What's inside
- CY15 in one line: flat MSRPs, rising street prices
- Supply: 28nm was everywhere, the flagship tier was not
- Demand: gaming plus the first mining wave
- Retailer markup: where the extra 5–15% came from
- Memory costs: GDDR5 fell, HBM1 did not
- Worked example: what a “rising” price actually cost you
- Buying timing: match your situation to an action
- Ownership realities: what fails first on a CY15 card
- Frequently Asked Questions
- CY15 in one line: flat MSRPs, rising street prices
- Supply: 28nm was everywhere, the flagship tier was not
- Demand: gaming plus the first mining wave
- Retailer markup: where the extra 5–15% came from
- Memory costs: GDDR5 fell, HBM1 did not
- Worked example: what a “rising” price actually cost you
- Buying timing: match your situation to an action
- Ownership realities: what fails first on a CY15 card
- Frequently Asked Questions
As an Amazon Associate I earn from qualifying purchases. This post may contain affiliate links at no extra cost to you.
Nvidia’s suggested prices barely moved during CY15 — calendar year 2015 — yet the price buyers actually paid for cards like the GTX 970 and GTX 980 Ti rose roughly 10–25% in many markets between January and December of that year. The surge was not one event but four pressures stacking at once: tight supply of the new flagship tier, a demand bump from gaming plus the first Ethereum-driven GPU mining wave, retailer and marketplace markup layered over thin channel stock, and a memory bill that split sharply between cheap GDDR5 and expensive first-generation HBM. Working out which of those four you were actually paying for is what tells you whether to buy immediately or wait.
As an Amazon Associate we earn from qualifying purchases at no extra cost to you.
CY15 in one line: flat MSRPs, rising street prices
“CY15” is simply shorthand for calendar year 2015. The distinction that matters is between the manufacturer’s suggested retail price — set once at launch and rarely changed — and the street price, which is whatever the card was actually selling for that week. In CY15 the first number was flat and the second number climbed.
The table below gives general market ranges for the cards that defined the year, based on launch MSRPs from the manufacturers’ own press materials and typical retail ranges reported through the channel. Treat them as bands, not quotes.
| Card | Launch | Launch MSRP (USD) | Typical US street, mid-CY15 | Typical non-US street (FX + VAT) |
|---|---|---|---|---|
| GeForce GTX 750 Ti | Feb 2014 | ~$149 | $130–$165 | $175–$215 |
| GeForce GTX 960 | Jan 2015 | ~$199 | $190–$225 | $240–$295 |
| GeForce GTX 970 | Sep 2014 | ~$329 | $320–$365 | $400–$475 |
| GeForce GTX 980 | Sep 2014 | ~$549 | $530–$605 | $660–$785 |
| GeForce GTX 980 Ti | Jun 2015 | ~$649 | $650–$755 | $800–$955 |
| Radeon R9 390X | Jun 2015 | ~$429 | $420–$465 | $520–$605 |
| Radeon R9 Fury X (HBM1) | Jun 2015 | ~$649 | $650–$785 | $800–$980 |
Two things stand out. First, the US mid-tier held close to MSRP all year. Second, the flagship tier and nearly every non-US market drifted upward — which is the actual “surge” people remember.
Supply: 28nm was everywhere, the flagship tier was not
Nvidia’s Maxwell generation stayed on TSMC’s 28nm process, a node that had been in volume production since 2012. That means the mid-range was never genuinely supply-constrained in CY15 — GTX 960 and GTX 970 stock was healthy, and prices reflected competition rather than scarcity.
The flagship tier was a different problem. AMD’s Fury X used first-generation HBM memory stacked on a silicon interposer, a packaging step that only a small number of lines could perform at the time. Yields on that assembly were low through 2015, so Fury X volumes were thin from day one. Nvidia’s GTX 980 Ti sat on the same 28nm wafer as the GTX 980 but used a much larger die, which means far fewer chips per wafer and a lower share of fully working parts. When a $649 part is both the fastest card available and the hardest to manufacture, retailers have no reason to discount it — and every reason to sell it in a bundle.
Demand: gaming plus the first mining wave
Two demand curves met in CY15. The first was ordinary: the 900-series was a large efficiency jump over the previous generation, and 1080p high-refresh monitors were getting cheap, so a lot of people upgraded at once.
The second was new. Ethereum’s Frontier network went live in July 2015 according to the project’s own launch notes, and unlike Bitcoin’s earlier ASIC-dominated mining, Ethereum was deliberately designed to be memory-hard and GPU-friendly. Through the back half of 2015, mining demand was a secondary factor rather than the main one — it became the dominant driver in the 2017–2018 boom — but it was enough to keep the mid-tier cards that offered the best hash-per-dollar in short supply in several regions. That is the honest split: mining was a contributing cause in CY15, not the whole story.
Retailer markup: where the extra 5–15% came from
Graphics cards are a low-margin, high-return category for retailers. Typical gross margins sit in the single digits to low teens, and a card that gets returned, opened, or price-matched eats that margin entirely. In a year when a flagship was scarce, the levers retailers pulled were predictable:
- Bundle-only sales. The card is listed at MSRP but only sold with a power supply, a monitor, or an extended warranty, which moves the real cost up 10–20%.
- Marketplace resellers. Third-party sellers on large retail platforms list the same card above the first-party price and capture buyers who only see the “in stock” label.
- Pre-order and allocation pricing. Cards that arrive in small weekly batches get sold at whatever the market clears at, not at the launch price.
- Currency adjustment. Retailers outside the US reprice whenever the dollar moves sharply, and 2015 was a sharp year.
If you want to know whether a price is markup or a genuine increase, check three things: whether the card is sold standalone, whether the seller is the retailer or a marketplace third party, and whether the local currency has moved. If all three are clean and the price is still above launch MSRP, it is a real market price.
Memory costs: GDDR5 fell, HBM1 did not
Memory was the quiet variable. GDDR5 was a mature, multi-supplier part in 2015, and 8GB configurations arrived on AMD’s 390-series at the same price point where 4GB had sat a year earlier — memory cost per gigabyte was falling, which pushed mid-tier prices down, not up.
HBM1 went the other way. First-generation HBM gave Fury X a 4096-bit bus and enormous bandwidth, but it capped capacity at 4GB and required that expensive interposer packaging. Per gigabyte, HBM1 in 2015 cost several times what GDDR5 did, and that cost landed directly in the flagship price. So “memory costs pushed GPU prices up in CY15” is only half true: they pushed the HBM flagship up while pulling the GDDR5 mid-range down.
Worked example: what a “rising” price actually cost you
Take a Canadian buyer looking at a GTX 980 at $549 USD. In January 2015 the exchange rate was around 1.20 CAD per USD; by December it was near 1.38. Add 13% sales tax.
- January: $549 × 1.20 = $659, plus 13% tax = about $745 CAD
- December: $549 × 1.38 = $758, plus 13% tax = about $857 CAD
Nothing about the card changed. Nvidia did not raise the price. The buyer’s cost rose roughly 15% purely from currency, which is why so much of the CY15 “surge” was regional rather than global. The same maths applies in reverse to cost of ownership: a $330 card held for 36 months and resold for $110 works out to $220 net, or about $6.10 per month. A $649 flagship held 48 months and resold for $250 works out to $399 net, or about $8.30 per month. The flagship costs roughly a third more per month of ownership, which is the number worth comparing rather than the sticker.
Buying timing: match your situation to an action
| Your situation | Best move | Why | Timing window |
|---|---|---|---|
| 1080p / 60Hz, budget under $250 | GTX 960 or R9 380 class | Flagship frames are wasted on a 60Hz panel; VRAM and shader count are sufficient | 4–8 weeks after a rival launch, when the older tier is discounted |
| 1440p, high refresh | GTX 980 Ti class with 6GB | Wider memory bus and more VRAM hold frame times steadier at 1440p | Launch week — flagship allocation is largest at the start |
| Buying outside the US | Track local FX and VAT, not the US MSRP | Local price = MSRP × exchange rate × (1 + VAT); FX moved 15%+ during CY15 | Wait for a currency rally or a VAT-inclusive price reset |
| 4K or VR | Wait for the next node | Single-GPU 4K was not reliable in CY15, and HBM1’s 4GB was the ceiling | Skip the generation; the 16nm parts changed the maths |
| Upgrading from a 2GB card | Any mid-tier with 4GB or more | At 1080p ultra in 2015 titles, VRAM capacity was the binding constraint, not compute | Any time — no reason to wait for a node shrink |
Ownership realities: what fails first on a CY15 card
If you are looking at one of these cards used today, the failure order is consistent. The blower or axial fan bearing is usually first, typically after four to six years of daily use — it starts as a rattle at low RPM and progresses to a stall. Thermal paste pump-out is second; the original compound dries and cracks, and hotspot temperatures climb 10–15°C before core temperatures move at all. On cards that mined, the VRM and the memory modules are the weak points, because sustained load at elevated memory temperatures ages the thermal pads faster than gaming ever would. The GTX 970’s split 3.5GB + 0.5GB memory configuration, which was widely reported in early 2015, is also worth knowing about because it affected resale value and still shows up in how those cards behave when the last 512MB is addressed.
Repasting and replacing thermal pads is a roughly 45-minute job and costs very little in materials; a replacement fan assembly is the harder part, since stock of original parts for a decade-old card is thin and aftermarket coolers are the practical route.
Frequently Asked Questions
Did Nvidia actually raise its prices during CY15?
No. Launch MSRPs for the 900-series were set in late 2014 and early 2015 and were not revised upward during the year. What rose was the street price — the amount retailers and marketplace sellers charged — driven by flagship scarcity, currency movement, and demand. That distinction matters because it means the increase was reversible, and it largely reversed once supply normalised.
How much of the CY15 price rise was cryptocurrency mining?
A modest share. Ethereum’s network launched in July 2015, so mining demand only had half a year to build, and it concentrated on the mid-tier cards with the best hash-per-dollar rather than on flagships. The dominant mining-driven price surge came later, in 2017–2018. In CY15, currency movement and flagship supply explain more of the increase than mining does.
Why was the GTX 980 Ti so expensive at launch?
It used a large die on a mature 28nm process, so yields per wafer were lower and the fully working chips were scarce. It was also the fastest consumer card available at the time, which removed any competitive pressure to discount. Retailers responded by holding it near MSRP or selling it only in bundles.
Was HBM memory the reason flagship prices rose?
Partly. First-generation HBM cost substantially more per gigabyte than GDDR5 and required silicon interposer packaging with limited capacity, which is why Fury X volumes were thin and its price stayed high. At the same time, GDDR5 prices were falling, which is why 8GB mid-tier cards appeared at 4GB prices. Memory pushed the top of the market up and the middle of it down.
Should buyers have waited for the next generation instead?
For 1080p and 1440p buyers, no — the 900-series was already a large efficiency improvement and the wait cost a year of use. For 4K and VR buyers, yes: single-GPU 4K was not reliable in CY15 and the following 16nm generation changed performance-per-watt enough to be worth waiting for.
Is a used CY15 card still worth buying?
Only at a low price and only with realistic expectations. Assume the fan bearing and thermal paste need attention, budget for a repaste, and check whether the card was used for mining — sustained memory load leaves the most wear. Driver support for that generation has also wound down, so newer titles may not run optimally.