Launch MSRPs, release cycles and deals windows as of November 9, 2026.
This is a cross-brand comparison between the AMD RX 9070 (RX 9000, High-end, current generation) and the NVIDIA RTX 4090 (RTX 4000, Enthusiast, superseded by the NVIDIA RTX 5090), released 2 years 4 months apart. Against the RTX 4090, the RX 9070 costs $1,050 less at launch MSRP ($549 vs $1,599, 66% cheaper), has 8 GB less VRAM (16 GB vs 24 GB) and draws 200W less at rated TDP (250W vs 450W).
Mid-cycle — next generation may be on the horizon
Best for: 1440p gamers who want 16GB VRAM and AMD's ecosystem at the most efficient power-to-performance ratio.
Full details →This GPU is no longer the current generation. It has been replaced by the NVIDIA RTX 5090.
Superseded by RTX 5090
Best for: 4K enthusiasts who can find a well-priced used RTX 4090 and don't need DLSS 4 or Blackwell's efficiency improvements.
Full details →| AMD RX 9070 | NVIDIA RTX 4090 | |
|---|---|---|
| Tier | High-end | Enthusiast |
| Generation | RX 9000 | RTX 4000 |
| VRAM | 16 GB | 24 GB |
| Memory Type | GDDR6 | GDDR6X |
| TDP | 250W | 450W |
| Upscaling | FSR4 | DLSS3 |
| Ray Tracing | ✅ | ✅ |
| Launch MSRP | $549 | $1599 |
| Editor score | 8.7/10 | 9.0/10 |
| Released | Mar 6, 2025 | Oct 12, 2022 |
| Cycle length | ~820 days | ~840 days |
| Cycle advice | Caution | Superseded |
| Deals advice | Caution | Clearance |
| Successor | — | RTX 5090 |
Same VRAM as the 9070 XT for $50 less — excellent value.
More power-efficient than both the 9070 XT and RTX 5070.
Full access to AMD's ML-powered upscaling.
More VRAM than the RTX 5080 (16GB), relevant for AI workloads and 4K texture packs even in the Blackwell era.
New retail units are scarce and priced above MSRP. The used market (eBay, local classifieds) is the only viable path to value.
Over 3 years of driver optimizations make this one of the most stable GPUs available.
Released Mar 6, 2025, the RX 9070 has been on the market for 20 months. The previous cycle for this line ran about 820 days, so it is roughly 75% of the way through a typical cycle, with around 207 days to go before a replacement would be due on past form. Our model rates it Caution.
Released Oct 12, 2022, the RTX 4090 has been on the market for 4 years. It has been superseded by the NVIDIA RTX 5090, which launched Jan 30, 2025, so it's no longer a current-generation buy — new stock is clearance, and the rest of the market is second-hand.
On timing, the RX 9070 is the only current-generation option of the two (Caution), while the RTX 4090 is a clearance or used purchase.
At launch MSRP the RX 9070 is $1,050 cheaper than the RTX 4090 ($549 vs $1,599, 66% less). Per gigabyte of VRAM that's $34.3 for the RX 9070 and $66.6 for the RTX 4090 — the RX 9070 gives more memory for the money. Launch MSRP is only a reference point for a superseded card; its real price is whatever clearance or used stock sells for today.
RX 9070 (deals window to Dec 5, 2026, rated Caution): Black Friday / Cyber Monday 2026 — prices climbed again heading into November, to roughly $780–$880 (40–60% above the $549 MSRP), and a fourth 2026 AMD price increase has been reported for November. Discounts are unlikely to reach MSRP; only buy at or near $549.
RTX 4090 (deals window to Jun 30, 2027, rated Wait): New units are listed above $4,400 as of October 2026 and used units sell around $2,500–$3,000 — far above the $1,599 MSRP. Only consider a used unit if it comes well below current RTX 5090 street prices. Otherwise, the RTX 5090 at MSRP is better value.
Next on the deals calendar: for the RX 9070, New Year clearance 2027, from Jan 2, 2027 to Jan 20, 2027.
Our pick: AMD RX 9070
Our pick is the RX 9070: it's the current-generation card here, while the RTX 4090 has been superseded by the NVIDIA RTX 5090, and it also costs $1,050 less at launch MSRP ($549 vs $1,599) and draws 200W less at rated TDP (250W vs 450W).
Go for the RTX 4090 instead if you want more VRAM: 24 GB vs 16 GB, or if you want newer GDDR6X memory rather than GDDR6.
Buy/Wait and deals ratings come from release-cycle and pricing data, not benchmarks. How we rate →