Contributor Paulo Santos published an article regarding the move in Advanced Micro Devices' (NASDAQ:AMD) share price on Tuesday based on the increased demand for GPUs from current increased interest in litecoin mining, essentially writing it off as a non-factor. He points out that ASICs (application specific integrated circuit) and FPGAs (field programmable gate arrays) are much better for mining, and litecoin is better suited for those tasks.
In this article, I will look at the differences between bitcoins and litecoins in relation to the significance for AMD, as well as look at some numbers to explain the impact.
Bitcoin Vs. Litecoin
This article is not a recommendation for/against mining virtual currencies.
The video above kind of gives a pretty simple breakdown as to what cryptocurrencies are and how they're used.
To call out the most important part concerning litecoins, notice the blurb about "mined efficiently with consumer-grade hardware." Also, the paragraph concludes by mentioning eventually ASICs are likely to take over the litecoin network.
So how do you create different kind of virtual funny moneys? Simple, you change the hashing algorithms in a manner that introduces latencies within the calculations by requiring access to L1/L2 cache; however, because scrypt hashing algorithms only require access to 128.5 Kb of cache, L3 & RAM is left untouched slowing the overall process down while simultaneously allowing consumer grade hardware to be used without drastically affecting system performance.
What Did You Just Say?
Most of my articles are very technical, so I will try and break it down in a more simplistic fashion to make this one a little easier to understand.
ASIC stands for application specific integrated circuit. Graphics cards are a type of ASIC; they render images.
FPGA stands for field programmable gate array. Think of FPGAs as slightly more versatile, with the tradeoff being that while they're cheaper to make, they're not quite as good as ASICs, at least in the instance of mining virtual coins.
In Paulos' article, he essentially points out that ASICs may be a ways off, but FPGAs are closer.
If you like to watch cars drive in circles for hours on end while drinking Miller Lite like I do, you'll recognize the above photo as a pace car. Pace cars slow the race down. This is the difference between bitcoin and litecoin mining.
Bitcoins can be mined by just performing calculations as fast as possible, whereas litecoins require access to larger amounts of memory (Source: LiteCoin.Info); this acts as a pace car, or at least this is how I think about it. The actual differences in how the coins are mined creates fundamental differences that affect the ability of the hardware.
But real world impact is what's most important, so let's look at the real world situation.
The FPGAs used for litecoin mining that Paulos mentions do not exist yet. Looking at the source for the articles he links to, we come to CryptoIndustries.com.
Let's look at some screenshots from CryptoIndustries' website.
All these things sound great, but you can't actually buy one yet. If you want to purchase an FPGA litecoin miner, you must sign up for a newsletter, and they'll let you know when you can buy one.
They claim their rigs will mine faster than anything you've had before, and this is probably true at a given power consumption. CryptoIndustries offers three separate miners: a 1 Mega-Hash/sec miner with a 150W power supply, a 5 Mega-Hash/sec miner with a 750W power supply, or a 10 Mega-Hash/sec miner with a 1500W power supply.
The bigger the "Mega-Hash," the more mining power in the FPGA.
Miners like Radeon 7950s. According to LiteCoin.Info 7950s have about a .7 "Mega-Hash/sec" rate and about 240W. So let's make-up a number to make this mean something, and let's call it hash/watt.
.7 "Mega-Hash/sec"/240W = .003
Now, let's make up the same number for one of the FPGA miners.
1 "Mega-Hash/sec"/150W = .007
So for the sake of argument, let's say that FPGAs that don't exist yet are just over twice as good as the Radeon 7950. Although 7950s are hard to come by, there are lower tiered Radeons that offer similar performance/watt as the 7950 that can be purchased right now.
The difference was much more drastic between the performance/watt offered by GPUs and FPGAs for bitcoin mining.
So, making up hash/watt ratios like I did above to understand the differences in relation to bitcoins, let's look at the various devices. I am just picking one example at random to demonstrate ballparks here; exacts are not important for what I'm trying to prove.
GPU: 550 "Mega-Hashes/sec" / 185Watts = 3
FPGA: 800 "Mega-Hashes/sec" / 40Watts = 20
ASIC: 10,752 "Mega-Hashes/sec" / 350Watts = 130
FPGAs are ~7x faster than GPUs in bitcoin mining; ASICs are ~40X faster! Note, there is some differences between various ASICs, FPGAs, and GPUs, but this demonstrates one huge difference; the need to access memory (the "pace car" example I have given), seems to shrink the difference between the various devices. Also, note that mining bitcoins is about 1000 times faster (.7 "Mega-hash" for litecoin vs. 700 "Mega-hash" for bitcoin).
Once FPGAs are introduced for LiteCoins, they look like they will have an advantage over AMD's GPUs. This will cause this increased demand for GPUs to drop. But there are none available yet; no FPGAs, no ASICs. Because of the differences between litecoins and bitcoins, the competitive standing between GPUs and FPGAs specifically for litecoin mining is more narrow than it was for bitcoin mining.
How Important Are Desktop GPU Sales to AMD?
Very. Newegg.com shows about 80 or so (I didn't count thoroughly, just a rough guess) Nvidia (NASDAQ:NVDA) 700 series GPUs powering notebooks. AMD has "about" 13 Radeon 8000 series notebooks. Nvidia is the dominant player in the notebook space, and in the professional GPU space, so consumer grade desktops are the most important source of revenue for AMD by far.
The cards being bought for mining are just your average, run-of-the-mill desktop cards.
Putting some numbers to AMD's unit shipment, I like to use research from Dr. Jon Peddie. In this press release from JPR for Q2 results, we can get some ballpark figures.
The same research report states that 14M add-in-boards (discrete cards) shipped during Q2. At a 38% market share, AMD ships about 5.32M units/quarter. In Q3, I estimate AMD's GPU revenues dropped from around the $330M level during Q2 to around $270M in Q3, based on total GVS revenues of ~$670M, with ~$400M of those revenues coming from consoles. This drop was explained during the Q3 conference call as being a direct result of transitioning from the 7000 series to the R9/R7 series of GPUs.
Just to ballpark these numbers (this is *very* rough), AMD likely shipped around 15% fewer GPUs in Q3 because of this transition, based on the sequential revenue drop from Q2 to Q3, which would mean around 4.5M discrete GPUs or so. Again, this is very rough, and could be off. But the exact quantity isn't quite as important as the point I'm trying to make.
If you've read my articles, you'll know I prefer technical details over strict financial analysis, so let me break from character. People like to talk about basis points of market share. Assuming a similar number for Q3 (again, this could be off - I am only using these numbers illustratively to describe relative magnitudes) of 14M units, each 140k units equals 100 bps of market share.
So How Much Is This "LiteCoin Craze" Impacting GPU Revenues?
I have been watching this unfold for a little while. After the drop of GPU revenues during Q3, I wanted to see how fast 7000 series inventory cleared out of the channel, so I would keep an eye at various e-tailers, and at the beginning of October, I noticed there were still some 7990s, and plenty of 7970s. As the R9 290X and R9 290 were the two new GPUs, I would also see how often they came in stock and how quickly they would sell out. As the majority of the R9 and R7 series cards were refreshed versions of the 7000 series, 7000 series cards remaining in stock meant retailers and add in board partners still had old inventory to clear out, which could prevent them from ordering new GPUs, which in turn could suppress GPU revenues while waiting for the channel to clear.
I noticed that specific GPUs were clearing out faster than others; those that are recommended for mining. CryptoBadger is a website with a "how to" guide for building mining rigs. And if you read that guide, notice the cards he recommended and how quickly they sold out - he has several updates within his guide. A simple check of Newegg inventory (I, II, III) shows that the cards he has recommended are the ones that are sold out. The R9 280X and 7970 are based on the same GPU. Notice there is a decent amount of 7000 series inventory left, but only the cards that have not been recommended for mining? At the time of writing (12/10/13, 7:57 PST) there was only 1 of each of the R9 280X, R9 290, and R9 290X in stock at Newegg, and these are all at a higher price over MSRP, indicating increased demand. When BSN published their article, Anshel noticed higher quantities of inventories available.
Again, at the time of writing, the AMD Sempron 145 (CPU recommended by CryptoBadger) is the #5 CPU on Amazon's best sellers list.
The dots connect: the recommended GPUs for mining are the hardest to find, and the recommended CPU for mining is hanging out among the Amazon top 5. Maybe people are feeling nostalgic and buying a 45nm processor that was released in May of 2010?
As far as quantity, this is the tricky one. As mentioned, the 7970 is essentially the same GPU as the R9 280X. To assume that all of the sudden demand for a GPU that originally launched around December 2011 spiked so high that supply could not be met due to Mantle or some other factor seems illogical. Mantle was announced during APU 2013, when 7970s and R9 280Xs were readily available.
Previously I had pointed out that 8000 total BF4 edition R9 290Xs released on launch day, at MSRPs of ~$550, which would equate to about $4.4M in sales. Since this card was a reference design, which was simply resold by AIB partners, assuming around 75% of this went to AMD would mean that AMD added around ~$3M to the company's topline within hours of the card going on sale. And the cards are selling out as quickly as they come in stock (source - Zoolert: I, II). Notice there is a slow trickle of R9 290Xs coming in, and selling out, according to the website. However, the R9 280X shows that the inventory has been coming in at a much slower rate compared to the R9 290X as of late.
To me this signals that AMD was not prepared for the demand for the company's R9 280X. This is supported by Zoolert showing the R9 290X being restocked many times throughout December, whereas the R9 280X has only been restocked twice during December.
And looking a little deeper:
On November 20th, the time interval used to measure how long the R9 280X could remain in stock shifted from days to hours or minutes, indicating the cards had a much higher turnover rate after this date. This date lines up with the dramatic rise in LTC value. If there is another plausible explanation for demand to spike so drastically for a GPU that was essentially launched in December 2011, there is a comments section below -- I am open to any alternative explanations.
The Real Proof
Nerds will realize the value of this chart; guys in suits and ties won't get it (unless those guys used to be nerds, but are now making more money and wearing suits and ties).
If you look at the difficulty of mining litecoins, it went on a tear on October 25th, jumping from a value of 1000 to 2600 - that's a 2.6x increase.
Mining difficulty is proportional to the total compute power of the network. This means that the amount of mining power went up 2.6x recently. Notice it rose most at the end of November? Notice this lines up with the rise in value of the litecoin, which coincidentally lines up with the very short turnover times in graphics cards? The mining difficultly jumped from 1500 to 2600 from November 21st until now.
You can lead a horse to water, but you can't make it drink. And if you try too hard, you might drown it. I'm pretty sure the spike in GPU demand is due to litecoin mining.
We can see that the total hash rate has grown around 50,000 "Mega-Hashes"/sec since the beginning of November.
Each card = ~.6 Mh/s, so this means around 85k cards have been added to the LTC network.
A good majority were likely 7000 series cards, which were being cleared out and are at end of life. I bring this up because 7000 series cards have more of a second order effect on revenue; those cards are already on the retailer's shelf, meaning that revenue has likely been booked in a previous quarter for AMD. By selling the old inventory, the way is now clear for AMD to turn over more of the new inventory.
During Q3, GPU revenues likely dropped to around $270M or so (based on looking at GVS revenues and console revenues). Dr. Su explained this during the Q3 earnings call as being caused by switching from the 7000 series cards to R7/R9 series. So with some of the 7000 series being cleared out, there is less resistance in the channel for new inventory; hence my descriptor of a second order effect.
Nvidia isn't competitive at LTC mining, and because of the differences explained above most of the ASICs and FPGAs used for bitcoin mining don't make sense here either. So if you assume that around half or so of those cards were the R9 280X ($300 card), R9 290 ($400 card), R9 290X ($550 card), and assume an ASP of $325 (I chose a lower value to account for retailer + AIB margins):
42.5k * $325 = ~$15M or so in revenues from this, give or take.
To put this in perspective, recall that ~4.5M desktop units sold during Q3 in a ~14M unit market. So AMD likely picked up around 30 bps of desktop GPU market share in Q4 that is solely attributable to litecoin mining.
On $15M in revenues, I'm not rushing out to buy champagne, but I'm also not quick to write-off the increase. AMD forecast 2-8% sequential growth on revenues of $1460M during Q3, so an extra $15M is 1% of that 2-8%. Also, when management made their Q4 forecast, this was before this increased demand, so it was likely not factored in. And using 40% GM and 750M shares outstanding, this could be worth around $.01 EPS or so to the bottom line. Consensus estimates for Q4 are $.04 to $.08, so an extra $.01 to the bottom line is a nice little addition.
Let's look at the totality of information presented here:
- LTC mining appears to have increased demand for GPUs
- It looks right now to be in an accelerating uptrend
- We have a source of information (network Hashrate) to gauge how quickly GPUs are being added
- Nvidia isn't competitive, FPGAs are just now being designed, and ASICs look to be a little ways behind FPGAs
- There are fundamental differences between bitcoins and litecoins, so the initial FPGAs look to be less competitive than they were when introduced for bitcoins. However, they are still an improvement over GPU mining
- This boost in GPU sales helps to clear out old inventory
If you read my previous articles (I, II), you'll see that nothing presented here changes those conclusions. Further strength in GPUs offsets AMD's dependency on PCs, and this strength could help gain GPU market share during this quarter.
There was virtually no interest in mining litecoins earlier this year. But when people started buying houses with bitcoins they forgot they had, this type of story makes national news and it attracts attention. If people want to mine litecoins, a GPU is the most effective way to do it right now. There are a lot of virtual currencies, many based on the same algorithms, so the competitive landscape between similar currencies is likely to be similar between GPUs, ASICs, and FPGAs. And there is re-use of this hardware between similar currencies, GPUs especially (GPUs are more versatile), so don't expect to see similar spikes in sales if another currency takes off. It's likely miners switching from litecoins to feathercoins or whatever flavor of coin is cool that month.
If you're an investor, you'll recognize the growth pattern of mining difficulty above as the start of a major trend. This is in its early stages, and is worth paying attention to. Going back to conclusions from previous articles, I have also said this is a volatile situation. Legislation, FPGAs, ASICs, or when Nvidia releases Maxwell GPUs in a few months down the road (based on leaks, I don't know exactly when these will be available), the competitive landscape could change. So this major trend could fall to 0 in relatively short order and leave a bunch of people with 40 GPUs stuck in milk crates in their garage feeling sad, trying to eliminate said sadness by selling these GPUs on eBay.
But until this happens, the best way to mine litecoins is to grab as many of the best shovels you can find and start digging. AMD sells the best shovels for the time being. The name of this game is to out-hash everyone else, so it's a race to the top. When all these crazy crypto-currency transactions are going on, there is a random chance that processing one of these transactions will net you some virtual funny money. You increase your chances by adding "MORE POWER" (if there are any old Home Improvement fans out there).
So he who has the biggest shovel laughs the loudest.
Forbes recently had an article in which they described the bitcoin network as being 256 more times powerful than the top 500 supercomputer combined. Throwing a GPU into the mix to mine bitcoins right now is about like trying to win the lottery while simultaneously being struck by lightning - good luck. However, the interest in litecoins and various other crypto currencies is just taking off, and right now while there are no/few FPGAs or ASICs, GPU miners have a decent shot. And by having more GPUs than the next guy, you're more competitive. Getting started early is key, because as more hashing power gets added to the network, the higher the competition to get that golden ticket in the form of being awarded some virtual crypto currency.
For right now there has likely been a decent chunk of inventory moved specifically due to litecoin interest. A good estimate for the sales right now I believe would be between $10M to $20M. Why is such a small number meaningful for AMD? I pointed out that after the Q3 earnings call, the bears started beating their bear drums to the tune of a declining PC market over a $50M sequential decline in PC revenues that was attributed to a loss of market share to Intel (NASDAQ:INTC). Small numbers matter when you're a smaller company. $15M is a rounding error for Intel; it's an additional 1% sequential growth for AMD.
The growth of the overall power of the litecoin network could provide some insight as to how high the increased demand would be for GPUs. And if the trend keeps accelerating and the hash rate grows, provided there are no FPGAs or ASICs introduced, most of that power is likely Radeon induced, as the red guys are the best option for litecoin mining right now.
Special thanks to Seeking Alpha readers "NeoSephiroth86," "SpinyNorman," SemiAccurate forum members "Chipshot64" and "Thrillseeker," and finally "CryptoBadger" and his website.
Disclosure: I am long AMD, INTC. 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. I actively trade my AMD position, and am long both shares and options. I may add or liquidate shares/options at anytime, or initiate a small hedge via puts at anytime.