AMD Vega: Frontier Edition Power & Temperature – Standalone
This is where the bulk of our testing was invested: Thermals, power, and noise. We’re looking at AMD Vega FE ($1000) primarily versus the Titan Xp ($1200), but also against some AIB 1080 Ti cards.
For power versus thermals, we’re starting with just Vega, then we’ll add the NVIDIA Titan Xp for comparison. Under our thermal benchmark and measured at the wall, the Vega FE stock configuration draws an average of 384W load, with the overclocked variant drawing 552W average from the wall. Our peak power consumption is 442W during the stock test and 569W during the overclocked test. Thermally, we’re averaging 79.9C with the stock configuration and 75.1C with the overclocked configuration, which is lower because the fan is manually controlled to 3700RPM.
Adding the Titan Xp stock card to the test, we see an average power consumption of about 357W – so that’s 27W lower than the Vega: FE system, and a maximum power consumption of 370.5W, or about 72W lower than the Vega: FE config’s max power consumption. The Titan Xp runs lower power consumption, but also higher thermals. This has been a criticism of ours since we first reviewed the Titan Xp, and is resultant of the blower design.
One item of note is that the Vega: FE card has much more sporadic clock behavior, as the clock rises and falls frequently throughout testing. We see this reflected in the thermal and power lines, while the Titan Xp remains more stable. We’re not yet sure the connotation of this behavior and aren’t sure if the root-cause is related to AMD’s execution of the software or to the architecture. Regardless, we will simply call it an “observed behavior” for now.
AMD Vega: FE Power vs. Thermals During SPECviewperf
This chart shows the power draw and temperature trend as plotted against a 22-minute test run of SPECviewperf 12, which performs trace-based testing of 3DS Max, Maya, and other production applications. The spikiness in the chart is from tests starting and ending. With Vega: FE, total system power consumption sits reliably around the 425W mark, with a few dips down to 340-380W in the middle. By-and-large, though, we’re at around 425W.
Adding the Titan Xp to the chart reveals its total system power consumption at around 340W for the initial quarter of testing and around 396-404W for the second half of the tests. Thermal performance for the devices is largely mirrored in this specific test pass, as neither is permitted enough time to heat up to the thermal throttle points.
Power vs. Gaming
For other power draw tests, our Vega: FE Air system draws 381W in a FireStrike workload, which is a delta of 40W more than the stock Titan Xp and a 1080 Ti SC2 stock card.
We also see a power draw of 401W in Ghost Recon, which is about 31W more than the Titan Xp and 1080 Ti. In For Honor, we’re at 393W, or about 37W more than the Titan Xp.
40dBA Noise Normalized Thermals
Testing thermal performance comparatively next, with a steady state chart plotting noise-normalized performance at 40dBA, we see AMD’s Vega: FE card operating at 59.9dBA. This ranks it alongside the worst GTX 1080 Ti partner card we’ve tested – the Armor – while still retaining a lead over the reference [Titan Xp card]. These numbers are Delta T over ambient, so if we account for ambient, the Titan Xp was operating around 96C and clock-throttling as a result. To be fair, AMD was also clock throttling on the Vega: FE card, dropping about 9% off its clock by the end of the test. The point is that while both of these devices can avoid clock throttling by running higher fan RPMs, neither of them is quiet while doing so. This is precisely why AIB partner models normally exist. Our liquid-cooled hybrid mod of the Vega: FE card will further emphasize this, and should be live shortly.
To provide perspective versus the AIB partner 1080 Ti cards, which is when we started using this testing methodology – hence their abundance – we see a prime example of a noise-efficient cooler in the ASUS Strix ROG card. The Gaming X isn’t far behind, using its dual-axial fans at ~100mm to drive down thermals with a lower noise cost. Blower fans have never been good at this, and that remains true with Vega: FE; that said, Vega does manage to retain a lower temperature at its noise level than the Titan Xp. Neither should be run at 40dBA, though, and should instead be left to run higher fan RPMs to reduce clock throttling that translates to performance loss.
Moving on to measurements of the hottest power MOSFETs on each board at 40dBA cooling, we see AMD’s Vega: FE hotspot FET at around 54.2C delta T over ambient which, although a few degrees warmer than the Armor, is still completely within operating spec. Remember that power components can take far more heat than a GPU, with many supporting 125C or 150C before any major problems develop. In this instance, we remain below 80C prior to our delta T calculation. That’s acceptable.
AMD Vega: FE Board Component Temperatures
Speaking of board components, let’s take a look at some of Vega: FE’s component temperatures under three different states: Auto, OC with a 3700RPM fan to prevent throttling, and our 40dBA noise normalized config.
The GPU certainly runs warm, but it’s nothing out of the ordinary for a blower card. We see this performance almost identically on both nVidia and AMD blowers, hence why we normally recommend buying AIB partner models (when they exist). For the Frontier Edition, though, it’s between air and liquid units from AMD right now. We’d recommend the DIY option, which we’ll soon do.
Regardless, FET temperatures in both our chosen hotspots are well within operating range. The backplate itself, which was measured on the inside of the aluminum plate, opposite the top FET measurement, posted a temperature of around 61C to 65C. This card gets physically hot to the touch when it’s running, just as an observation. Don’t touch it while it’s running, probably.
Noise Plot vs. RPM
This chart shows fan noise at each RPM, measured in 10% increments and at significant milestones – like lowest duty cycle and average auto duty cycle. Of note, the Vega FE card tends to auto-spin at around 40-41% in our bench setup, with the Titan Xp spinning at around 50-55% of its total fan speed in our setup. Keep in mind that the fan sizes and speeds are different, so these percentages mean different things for each device. The Vega FE card can spin up to 4900RPM, for instance, so its 40-41% is around 2000RPM. This places Vega: FE at around 47dBA output in its average auto configuration, with Titan Xp at around 47-47.9dBA output in its average auto configuration. These two are pretty close in noise output when configured to their auto, stock fan curve. Matching them in noise output is the best way to see a side-by-side, but we already did that in our noise-normalized charts earlier.
For reference, a 1080 Ti reference card follows effectively the same fan noise curve as the Titan Xp. Adding a liquid cooler to the card produces the results plotted in our EVGA SC2 Hybrid ($810) line on this graph, showing significant reduction in both noise and thermals.
And again, check the earlier page for the Blender render time vs. power & thermals chart.
Conclusion: AMD Vega: FE & RX – Too Soon to Call
Vega: FE ($1000) is not necessarily an appropriate tool to start speculating on the potential performance of RX Vega. We do not yet possess enough information to make definitive statements on RX Vega’s performance. It would be reasonable to suspect that RX Vega will perform within a reasonable stretch of Vega: FE, but that’s as far as we’re willing to speculate right now.
As it stands now, it certainly makes sense why the community was left feeling disheartened or disillusioned regarding Vega: FE’s gaming performance – but we’d suggest playing the “wait and see” game on RX Vega. Part of this is on AMD for naming its first card on Vega, well, “Vega.” That implies a flagship positioning, which seems to have unintentionally conveyed certain performance expectations to segments of the community. This could very well go either direction, and RX Vega’s viability as a product will almost entirely come down to its retail price. Vega: FE should, we think, have carried a different name to reduce confusion. It’s not a FirePro card, so that makes things difficult, but “Vega” just doesn’t seem appropriate. The wrong impressions are given, and that has led to a significant portion of the upset that we’ve seen.
Now, there are valid reasons to criticize Vega: FE – power consumption is certainly a bit high and the blower cooler is… a blower cooler—not very good. That said, until we have more confirmed specifications or pricing on the RX Vega card, it’s simply too soon to tell how well RX Vega will perform (for its price, which is all we ever look at) in gaming applications.
Our thermal, power, and noise tests do provide a solid, largely immutable understanding of how Vega responds to different scenarios. The noise aspect will change if AMD’s cooler changes with RX Vega – that’s still up in the air – but the power and thermals will perform similarly across all same-silicon GPUs. That’s why so much of the focus was on these more “boring” aspects today.
Vega: FE performs well in Catia, Creo, medical, and power simulation software (with SPECviewperf), though fell behind in our Blender benchmark, in 3DS Max, and in Maya. We also saw boosted power consumption and thermals over the Titan Xp in our Blender render pass – but again, as stated on that page, we cannot possibly test for every single use case for Blender. Regardless, in our testing, the Titan Xp pulled ahead in those 3D animation and modeling applications. It is also fair to say that Vega: FE is presently being crushed in all aspects of price:performance for gaming and, although it “isn’t a gaming card,” the intent was still to market Vega: FE and its odd positioning as a “content creator card.” This means, as AMD told us, that the user might be something of a developer who also needs the ability to game. Technically speaking, Vega: FE grants that ability – but mostly in the strictest of technical definitions. The card struggles with 4K gaming, something that GTX 1080 Tis can largely handle without issue. Gaming performance is similar to a 1070 & 1080, from what we’ve seen so far.
Some of that can be explained by drivers, some of it can be explained by limited focus on RX Vega. That, again, is why we’re suggesting that folks bury any expectations and simply wait for the more appropriate product for gaming. Vega: FE has its market – it’s just not us. Keep expectations checked for RX Vega, as it will be at least in the orbit of these results, but there's not yet reason to try and pretend we can guess at its performance, either. There will likely be improvement -- but don't expect 30-40%. Keep it realistic.
Looking again at Vega: FE and ignoring RX Vega, keep in mind that there are still plenty of kinks to work out of the drivers. WattMan is buggy right now, with overclocking at least partly broken, fan speed control somewhat delayed/buggy, and occasional colored screen hangs (like AMD fixed after the last time we reported them). Vega: FE has room to improve in at least these ways, if not in something more nebulous – like gaming.
We’re hanging onto this card, so we’ll re-evaluate when the time comes. As updates arrive, Vega: FE will remain on our list of revisited items to plot performance improvement or stagnation over time, as the case may be.
Editorial: Steve Burke
Video: Andrew Coleman