AMD's Ryzen CPUs Aren't Running As Hot As They Say They Are

Image: Twitter (@amd)

Following weeks of questions, reviews, re-reviews and more questions from the community and the media, AMD has come out and cleared out a couple of issues. And one of those: despite what your BIOS and sensor monitors are saying, your Ryzen CPU isn't really running that hot.

In a post on their official blog, AMD explained that their 7 series CPUs have a sensor called "T Control". Otherwise called tCTL, it's the primary sensor on the Ryzen series of CPUs that reports how hot everything is.

The sensor, according to AMD, takes the temperature from the interface point between the CPU die and the heatspreader. But a problem users have been noticing - including yours truly, who has been grappling with a Ryzen 7 1800X of late - is that the tCTL sensor is offset on the 1700X and 1800X CPUs, which AMD says is necessary so "all AMD Ryzen processors have a consistent fan policy".

"Specifically, the AMD Ryzen™ 7 1700X and 1800X carry a +20°C offset between the tCTL° (reported) temperature and the actual Tj° temperature. In the short term, users of the AMD Ryzen™ 1700X and 1800X can simply subtract 20°C to determine the true junction temperature of their processor. No arithmetic is required for the Ryzen 7 1700. Long term, we expect temperature monitoring software to better understand our tCTL offsets to report the junction temperature automatically."

Image: AMD Community Blog

The Ryzen 7 1800X I've been using, which is running off a Gigabyte AX370-Gaming 5 motherboard, has been reporting idle temperatures of approximately 52-60o at idle, and higher when workloads increase. The high idle temperatures were of enough concern that I emailed AMD to ask whether the samples I'd been supplied with (which I carried by hand back from San Francisco) were faulty in some way.

For reference, I'd been monitoring the CPU temperature through a variety of applications: HWMonitor Pro; a beta version of the Ryzen Master overclocking tool, which AMD supplied; the temperature monitor in the AX370's BIOS, and HWiNFO. All four reported temperatures of 52o and beyond.

Still, it's nice to know that things aren't really that hot. AMD also noted that users should still use the High Performance power plan within Windows 10, and that they "do not presently believe there is an issue" with the Windows 10 scheduler that was impacting upon Ryzen's performance.

Ryzen And Windows 10: Does AMD Have A Performance Problem?

A new architecture from a processor vendor is always going to snag the attention of hardware lovers and their magnifying glasses. AMD's Zen, found in its freshly stamped Ryzen line is the current target and amongst all the numbers, testers have discovered something interesting; a discrepancy in the Ryzen's performance and Windows 10. But which party is to blame -- if any?

Read more

"Going forward, our analysis highlights that there are many applications that already make good use of the cores and threads in Ryzen, and there are other applications that can better utilise the topology and capabilities of our new CPU with some targeted optimisations," AMD said.

They added that Ryzen's simultaneous multi-threading should "generally see a neutral/positive benefit" from having SMT enabled, although they linked to benchmarks from Techspot that showed the Ryzen 1800X matching Intel's 5960X and 6600K CPUs once SMT was disabled. That's partially down to optimisations that will come down the road, and that same analysis showed that disabling SMT had no effect in certain games.

"We have already identified some simple changes that can improve a game’s understanding of the "Zen" core/cache topology, and we intend to provide a status update to the community when they are ready."


Comments

    So are they saying that the 1700X and 1800X CPUs need their fan to be running (or running at a faster speed) when at lower temperatures than the 1700? Any particular reason for that?

    Fudging the temperature sensor data seems like a weird way to achieve that.

      Yeah, it doesn't make any sense. Why would a 1700X at 30C need the fan to operate higher than a 1700 at 30C? Fans should work off actual temperature, I can't see any reason why the CPU would need to lie about it.

      Has anyone taken temperature readings with external sensors to confirm this?

        It'd be very hard to measure junction temperature externally as the junction and surface temperatures vary greatly.
        Hence why the sensor is usually directly on the junction.
        They give a fudge it buy 20 degrese but as no materials are completely thermally efficient this would be a more middle ground and as junction temperature rises, especially with over clocking the temperature difference will increase.
        As I noted in my other post it surely must be some sort of manufacturing compromise, there is no logical reason to do this.

          If the CPU is reporting 60C when Tjunction is 40C (as claimed) a correctly positioned sensor on the centre-mid surface between the CPU and heatsink will say which one is correct. If it reads over 35 then AMD's claim is off, and if it's over 40 then the claim is altogether bogus. Tcase shouldn't be more than 5-10C different to Tjunction.

          Remember the Tjunction reading from these CPUs is claimed to be reading higher than actual. Architectural constraints requiring the sensor be moved from the traditional spot would result in lower readings, not higher. Tjunction by definition is the hottest point on the silicon, you can't get a Tjunction reading higher than the actual highest value on the chip.

          (also, this edit-moderation problem is getting ridiculous)

          Last edited 15/03/17 6:32 pm

            Architectural constraints requiring the sensor be moved from the traditional spot would result in lower readings, not higher. Tjunction by definition is the hottest point on the silicon, you can't get a Tjunction reading higher than the actual highest value on the chip.

            Well aware of that, didn't think that was up for discussion, though you did seem to miss the point that, PDW is not linear to Tj. So there simple 20 degrees addition to the read temperature is only valid for one condition. As PDW increases Tj will start increasing more rapidly, to compound it even further, Tj to Tc isn't linear either.

            AMD could get around the problem using custom fan profiles as they have all the information needed, but just telling people to manually adjust there fans seems a bit of a weak effort. They could have even put that ability within the CPU itself to report correct calculated temperatures, but failed to. It really seems a bit half arsed.

              Even if the sensor moved elsewhere on the silicon, Tj could still be calculated using the an adjusted thermal resistance value. If we call this offset position 'o', then the equation would be Tj = To + (PdW * θjo). There would never be a need to add a constant value to the To reading to get to Tj, and θjo is no harder to calculate than θja, which is already required on the datasheet.

              I'm not sure I follow your logic, but if you're saying that AMD had to do this because of an offset sensor then I don't agree. Adding a constant to an offset temperature reading to 'approximate' the onset value is like an architect making a skyscraper and then deciding to make the elevator shafts out of pudding. There's no way AMD would be that incompetent.

    Seems kinda weird to me. They say for a consistent fan policy. How is it consistent when depending what cpu you have you have to change fan profiles to suit between X series and other ryzen cpu's.
    Was there not enough room on the die? Was there some other technological issue that they could place it where it should be.

    That has to be rubbish

    then that would make their own overclocking software wrong

Join the discussion!