Snapdragon 765G vs Dimensity 1000L

The positioning of the Snapdragon 765 Reno3 Pro 5G versus the Dimensity 1000L Reno3 5G is an extremely interesting one. By all means, the regular Reno3 is supposed to be the lower-end device and in fact does cost less than the Pro variant of the phone. Logically, we would expect OPPO here to choose the lower-end SoC for the Reno3 whilst the Reno3 Pro gets the better performing chipset. Although that’s what everybody would assume at first glance, things aren’t quite as straightforward.

The although the two devices and SoCs are seemingly positioned close to each other, they’re very different in terms of CPU capabilities. There are two factors at play here: first of all, the MediaTek chipset outright has a core-count advantage over the Qualcomm design by simply having four big cores instead of only two. That’s already a major difference and usually the kind of differentiation you’d expect from SoC at different class categories.

Secondly, MediaTek’s design uses newer Cortex-A77 cores whilst the Snapdragon uses last year’s Cortex-A76 IP. The Cortex-A77 is generally 25% better clock-for-clock compared to its predecessor, which seemingly compounds the processing power ability difference between the two SoCs and Reno3 phones. Qualcomm does have an advantage in terms of clocks speeds which does slightly make up some of this difference: at 2.4GHz on one of the Cortex-A76 cores it’s 200MHz ahead of the MediaTek chipset, which should reduce the expected performance gap a little bit.

Another big difference between the two SoCs is the fact that MediaTek uses TSMC’s 7nm node, while the Snapdragon 765 is manufactured on Samsung’s 7LPP node, with the only other public silicon out there manufactured on this process technology being Samsung’s own Exynos chipsets.

Comparing single-threaded performance between the two SoCs, we see the expected performance gap between the two designs materialise in the measured results. In SPEC2006, the Dimensity 1000L outperforms the S765 by respectively 13% and 22% in the integer and floating-point suites.

Furthermore, the Dimensity-powered Reno3 5G is able to do this whilst consuming less power and energy than the Qualcomm chipset. Generally, we expect this difference to be process-node related as we highly suspect Samsung’s 7nm node being less power efficient than TSMC’s process technology.

GeekBench 5

Whilst we don’t usually run SPEC in rate mode (multi-process test), GeekBench5 very much showcases the multi-threaded performance advantage that the MediaTek-powered Reno3 5G has over the Reno3 Pro 5G: With a score of 2702, it’s 50% faster than the 1802 score of the Snapdragon 765.

The one area where I did note that the Snapdragon chipset outperforms the Dimensity counterpart is memory latency. On the part of the CPUs, the Qualcomm chipset seemingly has a lower latency memory subsystem that allows it roughly a 13% latency advantage in full random DRAM memory accesses – or around 22ns at a test depth of 128MB.

The memory characteristics between the two SoCs are quite different, as the MediaTek chipset does showcase stronger CPU-side memory bandwidth, particularly in memory reads that reach up to 19GB/s versus 13.3GB/s for the Qualcomm chipset, with a similar advantage in memory copies, although the Snapdragon seems to have faster accesses and more bandwidth for memory writes.

All in all, until now it seems that the MediaTek powered chipset has a clear advantage in terms of performance and power efficiency, which actually makes sense as the Dimensity 1000L on paper just is a higher specced piece of silicon.

3x Reno3's - All Different System Performance
Comments Locked

44 Comments

View All Comments

  • dotjaz - Wednesday, August 12, 2020 - link

    For years they only had interim licensing and they didn't have a license before starting the process, I'd call that stealing.
  • londedoganet - Wednesday, August 12, 2020 - link

    > For years they only had interim licensing

    Which years? From 2010 to 2014, Apple used stock PowerVR GPUs, and then paid license fees to Imagination from 2014-2017, stopped in 2017 and started again in January 2020 (https://www.theregister.com/2020/01/02/imagination...
  • deanylev - Monday, August 10, 2020 - link

    HiSilicon is spelled as "HiSilcion" on the last page.
  • LiviuTM - Monday, August 10, 2020 - link

    I'm no expert by any means, but reviews of mobile SoCs focus only on processing power and efficiency, which lead to one major issue: says absolute nothing about mobile connection quality. I suspect Qualcomm stack (modem/modem firmware, RF Front End, antennas and whatever else there is) is vastly superior.
    This is not very often talked of (if any at all), but it had a significant impact onreal usage.
    Remember Apple not just buried the hatchet with, but also signed a multiyear deal with Qualcomm for chipsets (mainly modems, I presume).
  • s.yu - Wednesday, August 12, 2020 - link

    I believe if you've read enough you'd notice that there's massive discrepancies of connection quality between QC powered models, so at least the front end and antennas should depend primarily on manufacturer implementation.
    OTOH Huawei diehards claim Huawei has superior connection quality, which I don't dismiss but instead attribute to Huawei's telecoms hardware discriminating against non-Huawei handsets, explaining how these claims only arise from regions where Huawei's telecoms branch comes close to a monopoly, like China, south Asia, the Middle East etc.
  • shabby - Monday, August 10, 2020 - link

    Why is there performance so poor with 4 a77 cores?
  • HyperText - Monday, August 10, 2020 - link

    Frequency...
  • dotjaz - Wednesday, August 12, 2020 - link

    Nope, if you looked at GB5 and SPEC06, the raw performance is not bad at all, far better than 765G, but the actual performance is lagging behind Reno3 Pro 5G especially with responsiveness/burst load.
  • [ECHO] - Monday, August 10, 2020 - link

    Fascinating stuff Andrei, thanks for the write up!
  • [ECHO] - Monday, August 10, 2020 - link

    It's interesting to see the Dimensity 800/820 actually show up in more phones (couple of Honor and Huawei's, at least one Xiaomi/Redmi) than the 1000/1000L/1000+ so far. Almost makes you wonder if the original wafers they ordered for the 1000's had a pretty bad defect, didn't get caught until far too long after tape out and then they were stuck salvaging some low percentage of good dies for 1000L's and the 1000+ is a pretty deep silicon respin, hopefully with more wafer starts...

Log in

Don't have an account? Sign up now