Whenever AMD and Nvidia launch a new GPU family, there’s always at least a few weeks of uncertain pricing and limited availability. Early shortages and short-term higher prices don’t mean much in the grand scheme of things, which is why most review sites keep a casual eye on store shelves, but don’t stay laser-focused on the issue. Now that the 14nm refresh cycle is essentially complete (at least for now), it’s time to take stock of where the market is — and it’s not good.
More than three months after Nvidia officially kicked off the 14nm refresh cycle, GPU prices still running hot. While the particulars vary, many of the products AMD and Nvidia have introduced are sitting significantly above their suggested MSRP (Manufacturer’s Suggested Retail Price) or SEP (Suggested E-Tail Pricing) price point. In some cases, this price-warping is enough to require rethinking whether buy a given GPU or switch to a different model.
All of the pricing data below is drawn from NowInStock.net. Because NIS is an up-to-the-minute tracker, we cannot guarantee that the exact in-stock notifications we saw when writing this story will correspond to what you see when you read it. Our price data is based solely on cards listed as in-stock for all of the given GPU SKUs at the time of writing.
Note: GPUs with extensive aftermarket modification, like water cooling plates, were omitted from this comparison. These cards tend to be $100 to $200 more expensive than comparative air-cooled cards and would inflate our average price calculations. In a few cases, we were unable to compare pricing against MSRP/SEP because no public SEP for certain GPU RAM configurations has been published.
Because GPU manufacturers have always built a variety of SKUs to address various market segments, we expected to see GPU average selling prices that were above the MSRP / SEP. We’ve also included data on the best-case price for a given GPU to account for this. General availability has improved compared to earlier this summer; all of the GPUs we’ll discuss below had at least several SKUs in stock around the country. We’ll break things down by manufacturer first, then compare their relative positions.
AMD introduced three new GPUs this summer: RX 460, RX 470, and RX 480. SEP for the RX 460 was $109 (AMD never specified if this was for the 2GB or 4GB variant), the RX 470 4GB was supposed to be $179, and the RX 480 came in at $200 for a 4GB card and $239 for an 8GB card.
At the low-end, there’s a single 2GB RX 460 at $109, but given the RX 460’s performance, that’s really where the 4GB GPU ought to sit. Performance-wise, the RX 460 is generally faster than the GTX 750 Ti, but is slower than the GTX 950 in DX11 titles. The least expensive RX 460 4GB GPU we saw was $129, and the low-end GTX 950 at $139 is arguably a better value. At $109, a 4GB RX 460 has legs. At $129, it doesn’t. Average price on the 4GB RX 460 was $140, which puts it firmly in GTX 950 territory.
The RX 470 is supposed to sell for $179; the least expensive RX 470 we found was $199, which is where the RX 480 4GB is supposed to sit. The 8GB RX 480 is consistently about 13% faster than the 4GB RX 470 — knock off a few percent for the difference in memory speed, and the RX 480 4GB should be 7-10% faster than the RX 470 4GB that’s sitting in its price bracket. Average selling price for the RX 470 4GB variant was $207. There are 8GB variants of the RX 470 as well, but these are truly terrible deals with an ASP of $242.
The RX 480 is meant to sell for $200 (4GB) and $240 (8GB). The cheapest RX 480 we could find was $229 for a 4GB GPU; the cheapest RX 480 8GB GPU was $269. This is problematic for AMD even if we ignore Nvidia. The cheapest 8GB R9 390 can be had for $259 and is faster, on the whole, than the RX 480 (though power consumption is markedly higher).
The RX 480 4GB is supposed to be matching the GTX 1060 3GB at $200 and slipping below the GTX 1060 at $240. Instead, the RX 470 4GB is matching the GTX 1060 3GB, a comparison which does AMD no favors. Average price on the 8GB RX 480 is $279, 17% higher than normal.
Compared with its own previous 28nm GPUs, AMD’s 14nm cards still typically offer significantly better performance at the same price point. The RX 470 4GB, for example, is markedly faster than the R9 380 or R9 380X. We’ll examine the competitive standings in greater detail once we break down Nvidia’s results.
Nvidia began its refresh cycle in late May with the launch of the GTX 1080 and continued it throughout the summer with the GTX 1070 and 1060 (this last is available in 3GB and 6GB flavors). Unlike AMD, Nvidia opted for a standard top-down refresh cycle. The net effect of this is that Pascal almost always offers better performance/dollar ratios than the old Maxwell cards did, even after accounting for price inflation. Prices, however, are still inflated compared with what Nvidia told us to expect three months ago.
MSRP on the GTX 1080 was set to $599, while the Founder’s Edition variant of the GPU had an MSRP of $699. More than three months after launch, only the FE card variants have hit their $700 MSRPs (Zotac even has a Founder’s Edition GPU for $676). Asus has several SKUs in the $619 to $629 range, but no GTX 1080 GPU tracked by NIS has ever hit the $599 price point. A $619 list price is only 3% above list MSRP. But again — after three months we’d expect to see at least some GTX 1080 cards hitting their suggested retail price. Average price for the GTX 1080 was $668, 11% over MSRP.
The GTX 1070 has been on the market for 2.5 months and shows the worst price inflation relative to the rest of the Pascal family. The lowest price we found on a GTX 1070 is $409, $40 over the $379 MSRP. Again, the only Pascal GPUs that have hit their MSRPs are the Founder’s Edition cards at $449. The average price for a GTX 1070 is $441, 16% above the recommended MSRP.
Both the 3GB and 6GB variants of the GTX 1060 are also priced above their MSRPs on average. But there are a handful of 3GB and 6GB SKUs available for their listed prices of $199 and $249, respectively. The GTX 1060 is the only Pascal card that competes head-to-head against new 14nm GPUs from AMD and it’s also the only 14nm GPU that’s managed to hit its MSRP targets. This is unlikely to be a coincidence. Average price on the GTX 1060 3GB is $217 (there are no Founder’s Edition of this GPU), while average price on the 1060 6GB is $290.
Now that we’ve discussed the impact on AMD and Nvidia products throughout their specific price bands, let’s look at how this plays out across both companies. First, here’s a comparison of SEP/MSRP pricing in green versus actual average pricing in yellow. The RX 460 (4GB) and RX 470 (8GB) are missing green bars because AMD never released formal price information on where it expected those GPU configurations to land. The slides below will walk through three different ways of comparing AMD and Nvidia’s respective GPUs, and what each tells us.
As the slides above show, AMD’s RX 480 8GB prices have been pushed high enough to put that card in conflict with the R9 390 — and the R9 390 is faster, overall, than its ostensible replacement. Nvidia doesn’t have this problem, since the GTX 1070 and 1080 are still faster than the cards they replace, even at inflated MSRPs, but that doesn’t let Team Green off the hook. Three months after Nvidia launched the GTX 1080 with an MSRP of $599 and 2.5 months after the GTX 1070 debuted at $379, it’s impossible to buy either GPU at that price. Nor is this a new development — when we first looked at 14nm availability last June, we saw the GTX 1080 selling for between $800 to $900, while the GTX 1070 was $525 to $609. Prices were still significantly inflated when we checked in a month ago.
Once you account for how long it’s taken Pascal prices to approach their MSRPs, it’s clear Nvidia’s guidance back in May was blatantly false and AMD’s has been only slightly better. The most charitable read of the available data suggests that both AMD and Nvidia were taken off-guard by general 14nm yield issues at both TSMC and GlobalFoundries. Even if this is true, it doesn’t excuse launching parts at price points that the market couldn’t support — and the market clearly couldn’t support the list prices both AMD and Nvidia published.
It’s not clear that AMD and Nvidia are actually making any money off these increased prices. Typically, both companies sell both completed boards and standalone chips to the various AIBs. Completed boards typically rely on AMD or Nvidia’s reference cooler but carry MSI/Gigabyte/Asus-specific branding, while standalone chips are used for the AIB’s customized product lines. We don’t know anything about what AMD or Nvidia charge for these products, but we do know that AIB partners set their own prices for retail hardware.
When the cryptocurrency craze blew AMD’s GPU pricing into the stratosphere several years back, sources I spoke to at the company expressed deep frustration over the situation. AMD, you see, hadn’t raised its prices at all. The price inflation and profit-taking came courtesy of the various AIBs. Nvidia has a much stronger market position than AMD does and so we can’t draw firm conclusions about where the money is going. But the one thing we can say is that neither company has done an acceptable job of delivering volume hardware shipments at their stated MSRP / SEP.
As for why this is happening, we can hazard a guess, as we suggested above. We see elevated prices on both AMD and Nvidia cards, including segments where this negatively impacts AMD’s ability to compete against Nvidia. This suggests that both companies have seen lower-than-desired yields at 14nm. Overall availability is much better for all GPUs than it was when they launched and prices have been coming down. AMD’s prices are more distorted than Nvidia’s, but the RX family has been out less time than the 1080 or 1070. The fact that Nvidia’s prices are finally approaching their MSRP targets, however, doesn’t excuse the months of inflation and limited availability. Similarly, AMD’s Suggested E-Tail Pricing (SEP) appears to be just that — a suggestion with only a limited relationship to the facts of the situation on the ground.
These price discrepancies are more dangerous for AMD than Nvidia. The RX family has been effectively shifted one price bracket out of launch alignment. There’s not enough performance headroom in the various RX GPUs to warrant that positioning, either against lower-end Maxwell hardware or Nvidia’s GTX 1060.