Summary of "Radeon Vega Frontier Edition LIVE Benchmarking"
Overview
Live benchmarking of the Radeon Vega Frontier Edition (Vega FE) was performed on PCPer’s GPU testbed. The stream covered driver behavior, test methodology, game and professional benchmarks, power/noise/thermal measurements, and ad-hoc mining/OpenCL attempts. Many tests used lossless capture plus frame‑rating analysis.
Test platform / hardware / software
- Testbed: Intel i7-5960X on X99, 16 GB RAM, stock CPU, Corsair AX1500i PSU (no power limitations).
- GPU: Radeon Vega Frontier Edition — 4096 shaders, 16 GB HBM2 (Micron), air-cooled sample (water-cooled variant also exists).
- Driver: Vega FE driver from Radeon site (driver branch was forked earlier).
- Capture chain: GPU output → capture card + monitor/4K TV → capture PC (lossless capture) → frame‑rating analysis tool (based on NVIDIA FCAT concepts).
- Power instrumentation: card-rail measurements via National Instruments DAQ (USB-6289 + LabVIEW/SignalExpress). Card-level power was measured (not just wall power).
Benchmark suite (settings and titles)
Typical resolution: 2560×1440 (also ran 4K 3840×2160). Most runs used Ultra presets, vsync off, and many tests used 4× MSAA.
- Synthetic/benchmarks:
- 3DMark Fire Strike / Extreme, Time Spy
- Unigine Heaven, Superposition
- Games (main tests):
- Dirt Rally, Fallout 4, Hitman (2016), GTA V, The Witcher 3, Rise of the Tomb Raider
- Pro / productivity:
- LuxMark, Cinebench R15, SPECviewperf (planned / queued)
- Mining / OpenCL:
- Claymore, SGminer, Ethminer, F‑miner, NiceHash/minergate, etc. (many failed or produced poor results on this driver/sample)
Key technical observations & measurements
Power / TDP
- Vega FE spec TDP shown as 300 W.
- Measured gaming power typically ~280 W (card total usually close to but under 300 W).
- Observed repeated step-like drops in card power during long workloads (e.g., Witcher 3 / Metro 4K); discrete steps down of roughly 40 W increments (examples: ~280 → 240 → 200 → 160 W). These steps correlated with frame-rate shifts.
- PCIe slot draw ≈ 25 W; each 8‑pin PCIe connector measured ≈ 120–135 W under load.
- Forcing the fan to 100% reduced the power-toggling behavior and kept power nearer 300 W, indicating thermal/thermal-target or fan-curve interactions may influence power limits.
Clocks & temperatures
- GPUZ reported idle clocks ≈ 852 MHz (higher than many other cards’ idle clocks).
- Observed steady gaming clocks around ~1348–1528 MHz (1440 MHz common as a settling point). 3DMark sometimes reported 1600 MHz, but that clock was not sustained in real gaming runs at stock settings.
- Temperatures stabilized around ~83–84 °C in gaming runs.
Driver / GUI
- AMD’s statement to the testers: “gaming mode” vs “pro mode” in Radeon settings is primarily a GUI/setting selection; AMD claimed no performance delta between modes for this product.
- The Vega FE driver appears to be a forked branch — no further game optimizations were expected between this driver and the RX Vega launch; RX Vega driver improvements might add modest gains (0–8–10% in select titles).
- Some monitoring tools had incomplete compatibility. GPUZ’s 1 s sampling resolution limits visibility into fast clock/behavior changes.
Memory / HBCC
- Vega FE includes HBCC (High Bandwidth Cache Controller) and 16 GB HBM2.
- Host expectation: HBCC benefits for gaming on the 16 GB card are minimal now but could be more important if a future consumer SKU uses 8 GB HBM2.
Cooling variants
- The air-cooled FE sample did not reach the highest advertised clocks in sustained loads. Water‑cooled FE or improved cooling might enable higher sustained clocks and power.
Performance summary (gaming)
Overall, Vega FE (air-cooled, stock) typically placed between a GTX 1070 and GTX 1080 at 1440p — often closer to the GTX 1070 and below the GTX 1080. Titan XP remained faster in most gaming runs.
- Dirt Rally: Vega FE ≈ 100 FPS (1440p); Titan XP noticeably higher.
- Fallout 4: some runs were surprisingly competitive (host planned to re-run to verify).
- GTA V, Witcher 3, Rise of the Tomb Raider, Hitman: Vega FE generally below GTX 1080; many tests showed performance in the GTX 1070–1080 class.
- 4K gaming: mid-to-low 40 FPS in very high settings for Witcher 3 and other heavy titles; generally under GTX 1080.
- Synthetic: 3DMark Fire Strike GPU scores around the mid‑10k range (lower than some early leaked numbers circulating online).
Professional / compute benchmarks
- LuxMark (OpenCL): Vega FE ≈ 4690 (single‑GPU FE); Titan XP ≈ 5877 in their runs.
- Vega FE was ~40% higher than a Radeon Pro Duo (one Fury X-equivalent) result on LuxMark, but lower than Titan XP.
- Cinebench / SPECviewperf:
- Some OpenGL/Cinebench GL tests looked competitive; host referenced beating Titan XP in some Cinebench OpenGL runs, but results were mixed.
- OpenCL / miners:
- Many mining applications failed or produced errors (OpenCL build errors, “cannot create DAG” for GPU0).
- Some tools (e.g., F‑miner) worked but produced low hash rates (~30 MH/s or lower on some algorithms).
- Overall miner support on this driver/sample was immature, so mining competitiveness was poor or unreliable.
Testing methodology notes
- Capture-based testing: 60‑second capture windows per run; typically two runs per title.
- Frame‑rating analysis captured frame times, dropped frames, runts, and related metrics.
- All stock, out-of-box testing was done first (no power-limit changes or overclocking) to capture reference retail behavior.
- Planned follow-ups included fan-curve tuning, raising power limits, re-running Fallout 4, and running more pro tests (SPECviewperf).
Operational / practical notes from the stream
- Many miners/benchmarks required extra OpenCL runtime/SDK tweaks (pagefile size changes, copying DLLs). The stream included ad‑hoc troubleshooting.
- Typical live-stream issues occurred: capture hiccups, background noise, long installs, some crashes/blue screens when running certain tools.
- Host emphasized they would re-run and validate suspect results (Fallout 4 retest, power/fan tuning) before finalizing the review.
Main conclusions (host)
- Vega FE (air-cooled sample, stock settings) did not show a large leap over GTX 1080 in gaming; performance often landed between GTX 1070 and GTX 1080 and below Titan XP.
- Power behavior showed unexpected step-like throttling/drops correlated with frame-rate steps. Increasing fan speed reduced this behavior, pointing to thermal or firmware/power-target interactions. A water-cooled variant might achieve higher sustained clocks/power.
- Driver maturity and miner/OpenCL support were incomplete at the time of testing. AMD may improve optimizations before RX Vega launch, but expected gains are modest in many cases.
- Vega FE appears well built and has strong pro/compute potential (HBM2 + 16 GB), but out-of-box gaming performance and miner support were not convincingly class-leading in these live tests.
Follow-ups / planned retests
- Re-run Fallout 4 to verify the anomalous result.
- Re-run with altered fan curve / higher thermal target and increased power limits to determine maximum sustained clocks and effects on power & performance.
- Run the full professional test suite (SPECviewperf, more LuxMark/Cinebench, Blender and other compute workloads).
- Re-test mining/OpenCL after installing official OpenCL runtime/SDK and after miner updates.
Main speakers / sources
- Primary host: Ryan (PCPer / PC Perspective livestream host; lead tester).
- Assistants / contributors: Ken, Alan.
- External context and statements from AMD representatives.
- Stream / outlet: PCPer / PC Perspective live stream (pcper.com).
Note: results came from a live early review session on a single Vega FE air‑cooled sample and the driver available during the stream. The host planned further controlled retesting and a full written review with finalized comparison charts.
Category
Technology
Share this summary
Is the summary off?
If you think the summary is inaccurate, you can reprocess it with the latest model.
Preparing reprocess...