Every price on this board is the lowest matching listing at one of the storefronts below, fetched on a schedule and recorded. Nothing is modeled, estimated or affiliate-weighted. Where a retailer refuses automated requests we say so and link you to their search instead of inventing a number.
12
sources
9
fetched live
32
parts tracked
91
snapshots recorded
Fetched server-side on every sweep. These are the storefronts whose numbers become the price series and the charts.
Marketplace
Official API · needs API keys
Official Browse API. The deepest pool of used and gray-market silicon — often the only place loose SXM modules and EOL datacenter cards trade at all. Needs free API keys.
Reseller
Fetched live
Refurbished datacenter and workstation specialist. Search ORs its terms, so queries are tuned per part; stock skews toward previous-generation Tesla cards.
Cheapest right now: $5,600 — PNY - NVIDIA Quadro - RTX 6000 ADA (48GB) - GDDR6 384 bit (VCNRTX6000ADA-PB)
Retailer
Official API · needs API keys
Official Products API — the storefront blocks servers, the API does not. Consumer cards only. Needs a free API key.
Retailer
Fetched live
Serves plain HTML server-side. The broadest single source here: consumer, workstation and even H100/H200 PCIe cards, with reliable stock signals.
Cheapest right now: $310 — ASRock Challenger Arc B580 12GB GDDR6 PCI Express 4.0 x8 ATX Graphics Card B580 CL 12GO
Retailer
Fetched live
Bay Area retailer on Magento. Consistently among the sharpest prices on current-generation consumer cards.
Cheapest right now: $300 — Asrock Intel Arc B580 STEEL LEGEND 12GB OC Graphic Card White
Reseller
Fetched live
Supermicro-authorized reseller on Odoo. One of the few places that lists real prices for H100 PCIe/NVL and A100 rather than quote-on-request.
Cheapest right now: $4,731 — NVIDIA PNY VCNRTXA6000-BLK Quadro RTXA6000 Graphics Card with 48GB GDDR6
Reseller
Fetched live
Refurbished enterprise hardware, read through Shopify's structured product JSON. Strong secondary-market signal for Tesla V100 and workstation Quadro cards.
Cheapest right now: $900 — Dell 0HR5WD Nvidia Tesla V100 32GB HBM2 250W PCIe 3.0 x16 FHHL Dual-Slot GPU
Reseller
Fetched live
Enterprise parts reseller, also read through Shopify's product JSON. Carries new and refurbished datacenter GPUs, frequently below the reseller average.
Cheapest right now: $669 — NVIDIA/HP Tesla V100 PG500-216 32GB HBM2 PCIe 3.0 x16 Refurbished Data Center AI Deep Learning HPC Accelerator GPU
Manufacturer
Fetched live
Supermicro's own e-store — the manufacturer selling direct. It lists no bare GPU cards, but it publishes real prices for complete GPU systems (8-way HGX H200/B300 and MI300X boxes), which almost nobody else in the channel does in public.
Cheapest right now: $60,683 — Supermicro 2U Gold Series GPU Server (SYS-212GB-FNR-01-G2)
These retailers refuse automated requests, so we never quote a price we can't verify. They still get a deep search link on every part they carry — often worth checking, especially in-store.
Retailer
Link only
In-store pricing is regularly the best in the country on consumer cards, but every request from a server — including robots.txt — is refused at the edge, so we link you straight to their search instead of quoting a price we can't fetch honestly.
Retailer
Link only
Blocks automated requests, so it gets a deep search link rather than a fetched price.
Marketplace
Link only
Blocks automated requests. Marketplace pricing swings hard on third-party sellers, so we link out rather than quote.
Step 1
Search
Each source gets the query shape its own search engine actually responds to — some want the full product name, some return nothing unless you send the bare model number.
Step 2
Match
A per-part title regex, plus an exclusion regex that throws out waterblocks, barebones servers, NVLink bridges and multi-unit lots.
Step 3
Sanity band
Anything priced outside a per-part floor and ceiling is not that part. A $12 “RTX 5090” is a bracket.
Step 4
Stock
Listings the source marks unavailable are dropped — a price you can't pay isn't a price.
Step 5
Outlier guard
A listing far below the median of its own matched set is treated as bait or a mispriced SKU, not as the day's low.
Step 6
Record
The survivor's price accumulates into the day's low and high for that source, so the chart shows the real daily floor rather than whatever the last sweep happened to catch.
The rental board is fed by a separate set of sources, and it has an easier time of it: none of these block automated requests, so there are no link-only entries. Five live feeds cover 16 named providers plus the smaller partner clouds behind them — AWS’s public pricing feed, Azure’s Retail Prices API, RunPod’s GraphQL endpoint, Vast.ai’s marketplace listings, and Shadeform’s catalog, which publishes live rates and per-region availability for a dozen-plus GPU clouds that have no public pricing API of their own. Rates are always attributed to the cloud that charges them.
Two things are static rather than fetched, and only two: how many GPUs each AWS instance type and Azure VM SKU carries. Neither hyperscaler publishes accelerator counts in its price feed, and a wrong count would silently scale every per-GPU rate — so shapes whose count isn’t unambiguous are skipped rather than guessed at.
AWS
direct feed
Microsoft Azure
direct feed
RunPod
direct feed
Vast.ai
direct feed
Lambda
Shadeform catalog
Crusoe
Shadeform catalog
Nebius
Shadeform catalog
Voltage Park
Shadeform catalog
Hyperstack
Shadeform catalog
Denvr Dataworks
Shadeform catalog
Massed Compute
Shadeform catalog
Latitude.sh
Shadeform catalog
Paperspace
Shadeform catalog
DigitalOcean
Shadeform catalog
Scaleway
Shadeform catalog
Vultr
Shadeform catalog