On Thursday Advanced Micro Devices (AMD) released the company's much anticipated R9-290X. Reviews across the interweb unanimously praise the GPU, with many websites trumpeting the R9-290X as the "Titan Killer."
During the summer Apple (AAPL) announced the company would be refreshing the aging Mac Pro line, updating both the look and the internals.
The internals happen to include dual AMD workstation class GPUs, used for video and photo editing, HD video playback, design work, etc.
In this article I would like to briefly explain why the AMD price points of the new flagship GPUs are important (meaning why Nvidia (NVDA) will have a hard time matching price), quantify the potential impact of revenue from the new GPUs, as well as quantify potential Mac Pro revenues.
R9-290X vs GTX Titan Performance and Pricing
The R9-290X shocked consumers when it launched at a price of $549 (~$579 for the BF4 edition). To contrast this to Nvidia's offerings, the Titan is a $1000 card, with the cheaper GTX 780 coming in around $650.
If you notice the reviews, most review sites explicitly compare the $550 Radeon against the $1000 Titan, not the GTX 780. The R9-290X presents incredible value to consumers, with the biggest cons regarding the new GPU calling out elevated temperatures, noise, and power consumption. Two of the concerns will likely be addressed later in the quarter by AIB (add-in board) partners using custom cooling solutions.
ExtremeTech's Joel Hruska has an excellent write-up of the card if you care to read the full review. One of the more interesting features I would like to touch on is the level of user control over thermals and performance.
If noise and heat are a problem, users can adjust the parameters to dial down performance to their own personal tastes for thermals and acoustics. If you want it quiet, you can lower fan and clock speed. If you have the same setup as I do with tower on the floor under your desk, if it's cold in the winter and you want to warm things up a bit, you can do that too. Coincidentally, lowering frequency (and thus performance) should also decrease power consumption, giving users the opportunity to mitigate the power consumption concern. But at the end of the day, some consumers may opt for a different GPU if this is a concern.
Regarding GPU comparisons between Nvidia and AMD, both the R9-290X and the GK110 GPU used in the Titan are made on TSMC's 28nm process. However, the R9-290X's GPU is ~440 mm^2, compared against the much bigger ~550 mm^2 for the Titan GPU; the Titan is the bigger GPU, making it much harder for Nvidia to lower the price enough to compete for performance/dollar.
AMD currently has a gaping hole in the company's lineup in regards to GPU price. The R9-280X is priced at $299 and the R9-290X is priced at $549. AMD plans to fill that gap with the R9-290. My guess on the R9-290 price is $399-$459, based on AMD's current pricing scheme.
Quantifying The Impact And Why Discrete Sales are Important For AMD
AMD launched 8,000 total Battlefield 4 edition R9-290X's at a price of ~$580 USD. Assuming roughly a 35% markup from the time the card leaves AMD until the card is in the hands of the consumer, this would give AMD revenues of approximately $350 per card.
The 8,000 BF4 units likely generated between $2M and $3M in revenues for AMD. This becomes significant when you think about the speed at which this happened.
Jon Peddie Research pegs the AIB market around 14M units during the second quarter of this year. Note these reports track all discrete cards, including workstation and server class cards.
38% of 14M units means AMD ships around 5M units/qtr. During Q2, AMD's GVS (graphics and visual solutions) revenues came in at $320M. During this same quarter, analysts stated they expected revenues on consoles of ~$30M (on the Q2 conference call it was estimated during the Q&A session that console royalties accounted for $100M-$150M/year). Factoring out royalties, along with ~$20M to account for AMD discrete GPU notebook shipments (the ~$20M is a guess based on the fact that Nvidia has a much higher notebook GPU market share), AMD likely generated around $270M in revenues, with an average GPU selling price of ~$55.
If this price seems low, realize there is mark-up added by both the retailer and the AIB partner. Looking at newegg, you'll find 297 AMD cards, with a large percentage of those cards priced below $150.
Nvidia generated ~$860M from GPU sales in its most recent quarter. Factoring out ~$30M for consoles royalties, along with $100M to account for notebook GPU sales (guess based on Nvidia having a higher market share), means Nvidia likely generated around $730M in sales on 8.7M units. This equates to ~$85 per GPU.
Regarding proof of notebook market share, JPR also tracks total GPU shipments. The numbers include both discrete and integrated GPUs. The totals also count both the iGPU and the discrete GPU separately in the event a PC ships with both. In the free report there are no total unit shipments listed, only percentages.
Nvidia is within a few percentage points of AMD in GPU shipments, despite the company not having an iGPU solution that factors into these totals. Also, looking on newegg at laptops shows that of the units that use discrete GPUs, there are far more listed using Nvidia than AMD.
Note, the exact values of the GPU ASPs could be off based on my assumptions, but I believe the magnitude is the more important factor. The fact that Nvidia cards carry a higher price tag shows the value of commanding a higher premium on the company's GPUs, and how it flows through to the financials.
Prior to the launch of the R9-290X, AMD's most expensive card during the quarter was around $350 for a 7970 GHz edition, with the prices of the cards being slashed leading up to the R7/R9 launch. This effect showed in GVS revenues which would have shown a significant sequential decline had it not been tremendously offset by console sales (speaking specifically about GPU revenues).
AMD's last major GPU launch took place at the end of 2011, with card availability somewhat scarce initially. With better availability during 2012 Q1, AMD's CFO stated during the earnings report:
Graphics segment revenue was $382 million, flat compared to the prior quarter. GPU revenue was up in a seasonally down quarter due to improved desktop GPU ASPs in the channel, offset by seasonally lower game console royalty revenue.
AMD increased GPU revenue in a seasonally down quarter, specifically due to having a refreshed line of GPUs available, and with these GPUs commanding higher ASPs.
Looking at CFO comments during the most recent quarter:
The first statement leads me to believe unit shipments were down as customers were waiting on new products. I believe GPU ASP declined as prices were lowered to clear inventory for the impending launch.
To tie this all together, because Nvidia has a bigger market share of notebook GPUs, desktop shipments become more meaningful for AMD, as that is where much of the company's GPU revenue is derived. AMD has had two quarters in a row of declining ASPs as the company has been competing with an aging line of GPUs, and has been lowering the price to clear inventory.
The R9-290X (and non-X version) represent important offerings from AMD, as they should help to serve raise the ASP of GPU shipments. The solid reviews the cards are receiving so far demonstrate the interest in the new GPUs. I believe at this point it will be more an issue of availability - meaning if AMD is able to meet demand for the cards they should sell well.
AMD actually powered the last generation of the Mac Pro, but given the age of the card (5000 series) that was used, the revenues generated for AMD were likely minimal.
Peeling back the layers reveals an extremely customized design, built specifically for acoustics and aesthetics. I bring this up because you will notice there is no place for an AIB. In order to mount the GPUs, Apple is likely purchasing silicon directly and has fabricated the rest of the design needed for mounting. This likely lowers the revenue AMD will generate on sales of the workstation cards, as AMD is not supplying the entire board.
The Mac Pro ships with dual AMD D300 GPUs. ExtremeTech writer Joel Hruska did some digging, and estimates the D300 GPUs are equivalent to AMD's W5000 GPUs, which retail for ~$400.
ZDnet provided the most detailed breakdown of Mac Pro sales I could find, but many of the numbers are based on the author's assumptions. Looking at the most recent time I could find a report by Apple which included Mac sales breakdowns:
"We sold over 4.9 million Macs establishing a new September quarter record. This represents 1% growth year-over-year compared to IDC's latest published estimate of 8% contraction of the global personal computer market in the September quarter.
Portable sales grew 9% year-over-year and represented an all time high of 80% of Mac unit mix, thanks to strong sales of MacBook Pro and MacBook Air."
Using a similar calculation to the one performed by the author of the ZDNet article, comments from Apple's Q4 2012 report indicate Apple shipped 980k desktop units. Assuming of these 980K unit shipments, assuming a range of 3% to 5% of these were Mac Pros means that ~30K-50K Mac Pro units were shipped during Apple's Q4 2012.
Given that $400 is the cheapest retail price of a single W5000, but the W5000 is the base model, I assume AMD revenues of $350 per Mac Pro unit sold. This should be on the conservative side, and account for volume discounts for Apple. Given the range of Mac Pro sales and my estimations of $350 per Mac, AMD could be looking at revenues in the range of $10M or so per quarter from this design win.
To put this in perspective, analysts cited a 6% sequential decline in CS (computing solution) revenues, equating to $50M during the most recent sell off of AMD. A $10M increase in revenues equates to 20% of the total decline AMD faced in their CS group in Q3.
A new GPU lineup and the Mac Pro design win represent great opportunities for AMD in Q4 and beyond.
The R9-290X and R9-290 will give AMD more expensive GPUs to ship, which will raise ASPs. The drop in demand for AMD's GPUs during Q3 could actually be a positive sign for Q4, as consumers could have potentially been waiting for the new flagship GPUs to upgrade. Now it is up to AMD to meet consumer demand.
The Mac Pro represents more of a steady stream of revenue since Apple seems to do pretty well even as the overall PC market is contracting. AMD maintained their spot in the Mac Pro design, but Apple doubled the GPU count in the machine. Contrast this to Nvidia's situation, in which Apple is using Intel's (INTC) Iris Pro graphics in select iMacs and MBPs, meaning Apple lost some sockets in the recent roll out of Apple's new products.
As the competition in the iGPU space heats up between AMD and Intel, this will only serve to squeeze Nvidia out of more sockets.
Looking at "Mantle" as an example, right now investors are wondering if this is a buzzword or something worth paying attention to. The truth is, who knows? But we will find out in December. The positive PR from the GPU launch is great news for investors. More positive PR from gamers finding out Mantle is more than a buzzword would be another big win. AMD has stated the company plans to more aggressively target the professional GPU market.
These developments are a solid positive for investors, but they only represent part of the story. A strong quarter from AMD's GVS segment in Q4 would show these new GPUs have clout. Nvidia's quarters are offset from AMD's somewhat, but next year if Nvidia reports weak financials during its holiday quarter, this could signal a positive for AMD investors depending on the reason for the weakness. Going forward, aggressive pricing and high performance will squeeze Nvidia at the top, with Intel and AMD simultaneously squeezing the budget section of the GPU market with more powerful integrated solutions.
Given the reaction to Q3 earnings, it is obvious the pundits following AMD care quite a bit about computer sales. To parallel AMD's CS and GVS segments, the company is looking at a similar situation going into Q4 as consumers could be waiting to upgrade until Kaveri is launched, similar to what caused declining GPU revenues during Q3.
I feel to take a short term stance about these new GPUs is the wrong view. Declining CS revenues could overshadow the effects of other positives AMD has going for the company during Q4, depending on how the market reacts to the report.
The longer term view is that strong Q4 GPU sales signal strength going into the first half of the year, which will coincide with AMD's Kaveri launch. If Kaveri is well received and AMD manages a strong quarter during Q1 or Q2 next year from both strong CPU and GPU sales, the $2.50 PT some bears have on AMD will likely need to be readjusted. But that is dependent on Kaveri performance and how well it is received, which we will not know more about until Q4 this year or early Q1 next year.
Additional disclosure: I am long AMD in both options and shares, and actively trade my position. I may add or liquidate shares/options in AMD at anytime. I am short NVDA via a very small number of puts.