AMD Radeon RX 9070 GRE 12GB Specs, Benchmarks & Pricing

The AMD Radeon RX 9070 GRE (Golden Rabbit Edition) is a mid-range gaming graphics card based on the RDNA 4 architecture using the Navi 48 GPU die. Originally launched in China on May 8, 2025, it received a global release on June 2, 2026 at $549 USD. The card features 48 compute units with 3,072 stream processors, 96 2nd-generation AI Accelerators, and 12 GB of GDDR6 memory on a 192-bit bus running at 18 Gbps, delivering 432 GB/s of memory bandwidth. It is positioned between the RX 9060 XT and the standard RX 9070 in AMD's lineup, offering a differentiated option with 12 GB VRAM (versus 16 GB on the RX 9070) on a narrower 192-bit bus at the same $549 price point as the RX 9070 at launch. Built on TSMC's N4P (4nm) process, the Navi 48 die measures 356.5 mm2 with 53.9 billion transistors. The boost clock reaches 2,790 MHz with a game clock of approximately 2,220 MHz, and total board power is 220 W. With 48 MB of 3rd-generation Infinity Cache and 48 3rd-generation Ray Tracing accelerators, the RX 9070 GRE delivers 34.3 TFLOPS of FP32 compute and 274 TOPS of dense INT8 AI performance. It supports FSR 4 (FidelityFX Super Resolution 4), AMD's machine-learning upscaling solution exclusive to RDNA 4, and uses a PCIe 5.0 x16 interface with DisplayPort 2.1a and HDMI 2.1b outputs.

  • Release Date: June 2, 2026
  • MSRP: $549 USD
  • GPU Architecture: RDNA 4
  • Hardware-Accelerated GEMM Operations:
    FP64 FP32 TF32 BF16 FP16 FP8 FP4 INT8 INT4
  • CUDA Compute Capability : n/a

Strengths

  • 12GB VRAM for model and data capacity

Considerations

  • Lower tensor core count compared to other GPUs

Specifications for AMD Radeon RX 9070 GRE

SpecificationPerformance Ranking
FP32 TFLOPs
52nd @ 34.3 TFLOPs (Mid Tier)(Mid)
FP16 TFLOPs
48th @ 137 TFLOPs (Entry Tier)(Entry)
Tensor Core Count
18th @ 96 Cores
Memory Capacity (GB)
35th @ 12 GB (Entry Tier)(Entry)
Memory Bandwidth (GB/s)
32nd @ 432 GB/s (Entry Tier)(Entry)
Int8 TOPs
51st @ 274 TOPs (Mid Tier)(Mid)
FP8 TFLOPs
27th @ 274 TFLOPs (Entry Tier)(Entry)
FP4 TFLOPsData not available

Real-time AMD Radeon RX 9070 GRE GPU Prices

We're tracking 16 of the AMD Radeon RX 9070 GRE GPUs currently available for sale. The lowest average price is $598 (what's this?)
Buy Now

Compare Price/Performance to other GPUs

We track real-time prices of other GPUs too so that you can compare the price/performance of the AMD Radeon RX 9070 GRE GPU to other GPUs.
Compare GPU Price/Performance

Compare AMD Radeon RX 9070 GRE to Another GPU

Compare the AMD Radeon RX 9070 GRE directly to another GPU to see specs, benchmarks, and prices side-by-side.
Compare GPUs Side-by-Side

Price History

AMD Radeon RX 9070 GRE Price History

This chart tracks AMD Radeon RX 9070 GRE prices over the past 6 months based on eBay listings. The yellow line shows the average of the three lowest-priced listings each day—a useful indicator for finding deals.

Trend: Over this period, the lowest average price has been down 6.6%. The current lowest average is $598, compared to a typical $566 over the period. The lowest average price is calculated as the average of the three lowest-priced listings each day.

gpupoet.com

Product Identifiers

Available from 5 Partners (5 products)
Sapphire
SAPPHIRE PULSE AMD Radeon RX 9070 GRE OC 12GB GDDR6
11354-01-20G(sku)
XFX
XFX Swift AMD Radeon RX 9070 GRE Triple Fan Gaming Edition
RX-97GRE29BA(model number)
PowerColor
PowerColor Reaper AMD Radeon RX 9070 GRE 12GB GDDR6
RX9070GRE 12G-A(model number)
ASRock
ASRock Radeon RX 9070 GRE Steel Legend 12GB OC
RX9070GRE SLD 12GO(model number)
Gigabyte
GIGABYTE Radeon RX 9070 GRE Gaming OC 12G
GV-R907GREGAMING OC-12GD(model number)

References

Notes

  1. tensorCoreCount tracks the 2nd-generation AI Accelerators as "tensor cores". The RX 9070 GRE has 96 AI Accelerators (2 per CU x 48 CUs), confirmed by AMD official product page and Tom's Guide review.
  2. fp32TFLOPS of 34.3 is sourced directly from the AMD official product page (https://www.amd.com/en/products/graphics/desktops/radeon/9000-series/amd-radeon-rx-9070-gre.html). Cross-check: 3,072 stream processors x 2,790 MHz boost x 4 (RDNA 4 dual-issue packed factor) = 34.3 TFLOPS.
  3. fp16TFLOPS of 137 represents FP16 Matrix throughput (dense) via the 2nd-gen AI Accelerators (WMMA), as published on the AMD product page (Peak Half Precision (FP16 Matrix) Performance = 137 TFLOPS; with 2:4 structured sparsity = 274 TFLOPS). This is the cross-vendor analog of NVIDIA tensor-core FP16. AMD also publishes FP16 Vector = FP32 Vector = 34.3 TFLOPS (SIMD path); that vector value is NOT used here because it understates AI throughput vs. the matrix path. The existing AMD RDNA 3/4 sibling files in this repo store fp16TFLOPS = 2x FP32 (packed-FP16 architectural capability); those files need migration to the matrix value — see project policy in .claude/agents/gpu-researcher.md.
  4. int8TOPS of 274 represents the non-sparse (dense) INT8 matrix performance per AMD official product page (https://www.amd.com/en/products/graphics/desktops/radeon/9000-series/amd-radeon-rx-9070-gre.html). The AMD product page also lists 549 TOPS with structured sparsity (2x dense), confirming the dense value is 274 TOPS. Dense value used for consistent cross-vendor comparison per project convention.
  5. fp8TFLOPS of 274 is the peak FP8 Matrix (dense, no structured sparsity) performance per the AMD official product page (https://www.amd.com/en/products/graphics/desktops/radeon/9000-series/amd-radeon-rx-9070-gre.html), which lists "FP8 Peak (Matrix) Performance" = 274 TFLOPS and "FP8 Peak (Matrix, Structured Sparsity)" = 549 TFLOPS. Dense value used for consistent cross-vendor comparison. This equals both int8TOPS (274) and 2x fp16TFLOPS (137), consistent with AMD's halving-bitwidth-doubles-throughput Matrix Core design pattern confirmed across all six RDNA 4 SKUs audited on 2026-09-02 (RX 9070 GRE, AI PRO R9700, RX 9070 XT, RX 9070, RX 9060 XT, RX 9060).
  6. fp4TFLOPS is intentionally omitted (not set to null-with-unverifiedSpecs) because RDNA 4's AMD official product page lists no FP4 Matrix performance row, and no FP4 WMMA input format is documented in the RDNA4 Instruction Set Architecture Reference Guide. This is a confirmed hardware negative (RDNA 4 has no FP4 support), not a research gap, so unverifiedSpecs does not apply per project convention.
  7. Phase 5 audit (2026-09-02): confirmed via direct fetch of AMD's official product pages that FP8 Matrix performance is published for all six RDNA 4 desktop/workstation SKUs (RX 9070 GRE, AI PRO R9700, RX 9070 XT, RX 9070, RX 9060 XT, RX 9060), spanning both the Navi 48 die (GRE, R9700, 9070, 9070 XT) and the Navi 44 die (9060, 9060 XT). This confirms FP8 WMMA support is an architecture-wide (GFX12/RDNA 4) Compute Unit capability, not a die-specific or SKU-specific feature. The prior absence of FP8 and fp8TFLOPS from the RX 9070 XT/RX 9070/RX 9060/RX 9060 XT spec files was a data-entry gap in those files (they were not checked against AMD's FP8 Matrix spec row at the time), not a real hardware difference from this GRE file. See the canonical RDNA 4 table in .claude/agents/gpu-researcher.md for the full determination.
  8. Memory bandwidth of 432 GB/s is calculated as 192-bit interface x 18 Gbps = 432 GB/s, confirmed by AMD official product page and multiple third-party reviews (Wikipedia, pcvenus.com, guru3d.com).
  9. Infinity Cache is 48 MB (3rd-generation), confirmed by AMD official product page and Tom's Hardware review. This is smaller than the 64 MB on the RX 9070 (256-bit bus), proportional to the 192-bit bus width.
  10. TDP is 220 W Total Board Power per AMD official product page. Some AIB partner cards (e.g., Sapphire PULSE OC) may have higher factory power limits. HotHardware notes up to 375 W available to the card in boosted scenarios.
  11. Process node is TSMC N4P (4nm) with the Navi 48 GPU die, measuring 356.5 mm2 with 53.9 billion transistors per Tom's Hardware review. The Navi 48 die is the same die used in the RX 9070 and RX 9070 XT, with 16 of 64 CUs disabled (48 active CUs vs 56 for RX 9070).
  12. Clock speeds: reference boost clock is 2,790 MHz and game clock is approximately 2,220 MHz per Wikipedia RX 9000 series specifications table. AIB OC models may have higher boost clocks (e.g., Sapphire PULSE OC reaches 2,920 MHz).
  13. Release date is 2026-06-02 (global launch). The card originally launched in China on May 8, 2025 as a China-exclusive product. AMD announced the global launch at Computex 2026, with worldwide availability beginning June 2, 2026 at $549 USD.
  14. MSRP of $549 USD verified from 3 sources: Tom's Hardware (https://www.tomshardware.com/pc-components/gpus/amds-formerly-china-exclusive-radeon-rx-9070-gre-goes-global-for-usd549-on-june-2-rdna-4-gpu-will-bridge-the-gap-between-rx-9060-xt-and-rx-9070), Tom's Guide review (https://www.tomsguide.com/computing/gpus/amd-radeon-rx-9070-gre-review), and HotHardware review (https://hothardware.com/reviews/amd-radeon-rx-9070-gre-review). All three independently confirm the $549 USD launch price.
  15. The RX 9070 GRE is positioned between the RX 9060 XT and the RX 9070 in AMD's lineup. At $549 it matches the RX 9070's launch price but offers 12 GB on a 192-bit bus versus the RX 9070's 16 GB on a 256-bit bus. Internal AMD benchmarks show approximately 21% higher average performance over the RTX 5060 Ti 16 GB at 1440p.
  16. RDNA 4 introduces optimized Compute Units, 3rd-generation Ray Tracing accelerators, and 2nd-generation AI Accelerators. FSR 4 is exclusive to RDNA 4 GPUs and leverages the AI accelerators for ML-based upscaling.
  17. manufacturerIdentifiers is empty because AMD OPN (Ordering Part Number) codes for consumer GPUs are typically not published in consumer-facing documentation. These internal B2B identifiers are not available through AMD's public-facing resources.
  18. thirdPartyProducts includes verified AIB partner model numbers from Sapphire (11354-01-20G for PULSE OC, confirmed from Sapphire official product page search snippet), XFX (RX-97GRE29BA for Swift Triple Fan, confirmed from xfxforce.com official product page), PowerColor (RX9070GRE 12G-A for Reaper, confirmed from eBay and nanokomputer.com listing), ASRock (RX9070GRE SLD 12GO for Steel Legend OC, confirmed from Newegg listing), and Gigabyte (GV-R907GREGAMING OC-12GD for Gaming OC, confirmed from Newegg listing). All model numbers sourced from https://www.newegg.com/p/pl?d=RX+9070+GRE and manufacturer product pages.
  19. FP8 (8-bit floating point) is listed in supportedHardwareOperations because the RDNA 4 2nd-generation AI Accelerators provide hardware FP8 matrix compute support, confirmed by the AMD official product page listing FP8 Peak (Matrix) = 274 TFLOPS and FP8 Peak (Matrix, Structured Sparsity) = 549 TFLOPS (https://www.amd.com/en/products/graphics/desktops/radeon/9000-series/amd-radeon-rx-9070-gre.html).
This site contains affiliate links. As an eBay Partner and Amazon Associate, gpupoet.com may be compensated if you make a purchase, at no cost to you. Learn more.