When the first Zen Engineering Samples leaked out, many were surprised at the relatively low clock speeds. At stock, the 8 core 16 thread chip managed only 2.8 GHz stock and 3.2 GHz boost. Given the engineering sample nature of the chip, it was reasonable for the lower than usual speeds. Now it seems that AMD has released some new variants of their Zen ES CPUs, this time clocked a bit higher.
At the same A0 and 95W TDP, AMD has improved the base clock to 3.15 GHz, all core boost is up to 3.3 GHz, and single thread boost is all the way up at 3.6 GHz. These speeds are quite respectable given that Broadwell-E i7 6900K peaks out at 3.7 GHz and reports are suggesting a similar IPC. This means AMD may have a real winner on their hands though Intel is moving onto Skylake-E soon.
In addition to the new 8 core information, we’re also getting some more information about the 4 core 8 thread variant engineering sample. Keeping the 65W TDP, the chip manages 2900 GHz with all core turbo at 3.1 GHz and max turbo at 3.4 GHz. While respectable for the mid-range, we will have to see how it fares in overclocking before it is a worthy competitor to Intel’s K series. I suspect we will see higher speeds for launch.
Having said all that, it’s important to note that this is an unconfirmed report at this time. The source is somewhat trustworthy since they were the ones to provide the initial ES specifications that were later confirmed. The source is also reporting that AM4 motherboards have finally hit production so we’ll hopefully see Bristol Ridge soon and Zen shortly after. The expected launch date in February 2017 but I suspect it will be a bit longer before we get real hardware. It will be interesting to see how Kaby Lake and Zen battle it out next year.
Depending on the application, high clock rate isn’t always that important. This article seems to be giving it greater importance than it often has in practice…
Many high-performance computing applications can be fairly insensitive to clock rate, as they are often memory bound or I/O bound.
This doesn’t mean high clock rate isn’t useful in other scenarios. But the high memory bandwidth of the AMD Zen is going to help a lot.
The back story is that there was a demo showing AMD zen was competitive with last gen Intel even though Zen was running at a lower clock. So this higher clock is supposed to be exciting as AMD can finally compete with current gen i7.
That makes sense. It may add a few percentage points in performance. And for some workloads it is important. Games often do better with boosted freq.
Yes, especially when looking at boost clock speed in regards to single core performance, which in the past AMD had been severely lacking in comparison to Intel…
2900gHz O_O heeeelllll yeeeaah!
Dang 2900GHZ. Wouldn’t need to upgrade my computer for a really long time with those fairy tale numbers
The 8 core 3.2GHz with a 40% improvement in IPC is about the same speed as my FX-8320@4.5GHz in single core performance. But of course it has “SMT” to double the threads it can handle.
I’m going to wait for overclocked benchmarks on both AMD’s ZEN and Intel’s Kaby Lake to finally upgrade my system that I’ve had for 4 years now. Sure… It’s still a super solid system and capable, but… I want! Gimme!
BD/PD doesn’t scale nearly that good, there is a lot more to it than just clocking the cores, eg: branch misspredicts will still take it’s time (huge pipeline) and cache lookups will be no faster.
First, look at a few benchmark, clocking the 8320 to ±4.5GHz will give you roughly 15% performance increase (on a sunny day). So, multiplying the clock by a ratio of 1.28 (28%) gives half that increase in performance (note that a lot of applications doesn’t even come close to 15% performance increase). Then we can add the clock diff between Zen and FX-8320 which gives us another 10%. So overall that gives you nearly 25% in _some applications_, overall we’d likely be talking about notably less.
On another note, BD/PD modules does not make up 2 cores, in fact there is only one FPU per module. So if whatever you do mainly relies on floating point computation you basically have a quad-core. Even the integer units (two per module) doesn’t scale well, this is obviously due to the numerous shared resources (scheduling, higher level cache, pipeline and more). So you will find that it quite closely resembles SMT/HT scaling (which CMT is meant to be an alternative to).
So, hypothetically speaking, say you managed to get an FX-81xx comparable (performance wise) to Summit Ridge, it would only be comparable to a quad-core.