“Disillusioned and confused” could describe much of the response to initial AMD Vega: Frontier Edition testing and reviews. The card’s market positioning is somewhat confusing, possessing neither the professional-level driver certification nor the gaming-level price positioning. This makes Vega: FE ($1000) a very specifically placed card and, like the Titan Xp, doesn’t exactly look like the best price:performance argument for a large portion of the market. But that’s OK – it doesn’t have to be, and it’s not trying to be. The thing is, though, that AMD’s Vega architecture has been so long hyped, so long overdue, that users in our segment are looking for any sign of competition with nVidia’s high-end. It just so happens that, largely thanks to AMD’s decision to go with “Vega” as the name of its first Vega arch card, the same users saw Vega: FE as an inbound do-all flagship.
But it wasn’t really meant to compete under those expectations, it turns out.
Today, we’re focusing our review efforts most heavily on power, thermals, and noise, with the heaviest focus on power and thermals. Some of this includes power draw vs. time charts, like when Blender is engaged in long render cycles, and other tests include noise-normalized temperature testing. We’ve also got gaming benchmarks, synthetics (FireStrike, TimeSpy), and production benchmarks (Maya, 3DS Max, Blender, Creo, Catia), but those all receive less focus than our primary thermal/power analysis. This focus is because the thermal and power behavior can be extrapolated most linearly to Vega’s future supplements, and we figure it’s a way to offer a unique set of data for a review.
AMD Vega: FE & Who It’s For – Clarification of Marketing
Again, a large part of our review is heavily focused on thermals, power, and noise, as we think we can cover those aspects of this card most adequately, with some additional tests for production, gaming, and FireStrike. Before anyone starts screaming that this “isn’t a gaming card,” let’s clarify a few things:
For our thermals, power, and noise testing, that doesn’t matter. Those things don’t change too much based on the use case (power does a bit, but as long as the GPU is loaded down, we’re mostly good). We focus a lot on thermals and noise in our reviews, so it follows that we should give Vega: FE the same treatment.
Secondly, and more importantly, Vega: FE and its marketing has been enshrouded at least partially in either confusion or disillusionment by the community, depending on who’s asked. We spoke with AMD in length about the Radeon Vega: FE card’s unique position in the stack. The device isn’t quite a card targeted at the core gamer audience, but also isn’t exactly a professional-class card for simulation-heavy applications where you need and pay for driver certifications for assurance of accuracy. The Radeon FirePro line is reserved for that side of the market, into which the Radeon Vega: FE card does not fall.
This card is split-marketed at gaming and developer applications. This isn’t for users who would be better served by Firepros & Quadros (think: Boeing) with application and driver-specific certifications or assurances. In our conversation with AMD, the company was big on the fact that it's a pro-ready card with pro drivers, but that it also has gaming drivers packaged in; this is somewhat similar to a Titan Xp, except (theoretically) with more pro-level optimization. That makes Vega: FE a very odd card, because it's positioned outside of part of the professional engineering space (no certification, no purchase), but also outside of the core gamer space – just like the Titan Xp is. Your typical buyer would likely be someone like a game or application developer, who might create content on a gaming-class machine and also require the ability to playtest said content on the same machine, while keeping cost low. Certified Pro-class cards aren’t expensive if you’re designing a $352 million aircraft, but are expensive if you’re designing a game -- or most other software, really.
As for why someone would go Vega versus a 1080 Ti or (potentially) future RX Vega, there's still a difference between "I'm a game developer" and "I lose millions of dollars if a simulation is wrong," and "I'm a game developer" is still a business, and can still justify a larger business purchase. Vega: FE, theoretically, is one of those larger business purchases that could be justified – just like the Titan Xp. It doesn’t make sense for our core consumer audience, but going cheap on hardware for a business will only cost money later on. Cutting pro-class certifications cuts cost significantly, but does so without hampering hardware performance in equally significant ways (for these types of people). Vega: FE is still significantly cheaper than Quadro or Firepro equivalents, here. Freelance and home users should go Ti or maybe RX Vega – we’ll see – but that doesn’t fit the bill for everyone. There are other audiences where the high-end devices make sense, like machine learning that's memory-dependent, for instance.
Now, that’s not to say Vega: FE is the best fit even for that audience, but it is to hopefully clarify some of the marketing language and AMD’s intentions, based on our conversations with officials.
Let’s get into this card.
AMD Vega: Frontier Edition Tear-Down
Our earlier article and video present the disassembly process of AMD Vega: FE. All told, it’s not too difficult to disassemble. The card uses a Delta fan capable of 4900RPM maximally, which rests over a barren area of PCB and funnels air through a standard aluminum baseplate. Atop the GPU rests a vapor chamber cooler, with HBM2 present on an interposer that sits atop the substrate. The HBM is not “on die,” but is “die-adjacent,” and so can be cooled by the same finstack and coldplate as the GPU proper.
As for the VRM, our PCB analysis should be posted later today with full details. The AMD Vega: FE uses a doubled-up 6-phase for the Vcore VRM, resulting in an effective 12-phase Vcore, using IR direct FETs for the staging (IRF6894 low-side MOSFETs and IRF6811 high-side MOSFETs). The HBM VRM uses a single phase, shown in the bottom-right (black module) of the Vcore VRM.
Vega’s FE variant also runs an OnSemi Dual-NFET (NTMFD4C86N), with an International Rectifier IR35217 voltage controller. This is comparable to the 290X Lightning VRM, making it one of the most powerful Vcore VRMs we’ve seen on a reference card to-date (though the 1080 Ti was legitimately good).
Subscribe to our YouTube channel to catch the full Vcore VRM & PCB analysis later today.
Overclocking Challenges with Vega: FE
Before getting started, some notes on overclocking on this card: It seems that there’s an HBM clock bug when using the WattMan software in nearly any capacity. When overclocking, underclocking, or even restoring a change to defaults, HBM drops from 945MHz (reported) to 500MHz (reported). Again, restoring to defaults does not seem to fix this, and a system reboot must take place to correct the HBM clock. This means that, despite managing a somewhat stable 1682MHz clock (and fully stable 1667MHz clock at 1250mv), we were unfortunately stuck with slow memory due to not-understood behavior of either the card or the driver package. We also encountered a bug with fan speed in WattMan – these are on multiple systems, too – where it’s difficult to get fan speed to ramp back down once it’s been manually incremented. We found it best to either restart for the fastest fan speed reset or restore to “auto,” then wait patiently. Ramping-up isn’t a problem, but returning down is.
Also on the front of fan speed, we found that the user-configured fan speed tends to be about 200RPM lower than what the card chooses to run. Keep this in mind if testing.
For all these reasons, but primarily because dropping half the clocks off of HBM kills performance, we have excluded the impact of overclocking from most of our tests. One or two FireStrike charts contain the data just to establish an understanding of how negative a -445MHz HBM2 clock can be. Hang tight for updates as we work with AMD to better understand what’s going on, as we presently are unsure. We have been in contact with AMD re: Vega for some time now.
UPDATED OC NOTES
Since this review, we have discovered a few things of which potential buyers should be aware: WattMan is in fact bugged to drop HBM speeds to 500MHz when making any changes. This is a driver/software-level issue, and we think it's from left-over code from the Fury X (500MHz). Second, we discovered that "reset to defaults," the button in Afterburner that's useful after a failed OC, will lock the PWR + TEMP targets, drag power target (expectedly) down to 100%, and then drag temp target to 35C. This means that the clock will not exceed ~850MHz as the GPU is effectively never cold enough to permit that operation. Manually unlinking the two and pushing the temperature target back to 85C will resolve this issue, and will permit the clock to return to normal behavior. Restarting will also work.
Just wanted folks to be advised. We are working on OCing still, just running into lots of bugs.