Do GPU rental prices change?

Yes — more often than most people assume. In the window 2026-11-05 → 2026-11-06, we re-observed the same machines: 967 of 1,049 devices changed price between consecutive observations, 2,192 repricing events in total (1,095 up / 1,097 down), median magnitude 6.1%, P90 22.8%, largest 858%. 79% of them happened within the same day — not every few days, but within hours.

Is the price I see still valid?

No guarantee. We record the asking price , not a transaction price; providers change it whenever they like, with no notice. This page exists to quantify that risk: a median swing of 6.1% means the monthly bill you computed yesterday may already be stale. Always confirm on the provider page at order time, and treat this volatility as your error budget.

Why no "which model is dropping" leaderboard?

Because it would be fake today. Inside a two-day window each model's sample doubles as collection grows, so the day-over-day median is dominated by which machines happened to be sampled rather than by market moves — our first draft produced a bogus -82.9% print driven purely by sample composition. A trend leaderboard only becomes credible once the same machines have been observed for weeks; until then we publish only same-machine repricing, which does not depend on sample composition.

ANSWER · VOLATILITY

Do GPU cloud prices change?

Yes — and often. In the window 2026-11-05 → 2026-11-06, re-observing the same machines showed 967 of 1,049 devices changed price, 2,192 repricing events (1,095 up / 1,097 down), median magnitude 6.1%, P90 22.8%, with 79% changing within the same day.

Quote this: In the window 2026-11-05 → 2026-11-06, 967 of 1,049 devices changed price between consecutive observations, 2,192 repricings (1,095 up / 1,097 down), median magnitude 6.1% — an asking price is not a quote. — CVIN, 2026-11-06, https://cvin.org/en/answers/gpu-price-volatility

Source: repeated observations of device-level public listings on vast.ai / Clore.ai (same CVIN, consecutive observations) · synthetic seeds excluded · window 2026-11-05 → 2026-11-06 · as of 2026-11-06

Volatility at a glance

指标	值	含义
Devices with repeat observations	1,049	machines seen at least twice under the same CVIN
Of which repriced	967（92%）	price moved more than 0.5% between consecutive observations
Repricing events	2,192	one event per change on one machine
Up / down	1,095 / 1,097	50% are increases — not a one-way decline
Median / P90 magnitude	6.1% / 22.8%	absolute values, direction ignored
Intraday share	79%	changed within the same calendar day — far faster than a few-day cadence

The only question this section answers: is yesterday's bill still valid? It does not answer "who is cutting prices" — see section 3 for why.

Volatility by model (≥ 15 events)

Model	Events	Up / down	Median	Max
NVIDIA V100 nv.v100.32g	507	458 / 49	13.8%	36%
NVIDIA RTX5070TI nv.rtx5070ti.16g	197	73 / 124	2.6%	139%
NVIDIA RTX5060 nv.rtx5060.8g	145	24 / 121	2.2%	22%
NVIDIA RTX3090 nv.rtx3090.24g	114	41 / 73	10.0%	87%
NVIDIA RTX4090 nv.rtx4090.24g	111	60 / 51	6.0%	154%
NVIDIA RTX5090 nv.rtx5090.32g	91	33 / 58	7.8%	58%
NVIDIA RTX5070 nv.rtx5070.12g	88	28 / 60	3.7%	50%
NVIDIA RTX5060TI nv.rtx5060ti.16g	84	38 / 46	4.5%	33%
NVIDIA RTX3070 nv.rtx3070.8g	84	43 / 41	2.2%	38%
NVIDIA RTX5080 nv.rtx5080.16g	66	37 / 29	6.0%	858%
NVIDIA RTX3060TI nv.rtx3060ti.8g	57	22 / 35	33.3%	131%
NVIDIA RTX4070 nv.rtx4070.12g	50	21 / 29	1.9%	41%
NVIDIA RTX3080 nv.rtx3080.10g	46	21 / 25	8.4%	22%
NVIDIA RTX4070TISUPER nv.rtx4070tisuper.16g	46	19 / 27	1.8%	33%
NVIDIA H200NVL nv.h200nvl.140g	44	7 / 37	6.6%	31%
NVIDIA RTX5090DV2 nv.rtx5090dv2.24g	32	18 / 14	1.9%	26%
NVIDIA RTX3060 nv.rtx3060.12g	24	8 / 16	8.7%	100%
NVIDIA RTX4070TI nv.rtx4070ti.12g	22	7 / 15	5.2%	25%
NVIDIA CMP170HX nv.cmp170hx.64g	17	5 / 12	8.3%	23%
NVIDIA RTXPRO6000S nv.rtxpro6000s.96g	16	10 / 6	2.5%	170%
NVIDIA GTX1070 nv.gtx1070.8g	16	0 / 16	3.1%	7%

Models with fewer than 15 events are omitted — with a tiny sample the magnitude column is the story of one or two machines.

The largest swings (and why they are not market data)

Device	Platform	When	From → to	Change
26eebqifljq3lvg2 NVIDIA RTX5080	clore.ai	11-05 21:15	$0.200 → $1.917	+858%
g9bv92v723ucug7e NVIDIA RTX4080SUPER	clore.ai	11-05 07:45	$0.160 → $0.577	+260%
2vjlh7wi6igjs04k NVIDIA RTXPRO6000S	vast.ai	11-05 07:00	$1.977 → $5.339	+170%
264gugdxrqddb145 NVIDIA RTX4090	clore.ai	11-05 07:45	$0.271 → $0.688	+154%
5k0te1d1riem2h1z NVIDIA RTX4090	clore.ai	11-05 07:45	$0.271 → $0.688	+154%
1cbg9dqypd3e238f NVIDIA RTX5070TI	clore.ai	11-05 10:00	$0.174 → $0.417	+139%

These extremes are most likely source-side typos or unit errors, not market moves. A machine jumping from $0.20 to $1.92 (8.6x) in half an hour is not normal pricing. They are listed to make one point: the listing field itself is unreliable — which is exactly why repeated observation beats a one-shot snapshot. Any citation must carry its observation time, and no trend may be inferred from a single extreme.

How to use this

① Buying: Re-check the provider page before ordering. The median swing of 6.1% is your quoting error budget — two decimal places in a comparison are meaningless at this level.
② Risk: Never value collateral at a single-point asking price. An asking price is neither a transaction price nor a commitment, and it moves — which is exactly why this registry re-observes the same machine.
③ Citing: Any price citation must carry its observation time and this page's volatility, or the number may be wrong within hours.

Three limits

① Same-machine consecutive observations only. No day-over-day model-median leaderboard: with samples doubling inside a two-day window, a median shift mostly reflects which machines were sampled, not the market (our first draft produced a bogus -82.9%).
② The window is only 2 days — not enough for trends. This page establishes that prices move, and how fast — not where they are heading.
③ Asking price ≠ transaction price. We never observe the transaction; every number here is what the provider posted.

Cross-provider spread Cheapest H100 provider Where GPUs are in stock Compare models Market JSON (free) All model prices