Winter 2026 is the first buying window in years where “buy now or wait” is not a question about your patience — there is nothing dated to wait for. Tom’s Hardware’s October 2026 buying guide puts it flatly: “With no new graphics card launches on the horizon, now is as good a time as ever to buy, especially as prices continue to rise.”
Our RTX 60 timing breakdown already owns the hold-out-for-the-next-generation question, and its conclusion — that a rumoured 2028 date is not a schedule — has not been overtaken. This piece carries the other half: the price index as fetched on 29 October 2026, the performance medians from Tom’s guide of 2 October 2026, the stack sorted by cost per frame, and the VRAMVRAM. Video RAM — memory on the graphics card used for textures and frames; more of it lets you run higher settings. floor printed on the winter window’s own store pages.
What this covers: US street prices as fetched on 29 October 2026 and performance medians from Tom’s guide of 2 October 2026, the GPUGPU. Graphics Processing Unit — the chip that renders the game's visuals; the main driver of framerate and image quality. stack ranked by frames per dollar at 1440p and 4K, the five supply signals worth tracking, and the stated requirements of games shipping between 6 October 2026 and 23 February 2027. Why it matters: the reported RTX 50 Super refresh is on hold and every RTX 50-series price in Tom’s Hardware’s own index rose between August and October, so the cost of waiting is now a measurable number. Who should care: anyone buying between now and February, and anyone holding a 6GB or 8GB card who needs to know which winter releases will refuse to run well on it.
Two dated snapshots sit behind the tables here. The “best US price” columns are Tom’s price index as fetched on 29 October 2026 — the lowest listing for an available card, live/retail branch, not pre-order, not regional exception. The median street column and the whole performance table come from Tom’s Hardware’s “Best Graphics Cards for Gaming in 2026”, last updated 2 October 2026, so the two were taken four weeks apart and are not the same measurement.
What the market actually is right now
The generation is not being replaced yet. The GeForce RTX 50 series and Radeon RX 9000 series launched in early 2025 and remain the current stack. The reported mid-cycle refresh — an RTX 5080 Super and RTX 5070 Ti Super at 24GB of GDDR7 on a 256-bit bus, an RTX 5070 Super at 18GB on 192-bit, and an RTX 5050 9GB — was reported on 18 July 2026 as “reportedly ready, but stuck in limbo due to excessive GDDR7 pricing.” That is a rumour resting on a VideoCardz report and one unnamed board partner; NVIDIA has announced none of those cards, and our earlier breakdown of that refresh still describes it correctly.
Prices are still climbing. On 29 October 2026 Tom’s Hardware reported the cheapest reputable in-stock RTX 5080 at $1,999 — a PNY card at Best Buy, i.e. exactly the RTX 5090’s on-paper MSRP — against the 5080’s $999 launch price, with RTX 5090 cards “found for as high as $12,000.”
Here is the same price index, same metric — Tom’s “best US price,” the lowest listing for an available card — at two dates three months apart:
| Card | Best US price, 3 Aug 2026 | Best US price, as fetched 29 Oct 2026 | Launch MSRP |
|---|---|---|---|
| GeForce RTX 5090 | $4,288 | $8,250 | $1,999 |
| GeForce RTX 5080 | $1,256 | $2,199 | $999 |
| GeForce RTX 5070 Ti | $949 | $1,249 | $749 |
| GeForce RTX 5070 | $629 | $899 | $549 |
| GeForce RTX 5060 Ti 16GB | $569 | $799 | $429 |
| GeForce RTX 5060 Ti 8GB | $429 | $529 | $379 |
| GeForce RTX 5060 | $339 | $449 | $299 |
| GeForce RTX 5050 | $289 | $419 | $249 |
A doubling at the top of the stack and between 23% and 45% everywhere else, in one quarter, measured by the same publisher on the same metric. AMD has moved less: the August survey found only single-digit percentage increases on RDNA 4 cards and noted “heavy emphasis on yet.”
The current stack, ranked by what it costs per frame
Tom’s Hardware’s best-GPU table, last updated 2 October 2026, is the most useful artefact in this market because it carries performance and price in one row: geometric-mean frame rates across 19 games with no upscaling or frame generation, paired with the median of the street prices its own survey recorded rather than MSRP.
| Card | 1080p | 1440p | 4K | Median street (2 Oct 2026) | Launch MSRP |
|---|---|---|---|---|---|
| GeForce RTX 5090 | 166.3 | 135.15 | 88.02 | $4,299 | $1,999 |
| GeForce RTX 5070 Ti | 123.7 | 92.0 | 52.8 | $1,099 | $749 |
| Radeon RX 9070 XT | 116.7 | 85.3 | 47.4 | $759 | $599 |
| GeForce RTX 5070 | 103.8 | 74.0 | 37.6 | $659 | $549 |
| Radeon RX 9070 | 103.4 | 74.8 | 41.1 | $634 | $549 |
| Radeon RX 9060 XT 16GB | 70.9 | 48.6 | 24.5 | $464 | $349 |
| GeForce RTX 5060 | 64.0 | 41.2 | 13.4 | $369 | $299 |
| GeForce RTX 5050 | 49.5 | 31.2 | 11.1 | $309 | $249 |
Those figures are a mixed raster and ray-traced index, not a prediction for any one game. Dividing price by them produces the most useful finding here:
| Card | Cost per 1440p frame | Cost per 4K frame |
|---|---|---|
| Radeon RX 9070 | $8.48 | $15.43 |
| Radeon RX 9070 XT | $8.90 | $16.01 |
| Radeon RX 9060 XT 16GB | $9.55 | $18.94 |
| GeForce RTX 5060 | $10.90 | $33.51 |
| GeForce RTX 5070 | $12.15 | $23.91 |
| GeForce RTX 5050 | $13.43 | $37.75 |
| GeForce RTX 5070 Ti | $13.58 | $23.66 |
| GeForce RTX 5090 | $61.04 | $93.73 |
Every GeForce row is divided from the 29 October best US price in the price index above; the three Radeon rows are divided from the 2 October medians, because the 29 October snapshot in this piece covers the GeForce cards only.
At 1440p the market is priced almost linearly: every non-flagship card lands between $8.48 and $13.58 per frame of index, with the flagship alone at more than four times the cheapest rate. At 4K the ratio becomes a real decision: the RX 9070 costs $15.43 per frame against the RTX 5090’s $93.73, so the flagship is roughly 6.1 times worse per frame there. That does not make the 5090 a bad card; it makes it a 4K card bought for a reason other than frames per dollar.
One apparent contradiction resolves once both tables are dated. The RTX 5090’s $4,299 is a 2 October median; the $8,250 is the lowest available listing on 29 October. A median cannot sit below the cheapest listing of the same card four weeks later, which is the tell that the two columns are not the same measurement. The same guide’s prose agrees with the newer reading — it says the card “is now more than four times its launch price”, and four times $1,999 is $7,996 — as does the $12,000 ceiling in the 29 October report. Treat the median column as a 2 October reference for the mid-range and a month stale at the top, which is why the cost-per-frame table prices every GeForce row at the 29 October best US price. The lower cost per frame is also not the whole bill: the same guide counts DLSS Ray Reconstruction in 30 games against AMD’s equivalent in two.
Five supply signals, and what each one changes
Five signals bear on a purchase in the next four months.
| Signal | What the source states | What it changes |
|---|---|---|
| 3GB GDDR7 pricing | The RTX 50 Super is “reportedly ready, but stuck in limbo,” with 3GB modules costing “twice or thrice as much” as 2GB (18 Jul 2026) | Removes the mid-cycle refresh as a reason to wait |
| GB202 reportedly diverted | A leaked factory notice, repeated by two leakers, says NVIDIA will stop making RTX 5090 boards and that “the highest-end GeForce graphics card will be the RTX 5080 24GB” (10 Oct 2026). Unverified | Halo-tier supply gets worse, not better |
| Micron ends 2GB GDDR7 | 28 and 32 Gbps 2GB GDDR7 modules marked End of Life; immediate shortage risk called minimal, since Samsung and SK hynix supply the bulk (24 Sep 2026) | 8GB and 16GB cards built on 2GB modules get less substitutable |
| Revenue per gigabyte | Gigabyte’s CEO described allocating by memory revenue per GB; $500/16GB works out at $32 against $50 for a $400 8GB card (16 Jan 2026) | Explains why high-VRAM mid-range SKUs are the scarcest |
| Memory shortage duration | SK hynix reportedly says 2027 will be the “worst year” and the crunch lasts until 2030 (reported 18 Jul 2026) | Argues against waiting for cheaper memory at any point in this build’s life |
The last row reframes the whole decision. If the shortage is a 2027 problem rather than a Q4 2026 one, “wait for prices to normalise before the winter games arrive” is a deferral into a worse market. The counter-argument is equally sourced: the August survey found RDNA 4 prices rising far more slowly, so a Radeon purchase is a partial hedge. One caution — the Gigabyte interview was published 16 January 2026 and describes a potential allocation strategy as its CEO characterised it, a model of why SKUs vanish rather than published policy.
The VRAM floor the winter window sets
This is where the winter window diverges from the story of the last three years. Fetching the store pages of games shipping between 6 October 2026 and 23 February 2027 does not produce a raised VRAM floor. It produces something more specific: the ray-tracing requirement has moved into the minimum spec, while the stated card class has barely moved.
| Release | Store-page date | Stated minimum graphics | Stated recommended graphics |
|---|---|---|---|
| Gears of War: E-Day | 6 Oct 2026 | “Hardware Ray Tracingray tracing. A rendering technique that simulates how light bounces, for realistic shadows, reflections, and lighting — costly to compute. capable GPU is required” — RTX 5050 / RTX 2060 / RX 6600 / RX 9060 | RTX 5060 / RTX 3060 Ti / RX 6700 XT / RX 9060 XT |
| Call of Duty: Modern Warfare 4 | 22 Oct 2026 | RX 470 / GTX 970 / GTX 1060 / Arc A580 | RX 6600 XT / RTX 3060 / Arc B580 |
| Phantom Blade Zero | 28 Oct 2026 | GTX 1660 6GB / RX 5500 XT 8GB | RTX 3060 Ti 8GB / RX 6700 XT 12GB |
| Crimson Desert Enhanced: Charting the Unknown | 29 Oct 2026 | RX 5500 XT / GTX 1060 | RX 6700 XT / RTX 2080 |
| Football Manager 27 | 10 Nov 2026 | 1 GiB VRAM, Shader Model 5.0 | 2 GiB VRAM |
| ANANTA | 15 Jan 2027 | Processor and graphics listed as undecided | Processor and graphics listed as undecided |
| Fable | 23 Feb 2027 | RX 6600 / RTX 5050 / RTX 2060 / Arc A580 | RX 6700 XT / RTX 3060 Ti |
Only Phantom Blade Zero’s page states the resolution its minimum targets — 1080p at 30 fpsframerate. How many images (frames) the game shows per second; higher = smoother motion. 60 fps is a common target. with upscaling, moving to 1440p at 60 for the recommended card.
Read that against the 8GB minimum-spec analysis from August. The floor is still an 8GB-class card from two generations back — an RX 6600 or an RTX 2060 — which the RTX 5060, the RX 9060 XT 16GB and the RTX 5050 all exceed. What changed is that Gears of War: E-Day requires a hardware ray-tracing capable GPU at minimum, an architectural line rather than a capacity one. An RTX 2060 or RX 6600 clears it; a GTX 1660 does not, however much VRAM it has.
Only two of those seven statements name a VRAM figure. The rest name card models, which is fine if you know them and useless if you are comparing capacities — which is why the VRAM tier analysis still matters more than any spec sheet, and why the 8GB-versus-16GB choice inside one SKU family stays the highest-leverage call in the mid-range. The RTX 5060 Ti’s two variants are the same GPU with a $270 gap at current pricing.
ANANTA is the honest case: it ships on 15 January 2027 and states nothing about processor or graphics. A recommendation built on its specs today would be invented.
The decision
| Your situation | What the evidence says | The call |
|---|---|---|
| Card works, you play at 1080p or 1440p | No dated replacement; prices rising; RTX 50 Super reportedly on hold | Keep it. There is no cheaper window visible in the sourced record |
| 6GB or 8GB card, planning on Gears of War: E-Day | Minimum spec requires hardware ray tracing | Check the architecture before the capacity. A 6GB RTX 2060 qualifies; a 6GB GTX 1660 does not |
| Buying at 1080p today | 8GB-class cards clear every minimum above that does not demand hardware ray tracing | An RTX 5060 at a $449 best US price is the cheapest correct answer |
| Buying at 1440p today | Pricing is near-linear across the non-flagship stack, from $8.48 to $13.58 per 1440p frame | Buy the cheapest card whose absolute 1440p figure clears your monitor, plus headroom |
| Buying at 4K today | RX 9070 at $15.43 per 4K frame against the 5090’s $93.73 | Price the 9070 and the 5070 Ti before the flagship; go flagship only for path tracing |
| Waiting for the next thing | RTX 50 Super reported on hold over 3GB GDDR7 cost; RTX 60 rumoured late 2027 or 2028, with no announcement for either | Do not hold money for an unannounced product when a three-month wait cost 23–45% across this stack |
The logic under every row is arithmetic, not sentiment: the cost of waiting has been positive for months on NVIDIA’s side and no announced product has a date that could reverse it. That is the opposite of 2024, when a refresh was imminent and patience paid.
What is not confirmed, stated plainly
- The RTX 50 Super lineup and its hold. Rumour, from a VideoCardz report and one board partner, published 18 July 2026. The 24GB/18GB/9GB configurations and 256-bit/192-bit bus widths are leak detail, not specification.
- The RTX 5090 halt and the RTX 5080 24GB. A leaked factory notice quoted by two leakers (10 October 2026), naming the 5080 24GB as the future flagship, with no NVIDIA confirmation — and headlines on the same site show NVIDIA denying the card is dead, which flags the halt as contested.
- Whether AMD’s prices stay lower. The August survey found single-digit RDNA 4 increases and explicitly flagged “yet.”
- The RTX 60 date. Rumoured; covered separately and in more depth.
Confirmed and dated: the median street prices and performance index from Tom’s guide of 2 October 2026, the 29 October RTX 5080 reporting and price-index snapshot, the 3 August price-index snapshot, and Micron’s 2GB GDDR7 end-of-life.