Conclusion

For as much hype and excitement that has been generated around Zen and the Ryzen products so far, with everything focused on the high-end when we hit the lower elements of the stack and the volume parts, not much ‘excitement’ is to be had. We’ve already gone through the new fancy microarchitecture and the platform, and what matters at this end of the spectrum is a pure performance per dollar metric. So far the Ryzen 7 parts have certainly hit that goal, especially when originally compared to Broadwell-E when the Ryzen 7 parts per launched. For the Ryzen 3, the direct competition is Kaby Lake, and CPUs with a much higher IPC. But where Intel has two cores, AMD has four.

Diving straight into the graphs, the Single Thread performance graph is as follows. All data is shown relative to the performance of the Ryzen 3 1300X.

For the AMD CPUs, the 1300X and 1500X are near in base/turbo frequency (3400/3700 vs 3500/3700) and so score about the same, and the Ryzen 3 1200 at 3100/3400 scores about 13% lower.

The Intel CPUs here show a distinct curve from the Pentium G4560 at 3.5 GHz through the Core i3s at 3.9 GHz, 4.0 GHz and 4.1 GHz. The Core i5 7400 scores a lot lower here, with its base frequency of 3.0 GHz and a turbo up to 3.5 GHz.

For the multi-thread performance:

We shopped the graph here at +20% otherwise it would look odd, but the Ryzen 5 1500X with simultaneous multithreading gets a +40% boost over the 1300X, while the Ryzen 3 1200 sits again at around -12%. All the dual core intel parts lag behind here compared to AMD’s quad cores, although the Ryzen 3 1200 and the Core i3-7300 are closely matched – but the Ryzen wins on price by being ~$40 cheaper.  The Core i5-7400, competes against the Ryzen 3 1300X here as they are both quad cores, and the Intel wins despite the lower frequency due to higher IPC – which comes at a $50+ premium.

For our combined all-in-one graph, we included our mixed workload data and weighted the results 40:50:10 for single:multi:mixed thread workloads.

If we ignore the Ryzen 5 1500X in the top right corner, there are a few stories here.

First is that the Ryzen 3 1200 does not look like an attractive option. It performs +2-3% of the Pentium but is $30 more expensive, and the Core i3-7100 beats it by 8% for only a sub-$10 cost.

Then there is the Ryzen 3 1300X. Compared to the Core i3-7300/7320 and the Core i5-7400, it clearly wins on performance per dollar all around. Compared to the Core i3-7100 though, it offers almost 5% more performance for around $10-15 more, which is just under 10% of the cost. Depending on budgets, each one could be an attractive option.

We’re still working through our gaming testing as this review goes live, and we’ll add graphs for that in a bit.

Power Consumption
Comments Locked

140 Comments

View All Comments

  • ampmam - Thursday, July 27, 2017 - link

    Great review but biased conclusion.
  • tvdang7 - Thursday, July 27, 2017 - link

    No overclock?
  • Oxford Guy - Thursday, July 27, 2017 - link

    No, just a RAM underclock.
  • zodiacfml - Thursday, July 27, 2017 - link

    overclocking tests on the ryzen 3 1200 please. the only weakness of the chip is for non-gaming or htpc usage as it will require purchasing a discrete graphics card. otherwise, it presents good value for most things like gaming and multi-threaded applications, add overclocking, and it gets even better.
  • kaesden - Thursday, July 27, 2017 - link

    one thing to not overlook with the ryzen 1300x is the platform. Its competitive with budget intel offerings and can take a drop in 8 core 16 thread upgrade with no other changes except maybe a better cooling solution, Something intel can't match. Intel has the same "strategy" at their high end with the new X299 platform, but they seem to have lost focus of the big picture. The HEDT platform is too expensive to fit this type of scenario. Anyone who's shelling out the cash for a HEDT system isn't the type of budget user who is going to go for the 7740x. they're just going to get a higher end cpu from the start if they can afford it at all, not to mention the confusion about what features work with what cpu's and what doesn't, etc...

    TLDR; AMD has a winner of a platform here that will only get better as time goes on.
  • peevee - Thursday, July 27, 2017 - link

    From the tests, looks like Razen 3 does not make much sense. Zen arch provides quite a boost from SMT in practically all applications where performance actually matters (which are all multithreaded for years now), and AMD artificially disabled this feature for that stupid Intel-like market segmentation.

    Also I am sure there are not that many CPUs where exactly 2 out of 4 cores on each CCX is broken. So in effect, in cases like one CCX has 4 good cores and another has only 2 they kill 2 good cores, kill half of L3, kill hyperthreading...

    It would be better to create a separate 1-CCX chip for the line, which would have much higher (more that twice per wafer) yield being half the size, and release 2, 3 and 4 core CPUs as Ryzen 2, 3 and 4 accordingly. With hyperthreading and everything. I am sure it does not cost "tens of millions of dollars" to create a new mask as even completely custom chips cost less, let alone that simple derivative.
  • Oxford Guy - Thursday, July 27, 2017 - link

    "It would be better to create a separate 1-CCX chip for the line"

    Or, it could be explained by this article why AMD can't release a Zen chip with 1 CCX enabled and one disabled. Instead, we just get "obviously".
  • silverblue - Friday, July 28, 2017 - link

    He did explain it. Page 1.
  • Oxford Guy - Saturday, July 29, 2017 - link

    Where?

    All I see is this: "Number 3 leads to a lop-sided silicon die, and obviously wasn’t chosen."

    That is not an explanation.
  • peevee - Tuesday, August 1, 2017 - link

    That is still be half the yield per wafer compared to a dedicated 1-CCX line. Twice the cost. Cost matters.
    And the 3rd chip must be 1CCX+1GPU. SMT must be on everywhere though, it is too good to artificially lower value of your product by disabling it by segmentation.

Log in

Don't have an account? Sign up now