Advanced Micro Devices: Cutting Through The 'Noise' And FUD

Nov. 8.13 | About: Advanced Micro (AMD)

The easiest way to scare the consumer, or an investor, is to spread fear, uncertainty and doubt. This becomes infinitely easier if it is done so surrounding issues that are hard to understand and are therefore more easily blown out of proportion. I typically ignore these "issues" because I find them to be largely irrelevant, but I have received many questions asking my take.

In this article I will dissect some of the negative reviews Advanced Micro Devices' (NASDAQ:AMD) new GPUs have received, explain my thoughts on the implications, and state what my biggest concerns are regarding the GPUs and their potential success.

Negative Reviews Regarding Sound

Most of these reviews test GPUs for sounds either on a test bench or with the case open measuring from a relatively close distance - closer than your head would be to the case while gaming.

A cheap $25 pair of headphones offer enough noise isolation to make sound a non-issue for those that use headphones for gaming, or $15 sound dampening foam placed inside the case to dampen the compressional vibrations that are sound -- the foam just eats the sound energy to make it go away (somewhat). The price of the R9 290, sound foam, and cheap Sennheiser headphones comes out to still $60 cheaper than the GTX 780.

Given that 3rd party coolers supplied by AIB (add in board) partners are usually much quieter and more effective than reference coolers, I feel the overall maximum performance is much more important than sound. Those that want to buy an AMD R9 290 but are sensitive to noise may be better off waiting on 3rd party cooling solutions. Sound doesn't bother me personally, so I ordered the R9 290 so I could test the "issues" I will describe later.

As an investor, I care more about overall performance than sound, since sound is easily remedied by the third party cooler.

Source: AnandTech's R9 290 Review, Crysis 3 Noise

(click to enlarge)Click to enlarge

Source AnandTech's GTX 680 Review, Metro 2033 Noise

Although Ryan stated he could not recommend the R9 290 based on noise levels, you'll notice there is only 5 dB separating the GTX 680 and R9 290. 5 dB would likely be somewhere around a 30% perceived increase in volume (10 dB on a logarithmic scale doubles perceived loudness). Cooler "loudness" is a subjective quantity, and according to Ryan's comments in his article, noise level is measured one foot from the card with the case open, so the perceived loudness will be further muffled from the case, drowned out by headphones or speakers while gaming, and any acoustic measures (such as installing dampening foam) inside the case.

For perspective, the following graphic is taken from SengpielAudio:

57 dB while gaming with the fan speed set to 47% rests somewhere between average background noise in a home and conversational speech, and this is if you feel like playing games while laying on the floor beside your tower with the side of the case open and no headphones.

For some consumers, noise may be an issue. For others, the compelling price/performance ratio will be the deciding factor.

Negative Reviews Regarding Variance

The biggest issue in my mind is Tom's Hardware Guide finding that cards they had purchased via retail performed worse than the samples they were provided for review. This has led to further investigation by members of the press, which has actually caused some sites to pull recommendations for the new GPUs.

Design Goals of Hawaii

First, keep in mind the design goals of the R9 290. During the THG Q&A with AMD, a representative stated:

"Product cost is a function of die size (and other parameters). We were confident that we could achieve industry-leading performance on a twenty-something percent smaller die using GCN, and we knew that would, in turn, give us a more attractive price for gamers.

We went for it. And thus $399-$549 hella fast GPUs were born."

From the beginning, AMD has stated they were not targeting the $1000 market. Based on the design choices, along with statements from AMD, it seems like the design goal of the new GPUs from the beginning has been to compete with Nvidia's (NASDAQ:NVDA) best on performance at a ridiculously lower price.

You do this by packing as much performance as you can into the cheapest possible package. Joel Hruska (in the comments section of his article on ExtremeTech) describes the AMD cooler "powerful", "capable", and "loud."

Nvidia's reference cooler for the high end GTX Titan is undoubtedly a quieter design; that is evident in the reviews; it is likely more costly as well.

Nvidia's GTX 780 and Titan cards also tend to run cooler based on the reviews. But AMD takes roughly equal performance, and squeezes it in to a much smaller die. And to do this, the card is operated to the limit of 95 C. The attempt is to balance acoustics (fan speed), cooling (fan speed), and performance (clock speeds), and to achieve the best balance the allowable temperature is set higher. Although AMD's cooler is capable, it is louder than Nvidia's, but likely also comes at a lower price allowing AMD to price the cards more aggressively.

When the original R9 290X was released last week reviewers, along with Nvidia, had jaw dropping reactions. Thanks to AMD, those that want to maximize performance per dollar get cheaper options, whereas consumers of Nvidia cards get cheaper prices and some games.

But because of Hawaii's smaller die size and likely overall lower BOM, AMD is able to offer a price/performance ratio that Nvidia will likely have a hard time matching without sacrificing margins.

The Issue With Operating At The Limit

AMD's Radeon R9 290 has a problem, but Nvidia's smear attack is heavy-handed -- ExtremeTech's article regarding the issues

The article above essentially explains how the press has managed to take a small issue with likely an easy fix in order to distort the severity of the problem -- FUD.

(click to enlarge)Click to enlarge

(click to enlarge)Click to enlarge

The above graphs depict the same throttling issue. Going back to AMD's attempt to balance noise with thermals, the blower operating at lower RPM isn't able to adequately cool the R9 290/290X to prevent clock throttling. The first graph shows that performance is stable with higher clock speeds. The second graph is, in my opinion, a smear campaign against proper operation. According to the article on ExtremeTech, the graph above from Nvidia is built using an R9 290X in "quiet mode." This makes the behavior of the R9 290X seem erratic, when in actuality it is controlling GPU clock speed at a lower fan RPM in order to maximize performance while minimizing fan noise. This represents an extreme case, and one Nvidia can better use for marketing.

Below are two screen captures posted by TechReport forum member "JohnC" while using AfterBurner to measure specs.

(click to enlarge)Click to enlarge

(click to enlarge)Click to enlarge

To explain what you're looking at, the top graphic shows the throttling behavior, whereas the bottom graphic depicts a more static behavior (the really crazy line is GPU usage which varies wildly based on game play, so although it looks highly erratic, it's not).

Between these two runs, the fan speed is adjusted up from 40% to 50%, which is enough to move the top number (GPU temperature) away from the operating 95 C point to allow stable operation at 1000 MHz clock speed.

The card that is designed to operate at 95 C, and with a constant fan speed, clock frequency is controlled to maintain the card temperature in spec. This is normal operation.

The most frustrating part of this I find is that, at least to me, the issue was blatantly obvious from the moment I read the reviews and looked at a few graphs. And not only to me, but to SWEClockers as well. Even Tom's, in their own article, show the problem is from frequency throttling due to inadequate cooling.

(click to enlarge)Click to enlarge

The issue was created by enforcing "the same 40% fan speed limit to give you an idea of how much more variance there is between the troughs and crests. AMD gives the 290 an "up to" rating of 947 MHz, but our seven games average 832 MHz."

Summary of Issues

This is the second time in a relatively short period of time where I have noticed issues surrounding AMD being blown out of proportion.

The first was regarding AMD crossfire issues and greater than high def gaming. It is something that AMD is actively fixing, has been largely eliminated in the new Hawaii GPUs, and affects a small percentage of the gaming community. Legit Reviews noticed the same thing.

Now the issue is the press focused largely on noise and GPU throttling, with the first issue being subjective and the second issue being fixed by a driver update.

Regarding noise levels, this is a completely subjective quantity, and in my opinion has little basis for being used as the sole means to recommend for/against any GPU. Using AnandTech's benchmarks of the GTX 780 Ti to illustrate a point, Nvidia's new flagship carries a 75% higher price tag, yet only offers ~20% or so higher performance on average over the R9 290. Using R9 290 noise to recommend against a card would be the same as using a price tag that is higher, but within reason, to recommend against the 780 Ti. Thomas Ryan of SemiAccurate published findings that support my own experience -- the noise isn't noticeable during game play, especially while wearing headphones.

Regarding variance between cards, this appeared to be an issue regarding inadequate cooling at low fan RPMs, and one that largely goes away when higher RPMs are used.

But rather than raising fan RPMs to show when throttling ceases, similar to what "JohnC" or SWEClockers has done, the graphs on THG are setup to show performance variance in AMD cards. My guess is that anyone capable of installing a GPU can also raise fan RPMs to get stable performance from AMD's cards -- it's easy and there are multiple ways of doing it.

This one I believe could have been a big deal, so I used it as an excuse to upgrade to an R9 290. I passed along my test results to Joel Hruska. I noticed absolutely no performance degradation using the Metro LL benchmarking tool ran through 6 loops using very high settings, the driver that was causing issues for Tom's, and at 1080p. Some consumers or press outlets may have gotten unlucky and received cards which exhibited higher fan variance, and with the card designed to operate up to the 95 C limit in order to maximize performance while minimizing die area, these fan variations translated into performance variance.

Prior to AMD releasing the company's new beta driver, the fix was simply installing a program such as MSI's AfterBurner and manually raising the fan speed. However, AMD has resolved this issue via a driver fix.

Conclusion

AMD has managed to squeeze a ton of performance in a smaller die. Nvidia has had the luxury of record margins as of late, but the competition from AMD has forced Nvidia into a round of price cuts on their older cards. It also forced Nvidia to release the GTX 780 Ti, which is a top binned, fully enabled GK110 GPU (read this as more expensive than the standard GTX 780 GPU), but it sells for $300 less than Nvidia's GTX Titan.

Because AMD uses a smaller die, and probably a cheaper blower, the company is likely in a better position to compete on price.

Regarding performance, once AIB partners such as Sapphire release cards with non-reference blowers (Sapphire's "Toxic", for example), noise and cooling will likely become less of an issue, giving these AIB partners more headroom to push performance closer to GTX 780 Ti territory at lower price points.

Lastly, during Tom's Ask-Me-Anything session with AMD, AMD's GPU team stated something to the effect of "we would not release Mantle if it were only for a 3-4% gain." The company also discussed partnerships with key game studios, with CryTek being the one that piqued my interest the most. CryTek owns the CryEngine studio, so if CryTek ends up supporting Mantle, this would be very good news.

I bring this up because based on looking at initial reviews of AMD's cards, I believe there is room to further push the performance envelope with better cooling. Games that take advantage of Mantle will likely give AMD another healthy performance boost, meaning it could push AMD's performance above GTX 780 Ti levels in Mantle games.

The Jon Peddie Research firm should be releasing add-in-board market share numbers this month. Although I believe it's entirely possible Nvidia gained market share during the most recent quarter, I feel AMD should be able to gain additional add-in-board market share during Q4. AMD's flagship products were released simply too late to have much of an affect during Q3.

AMD has used older GPUs, along with the new Hawaii GPU, to do a top to bottom refresh across the entire GPU product line. This becomes more significant in light of the fact that Mantle, AMD's new low level API, is only supported by GCN based graphics cards. Meaning AMD users that are interested in Mantle, but have graphics cards predating the 7000 series, will need to upgrade. But since the cheapest GCN based graphics cards start around $75, the price point to explore Mantle is easily accessible.

Next week, we also should get a preview of Mantle performance during AMD's APU 2013 event, along with the strong possibility of an in-depth look at the upcoming Kaveri APU.

Disclosure: I am long AMD. I wrote this article myself, and it expresses my own opinions. I am not receiving compensation for it (other than from Seeking Alpha). I have no business relationship with any company whose stock is mentioned in this article.

Additional disclosure: I am long AMD in both shares and options and actively trade my positions. I may add or liquidate shares/options at anytime.