So, last week I wrote an article regarding increased demand for Advanced Micro Devices' (AMD) GPUs due to increased interest in virtual currencies.
The demand for AMD's GPUs has risen so quickly it took me by surprise to be honest. In the previous article I focused primarily on the fact that increased demand could drive GPU sales back to historic levels, without really quantifying the impact. In this article, I would like to take a 10,000 foot view of the impact of consoles and overall importance of GPU revenues in relation to AMD as a whole.
Expected Sequential Revenue Growth From Consoles
It was estimated that AMD generated $401M in revenues from console sales during Q3. Given $671M in revenues total for the GVS (graphics and visual solutions) operating segment, this would imply GPU revenues of $270M.
Both IHS and TechInsights appear to have overestimated the console APU prices. I am basing this on CFO comments during the Credit Suisse Technology conference, which put console APU ASPs between $60 and $100, with no specifics given.
Based on the sum of these sources, I will be using an ASP of $90 for the console APUs.
Demand for consoles appears quite strong, with most retailers unable to keep stock for more than a few minutes. According to VGChartz (data taken on 12/8), console sales figures are approaching 4M units.
Sony (SNE) is almost at IHS forecasts with almost one month left prior to the quarter's end; same for Microsoft (MSFT), which is slightly lagging Sony in overall sales. Analysts have estimated that Sony should be able to manufacture ~1M units per month. Using this as a guesstimate for Microsoft's ability as well, this would imply that Sony should be able to reach ~3M consoles sold by year's end, with Microsoft coming in at ~2.5M units. These estimates would lead to 5.5M units sold, which is a 20% increase above IHS estimates. Given the strong current demand for consoles, I believe reaching this will be more an issue of being supply constrained rather than a lack of demand.
According to the Q3 conference call:
Now turning to the outlook. For the fourth quarter of 2013, AMD expects revenue to increase 5% sequentially, plus or minus 3%. Non-GAAP gross margin is expected to be approximately 35%, as our semi-custom business continues to represent a greater portion of our overall revenue. We expect to continue to maintain quarterly non-GAAP operating expenses in the $450 million range and expect to be profitable at the net income level.
This quote leads me to believe that AMD is expecting a good portion of the revenue growth to come from increased console demand as volumes continue to ramp.
Based on $1.46B in revenues for Q3, a 5% sequential increase (mid-point of guidance) implies a $73M increase in revenues. Assuming most of this comes from consoles, console revenues are likely to see an increase.
GPU Revenues: Never Look a Gift Horse In The Mouth
This article is not a recommendation for/against mining cryptocurrency.
Image Source: SudoSecure
In the article I linked to at the beginning of this piece, I pointed out that demand for AMD's GPUs is through the roof right now (I, II, III, IV, V, VI); this I believe is almost completely attributable to the increased interest in litecoin mining.
Without getting too technical, you have to understand the barriers to entry to understand AMD's moat in regards to mining. When you hear the term dual core or quad core CPU, CPUs are very good at doing extensive serial tasks. Think executing computer programs.
Graphics cards however are comprised of 1000's of tiny processors which are really good at doing things in parallel. Take their primary use case of rendering graphics. GPUs do post processing effects, which are little calculations applied to the million pixels on a computer screen. Although the calculations may not be that intensive, there are still a ton of simple computations that must be performed, and therefore are ideal for carrying out on highly parallelized GPU processors.
Because mining cryptocurrency is the equivalent of code breaking, the more calculations that can be performed per second, the higher the miner's chance is of successfully mining a unit of cryptocurrency. Because these types of calculations are highly parallel, processors from either AMD or Intel (INTC) are not really suitable replacements.
GPUs are a type of ASIC (application specific integrated circuit). Bitcoin mining has been largely taken over by ASICs designed to mine bitcoins. However, litecoins were designed to be mined with consumer grade hardware. The algorithms used to mine litecoins are also more bandwidth intensive, making it more expensive to design ASICs specifically geared for mining litecoins. However, if the price of the litecoin appreciates substantially, it may become feasible for litecoin mining ASICs to be produced which could make it less profitable to mine with GPUs. However, as far as I am aware, there are none available now, indicating that it is likely a few months away before AMD would have competition from ASICs.
Comparing AMD against Nvidia (NVDA) at the hardware level, Joel Hruska explains that AMD's focus on integer operations at the architectural level account for a major difference between the efficiency and performance between AMD and Nvidia GPUs. Unless Nvidia has already worked something into the company's GPU to increase mining efficiency, AMD is not likely to have competition from Nvidia GPUs at any point in the near future.
A Look At Mining Rigs To Understand the Impact
There will be a brief summary at the end of this section to briefly describe the significance to potential revenue for AMD if you would like to skip the details
The first thing you should notice is that a rig designed to mine litecoins consists of a $40 CPU, cheap motherboard, 3 more expensive GPUs and a good power supply. The GPUs are the workhorse of this operation, and the power supply is analogous to the oats that feed these stallions. The processor is like the jockey, and you want it to be as light as possible to minimize weight (costs); the processor is largely unimportant.
Notice how the recommended processor for mining is a cheap AMD Sempron?
A $30 antiquated Sempron is #5 on Amazon's best sellers list at the time of writing (12/8/13). Again, the CPU is unimportant for this operation. Notice the CPU recommended in the litecoin mining rig?
So the name of building a mining rig is to create a rig that has maximum efficiency in relation to the highest hash rate (calculations per second), at the overall lowest cost.
You may wonder why you would not want to use an Nvidia card if all the recommended GPUs from AMD are unavailable. That's because even the cheap GPUs from AMD do a much better job than Nvidia's flagship cards.
Looking at info on Tom's Hardware and LiteCoin.Info, you can see that a $200 AMD card hits the same, or higher, hash rates as a $600+ Nvidia cards. The Titan is a $1000 card that offers significantly less performance than AMD's R9 270X. You'll also notice that the R9 270/270X cards are much easier to find than the more expensive cards; AMD is selling out of the highest ASP, likely highest margin products.
And for anyone interested in mining, the AMD cards are still the best choice over Nvidia. This becomes even more apparent when factoring in power consumption (source: Tom's Hardware).
Normalizing for wattage, AMD's cards exhibit roughly 2.5x higher hash rate/watt. In order to profitably mine cryptocurrency, the value of the coins mined must exceed the cost of electricity + equipment costs. The cost of an entire three R9 270X mining rig setup would be roughly $900.
R9 280Xs are the backup choice for Radeon 7950's, but you can see the power consumption is significantly higher for the Tahiti based GPUs when compared against the R9 270X.
Normalizing for wattage, the R9 270X looks to offer an even slightly higher hash/watt metric over the R9 280X. But unlike the R9 280X, the R9 270X is currently in stock.
Summary of Section: The cards that are in highest demand are those that carry the highest ASP. Nvidia currently has no answer, and as far as I am aware, there are no ASICs to compete with AMD, giving the company a moat amongst miners for the time being. Given that the R9 280X and above are largely out of stock, and the demand has caused the prices of the cards to actually rise (feel free to check eBay, newegg, or your local Craig'sList to find used Radeon cards that are specifically recommended for LTC mining), the R9 270X is the next viable alternative, and is still better than anything Nvidia could offer. And in order to increase the power of mining rigs, consumers are placing 2-3 GPUs in a mining rig; not just one. You may think I'm being hard on Nvidia, but this operation is only profitable as long as the value of coins mined exceeds the cost of electricity. Although you could use Nvidia cards and hit the same hash rate, you'll be spending significantly more power to do it.
The Totality of PC Sales, Consoles, and GPU Revenues
AMD is currently projecting continued weakness in PCs, so the company's forecast likely includes lower CS revenues.
Regarding GPUs, AMD introduced two brand new GPUs during Q4: the R9-290 and R9-290X. The company also refreshed the 7000 series line by launching lower tiered R9 and R7 series GPUs. During the conference call Dr. Su stated the company was looking to gain GPU market share.
But looking more closely, the cards that are largely out of stock are the ones with the highest ASPs; the cards that will likely have the largest impact to the bottom line. And miners are not just buying 1 card; they are buying 2-3 AMD cards, as Nvidia is not really competitive in this space at this time.
The Mac Pro is slated to launch in December, which features two workstation class GPUs. During Q4, AMD typically sees higher relative console royalties from the outgoing generation consoles (Wii, WiiU, Xbox 360), however, some of this could be offset by the new generation of consoles.
These factors should drive sequential GPU revenue growth during Q4. Because these products carry higher ASPs, more of this revenue should flow through to the bottom line.
As an illustrative example, during Q1, AMD's CS segment achieved revenues of ~$750M. The GVS segment generated revenues of $337M (which at this point excluded consoles). I use these numbers because they would demonstrate a substantial decline (5%) in CS revenues, coupled with a substantial increase in GPU revenues (25%) that are likely ballpark guesses to where traditional markets could fall out.
Management's guidance has led me to believe the company expects console revenues from semi-custom to increase during Q4. If this is the case, using $450M as a guesstimate for console revenues, combined with the $337M from traditional GPUs, leads to just under $800M in GVS revenues. Compare this to Q3 CS (PCs + Servers) revenues of $790M.
During Q4, we could see AMD's revenues shift from depending mostly on PCs (~70% during Q1) to accounting for half or less of overall revenues.
This is a significant change, and shows more of a divorce of AMD from the overall PC market. Consoles give AMD a moat; there is no competition from Intel or Nvidia in the PS4 or Xbox One. Current estimates for PS4s + Xbox Ones sold show that total sales are only ~800k shy of IHS forecasts. Inventory turnover rates signal that in the near term, supply will be more of an issue than demand.
The only competition in consoles is from the Steam Machine. But it just so happens that some of these will also feature AMD hardware. The one linked to uses an AMD CPU and discrete GPU, meaning AMD is selling two pieces of silicon in each Steam Machine. Note that unlike in traditional consoles, AMD will also face competition from Nvidia and Intel, so you will want to pay attention to how Steam Machines affect traditional console sales, as well as the mix of hardware in Steam Machines.
AMD also has a temporary moat in cryptocurrency mining. For the lower budget DIYer that wants to mine, AMD GPUs have the lowest start-up costs, and Nvidia isn't competitive here. And specifically for litecoins, AMD has no competition from ASICs for the time being (unless I have missed some). Note this could change a few months from now, so AMD's window may eventually close. But for the time being, the increased interest in mining have helped retailers move a ton of AMD inventory, as well as clear the channel so that when more GPUs are manufactured, retailers will have to restock. I also do not believe this would have been factored into management's Q4 forecast.
Looking at a chart of USD vs. LTC, you can see that the value increased substantially (along with demand for AMD's GPUs) well into November, after Q4 revenues were forecast (Q3 earnings report was released on October 18th).
While I think we're likely to see strength in GVS, with a revenue decline in CS during Q4, I think this quarter will further illustrate AMD's shift away from being dependent on the overall PC market. Driving YoY revenue growth despite weaker PC sales speaks to AMD's turnaround strategy. Initiatives like Mantle and aggressive R9 290/290X pricing show calculated measures aimed at stealing marketshare from Nvidia. During the Credit Suisse Technology conference, AMD CFO Mr. Devinder Kumar stated that in 2014 console chips will be made at Global Foundries. This shows the company is actively managing purchase commitments and obligations to GlobalFoundries.
And the sooner the console chips can be shifted, at least somewhat, to GlobalFoundries, the more capacity that should be available at TSMC to manufacture GPUs, which would allow AMD to more readily meet demand for GPU products. However, I am taking an extremely short-term view on increased demand due to litecoin mining, as legislation, the introduction of LTC ASICs, or a collapse of the cryptocurrency markets would extinguish this increased demand in short order. But in the near-term, major financial publications and websites such as CNBC and Forbes publishing articles on cryptocurrencies will drum up interest. And when those interested Google "litecoin mining guide", the first tutorials they will find recommend AMD hardware. I did not stumble upon the CryptoBadger website I linked to earlier; it was one of the first websites I found when I used the Google.
In the end, the biggest takeaway I am trying to demonstrate is a further removal of dependence on the PC industry. Console revenues will be incremental for Q1 and Q2 2014, which should aid to drive YoY growth. AMD has stated the company intends to more actively pursue the professional GPU market. One thing I have not seen mentioned yet is that AMD managed to steal away one of the guys (Mr. Jean-Christophe Baratault) from Nvidia responsible for the success of Tesla GPUs. The article was featured on KitGuru, and is backed up by a quick check of his LinkedIn profile. All these moves target growth areas outside the PC sector.
Additional disclosure: I am long AMD both shares and options. I may add/liquidate shares/options in AMD at anytime, or initiate a small hedge at anytime.