Power & Temperature Testing
We again used the NZXT Kraken X62 at maximum fan speed to cool the CPU as much as possible--the 1700X has a TDP of 95W, higher than the 1700’s 65W but identical to the 1800X. The stock 1700X reached a peak of 76C under load, causing mild throttling from the usual 3.5GHz as the 75C limit was exceeded. We can see this in clock fluctuations at the higher temperatures (~76C), where the 1700X occasionally dips to control thermals. The OC 1700, meanwhile, never exceeded 71C and held steady at 3.9GHz--an odd result, but there’s a reason for this. AMD’s recent statement (and our original research) shows that the company intentionally offsets temperatures with a “tCTL” value, just like the FX series before. What’s new in the statement, and what wasn’t told to GN during our original detailings of tCTL’s continued existence, is the value with which AMD offsets temperature. The 1700 accurately reports temperatures, while the 1700X and 1800X report temperatures that are 20C higher in order to “maintain a consistent fan profile” for all three chips. The end result is this: we managed to stress the 1700X to the point of throttling, but not the 1700 when running at the same frequency.
HW Monitor’s TMPIN0 value (on the ASUS board) appears to be the proper Junction temperature, as it aligns with ASUS’ AI Suite. AI Suite, as a recap, takes the TSI temperature bus value and combines it with a thermistor on the CH6, then crunches the two through an algorithm to achieve a more realistic temperature. We are still not positive which numbers are actually Junction temperature and which are tCTL Junction temperatures, though. Given a TjMax of 75C, we are assuming that HW Monitor and AI Suite are presenting the true Junction temperature, given the throttle performance (thermal trip-point) at >75C in these tools.
Regardless, we are clear that our 1700 chip is running cooler than the 1700X when the 1700 is overclocked. The 1700 ($300) is also not throttling under the same conditions as the 1700X when tested, and generally seems to be advantaged thermally by way of its lower possible vCore (but same performance). This keeps the R7 1700 in its championing position for the R7 lineup.
Update: We have some more news on thermals, thanks to follow-ups from AMD and some testing. We'll follow-up with a piece dedicated strictly to temperatures on Ryzen.
1700X Power & Voltage
| Stock + balanced | Blender (multi-thread) | Cinebench (single-thread) | POV-Ray (single-thread) | Idle |
| Watts | 168 | 89 | 91 | 61 |
| Core Frequency MHz | 3499.18 | 3900 | 3900 | |
| Core Voltage | 1.155 | 1.373-1.41 | 1.417 |
AMD R7 1700 Auto Voltage Behavior
| 1700 Auto vCore 2666MHz DRAM | Blender (nT) | Cinebench (1T) | POV-Ray (1T) | Idle |
| Total System Draw (Watts) | 133 | 80 | 81 | 58 |
| Core Frequency (GHz) | 3.2 | 3.2-3.75 | 3.75 | |
| Core Voltage (v) | 1.068 | 1.286-1.308 | 1.286 |
AMD R7 1800X Power + Voltage
| 1800X Auto vCore, 2933MHz RAM | Blender (nT) | Cinebench (1T) | POV-Ray (1T) | Idle |
| Total System Draw (Watts) | 185 | 92 | 94 | 60 |
| Core Frequency (GHz) | 3.7 | 3.7 - 4.1 | 4.1 | |
| Core Voltage (v) | 1.221 - 1.243 | 1.395 - 1.461 | 1.417 |
As stated in the methodology section, our power testing is currently a weak link in our test course, but it still permits relative comparisons between same-arch products. The system under load, using identical hardware and settings except for the CPU, drew 133W with the R7 1700, 168W with the R7 1700X, and 185W with the R7 1800X. All three consumed about 60W at idle.
Energy dissipated as heat correlates directly with power consumption, which is of particular interest with the R7 CPUs, given that the 1700 runs so much “cooler” than the X-branded chips. Under full load (during the thermal tests), our stock 1700 reported 1.1v vcore and 46C tJunction, 1700X 1.2v and 75C, 1800X 1.2v and 75C, and our maximum 4.0GHz overclock on the 1700 reported 1.4v and 71C. These reported temperatures, although they should be accurate, seem to align with the offset tCTL temperatures that AMD has described. Moreover, the CPUs appear to be throttling based on this number. We’re reaching out to AMD for further clarification over how R7 chips decide to throttle and we are considering furthering our own testing with some in-house tools.
Our review of the Wraith Stealth cooler is coming soon. Until then, it’s best to assume that a stock cooler won’t be sufficient for overclocking, and that a 3.9 - 4.0GHz R7 1700 will need something beefier. If the Wraith Spire is actually up to the job of cooling the stock 1700X (early review samples shipped without stock coolers), that could increase its value by a few dollars and make up for some of the $70 gap between itself and the 1700, but it needs to be pretty impressive.
Continue to Page 3 for the Premiere, Blender, & synthetic tests.
