terms, SMT doesn't give more performance than without, but there is less loss on Zen 2 than Skylake. 2048k FFT is large, and Intel gets a slight advantage here when HT is on. I suspect this is related to the memory architecture. The Zen2 L3 cache is nearly sufficient to hold a single task, but not when running two with HT/SMT. Ram bandwidth then starts to impact more, but Zen 2 has half bandwidth writes from each CCD and this is probably starting to choke. Intel CPUs offer full bandwidth each way.

 

Aida64 has a bunch of synthetic benchmarks. Photoworxx in particular seems to be ram sensitive so that may in part be why it is relatively weak. AES has always been a strong area even with past Zen generations. They seem to have replaced Hash with AES3, and removed VP8. Hash was another one running particularly well on Zen, but AES3 swings towards Intel. VP8 I found to be highly variable between runs so I don't miss that at all.

 

All considered, Zen 2 has generally better IPC all round compared to SKylake (and by implication, other -lakes until Sunny Cove e.g. Ice Lake). For many things CPU throughput related, assuming it can sale, this would seem the optimal choice until we see a desktop Sunny Cove part from Intel.

 

Test setup

Common:

Windows 10 64-bit 1903 with August 2019 updates
3DPM v2.1

Aida64 6.00.5100

Prime95 29.8b5

Y-cruncher 0.7.7 9501

 

AMD system:
Ryzen 3600
Noctua D9L
Asrock B450 gaming-itx/ac, bios 3.50
Kingston HyperX Predator RGB 4000 2x8GB @ XMP3600 settings

 

Intel system:
i7-6700k
Deepcool Lucifer
Asus Maximus IX Apex, bios 1301
Kingston HyperX Predator RGB 4000 2x8GB @ XMP3600 settings

 

 

Similar testing was performed in the past comparing Zen, Zen+, Skylake and Skylake-X. Due to different software versions used, those results are not directly comparable to this newer testing.

 

 

[NunoLava1998]

Zen 2 has much inferior IPC compared to Skylake. Just look at SHA3! /s

[Stormseeker9]

It was always intel intel intel...

 

zen 3 has really shown a lot about what amd has to offer currently. Zen3 will be a very interesting cpu, intel is loosing their market share slowly I’d say. Much better price/performance at amd 

[porina]

Just now, VegetableStu said:

nothing against the OP, but could anyone explain a theory (I'm choosing to use this word for now just because I'm not qualified to... qualify it) that locking clock frequency below its intended peak would affect IPC measurements and comparisons? o_o

It entirely depends on your what you want to get out of the results. The goal of my testing is to see what the cores do as far as is practical. I choose to do so by a combination of lower core count (more effective cache per core) and lower clock (less ram bandwidth limiting). I aim to find what I call "peak IPC". What is the best the cores can do if not limited by other factors.

 

If you want to argue "real world" testing, look elsewhere.

[porina]

8 minutes ago, Stormseeker9 said:

zen 3 has really shown a lot about what amd has to offer currently. Zen3 will be a very interesting cpu, intel is loosing their market share slowly I’d say. Much better price/performance at amd 

Call it Ryzen 3000 (although not strictly correct as there are 3000 APU parts based on Zen+), or call it Zen 2. It is not Zen 3, which will be the next generation to come.

6 minutes ago, porina said:

I aim to find what I call "peak IPC". What is the best the cores can do if not limited by other factors.

To expand on that, this is in part why I titled the thread Zen 2 vs Skylake, not 3600 vs 6700k. The CPU model isn't terribly important. I'm more interested in the architecture as a whole. Limited scenario IPC is not really of much interest to me as it muddles too many factors. This is in part why I wouldn't use game benchmarks in this, as the GPU contribution significantly reduces the role of CPU.

[Stormseeker9]

1 minute ago, porina said:

Call it Ryzen 3000 (although not strictly correct as there are 3000 APU parts based on Zen+), or call it Zen 2. It is not Zen 3, which will be the next generation to come.

To expand on that, this is in part why I titled the thread Zen 2 vs Skylake, not 3600 vs 6700k. The CPU model isn't terribly important. I'm more interested in the architecture as a whole. Limited scenario IPC is not really of much interest to me as it muddles too many factors. This is in part why I wouldn't use game benchmarks in this, as the GPU contribution significantly reduces the role of CPU.

Typo was meant to be zen2 in the beginning. 

[porina]

17 minutes ago, VegetableStu said:

oh okay, understood. I'm just curious about the technicalities of that old tweet ._.

Just realised my earlier choice of words may sound overly harsh. It was a fair question, just "real world" CPU tests I find harder to digest as there are many more variables confusing the results if all you're interested in are the CPU cores.

[porina]

1 minute ago, VegetableStu said:

nah it's okay, no harm taken. Also puts in perspective how does a reviewer observes IPC comparisons in their tests

Unless it is a deep dive of some sort, I think mainstream reviewers tend to compare models against competitive models, as that is the real world choice "normal" buyers have to make. I find by knowing the peak under certain conditions, you can better understand if the CPU is the limit, or ram or other factor. It makes it easier to extrapolate to new core configurations that may not exist yet. Actually, in the current market, it is harder to reach peak core loading as caches, bandwidth, ram, are generally growing slower than the core potential.