diff --git a/app/igneum-app/tiers/class-v5-tiers.json b/app/igneum-app/tiers/class-v5-tiers.json index 612accda4..93087e6db 100644 --- a/app/igneum-app/tiers/class-v5-tiers.json +++ b/app/igneum-app/tiers/class-v5-tiers.json @@ -1081,9 +1081,9 @@ "memory_gb": 24, "label": "estimated", "stock": { - "mhs": 28.0, - "w": 300.0, - "uj": 10.7, + "mhs": 40.0, + "w": 310.0, + "uj": 7.75, "class": "v4" }, "tiers": [ @@ -1094,13 +1094,13 @@ "mem_mhz": 0, "core_offset_mhz": -500, "power_offset_pct": -30, - "limit_w": 355, - "mhs": 28.0, - "w": 224.0, - "mhw": 0.125, + "limit_w": 310.0, + "mhs": 40.0, + "w": 235.6, + "mhw": 0.1698, "source": "measured", "label": "estimated", - "uj": 8.0, + "uj": 5.89, "note": "not rentable, not owned: the 9070 XT's dependent-read rate per channel (150 M per second per 32-bit channel) on 24 channels; the knob by the 9070 XT grid" }, { @@ -1110,13 +1110,13 @@ "mem_mhz": 0, "core_offset_mhz": -300, "power_offset_pct": -15, - "limit_w": 355, - "mhs": 28.0, - "w": 255.0, - "mhw": 0.1098, + "limit_w": 310.0, + "mhs": 40.0, + "w": 266.6, + "mhw": 0.15, "source": "measured", "label": "estimated", - "uj": 9.1, + "uj": 6.67, "note": "half the offsets" }, { @@ -1126,14 +1126,14 @@ "mem_mhz": 0, "core_offset_mhz": 0, "power_offset_pct": 0, - "limit_w": 355, - "mhs": 28.0, - "w": 300.0, - "mhw": 0.0933, + "limit_w": 310.0, + "mhs": 40.0, + "w": 310.0, + "mhw": 0.129, "source": "stock", "label": "estimated", - "uj": 10.7, - "note": "modelled" + "uj": 7.75, + "note": "24 channels at the RX 7600's measured per-channel rate: about 40 MH/s at about 310 W (7.8); not rentable, not owned" } ], "src": "modelled on the 9070 XT rows" @@ -1149,9 +1149,9 @@ "memory_gb": 16, "label": "estimated", "stock": { - "mhs": 18.0, + "mhs": 27.0, "w": 230.0, - "uj": 12.8, + "uj": 8.52, "class": "v4" }, "tiers": [ @@ -1162,13 +1162,13 @@ "mem_mhz": 0, "core_offset_mhz": -500, "power_offset_pct": -30, - "limit_w": 263, - "mhs": 18.0, - "w": 171.0, - "mhw": 0.1053, + "limit_w": 230.0, + "mhs": 27.0, + "w": 174.8, + "mhw": 0.1545, "source": "measured", "label": "estimated", - "uj": 9.5, + "uj": 6.47, "note": "16 channels of GDDR6 at the 9070 XT's per-channel rate; the knob by the 9070 XT grid" }, { @@ -1178,13 +1178,13 @@ "mem_mhz": 0, "core_offset_mhz": -300, "power_offset_pct": -15, - "limit_w": 263, - "mhs": 18.0, - "w": 195.0, - "mhw": 0.0923, + "limit_w": 230.0, + "mhs": 27.0, + "w": 197.8, + "mhw": 0.1365, "source": "measured", "label": "estimated", - "uj": 10.8, + "uj": 7.33, "note": "half the offsets" }, { @@ -1194,18 +1194,85 @@ "mem_mhz": 0, "core_offset_mhz": 0, "power_offset_pct": 0, - "limit_w": 263, - "mhs": 18.0, + "limit_w": 230.0, + "mhs": 27.0, "w": 230.0, - "mhw": 0.0783, + "mhw": 0.1174, "source": "stock", "label": "estimated", - "uj": 12.8, - "note": "modelled" + "uj": 8.52, + "note": "16 channels at the RX 7600's measured 1.74 MH/s per channel: about 27 MH/s at about 230 W (8.5); not rentable, not owned" } ], "src": "modelled on the 9070 XT rows" }, + { + "card": "AMD Radeon RX 7600", + "match": [ + "7600" + ], + "vendor": "amd", + "arch": "rdna3", + "memory_gb": 8, + "label": "estimated", + "stock": { + "mhs": 13.88, + "w": 113.0, + "uj": 8.14, + "class": "v4" + }, + "tiers": [ + { + "id": "efficiency", + "clock_mhz": 0, + "power_pct": 70, + "mem_mhz": 0, + "core_offset_mhz": -500, + "power_offset_pct": -30, + "limit_w": 113.0, + "mhs": 13.88, + "w": 85.9, + "mhw": 0.1616, + "source": "measured", + "label": "estimated", + "uj": 6.19, + "note": "the 9070 XT's ADLX grid (24 percent of the draw for no rate) applied; this card's own 24-point grid runs on PC 1 by about 16:50 UK and replaces this row" + }, + { + "id": "balanced", + "clock_mhz": 0, + "power_pct": 85, + "mem_mhz": 0, + "core_offset_mhz": -300, + "power_offset_pct": -15, + "limit_w": 113.0, + "mhs": 13.88, + "w": 97.2, + "mhw": 0.1428, + "source": "measured", + "label": "estimated", + "uj": 7.0, + "note": "half the offsets, inside the flat band" + }, + { + "id": "max", + "clock_mhz": 0, + "power_pct": 100, + "mem_mhz": 0, + "core_offset_mhz": 0, + "power_offset_pct": 0, + "limit_w": 113.0, + "mhs": 13.88, + "w": 113.0, + "mhw": 0.1228, + "source": "stock", + "label": "measured", + "uj": 8.14, + "note": "the founder's RX 7600 (8 GB) on PC 1, the card-in job 15:46 to 15:50 UK: class v4 sub-version 3 at stock 13.88 MH/s at 113 W (8.14), the v4 and v5 fingerprints equal to the pinned readings, the app's own row 13.4 at 113 W; class v5 within 2 percent; the 1 GiB dataset fits in 8,176 MB (the 2, 4 and 5.5 GiB rows follow)" + } + ], + "src": "the card-in lane's PC 1 job, 8 October 2026 15:50 UK" + }, { "card": "AMD Radeon RX 7600 XT", "match": [ @@ -1217,9 +1284,9 @@ "memory_gb": 16, "label": "estimated", "stock": { - "mhs": 9.0, - "w": 160.0, - "uj": 17.8, + "mhs": 13.9, + "w": 120.0, + "uj": 8.63, "class": "v4" }, "tiers": [ @@ -1230,13 +1297,13 @@ "mem_mhz": 0, "core_offset_mhz": -500, "power_offset_pct": -30, - "limit_w": 190, - "mhs": 9.0, - "w": 117.0, - "mhw": 0.0769, + "limit_w": 120.0, + "mhs": 13.9, + "w": 91.2, + "mhw": 0.1524, "source": "measured", "label": "estimated", - "uj": 13.0, + "uj": 6.56, "note": "8 channels; the card-in queue holds one for a PC measurement" }, { @@ -1246,13 +1313,13 @@ "mem_mhz": 0, "core_offset_mhz": -300, "power_offset_pct": -15, - "limit_w": 190, - "mhs": 9.0, - "w": 136.0, - "mhw": 0.0662, + "limit_w": 120.0, + "mhs": 13.9, + "w": 103.2, + "mhw": 0.1347, "source": "measured", "label": "estimated", - "uj": 15.1, + "uj": 7.42, "note": "half the offsets" }, { @@ -1262,14 +1329,14 @@ "mem_mhz": 0, "core_offset_mhz": 0, "power_offset_pct": 0, - "limit_w": 190, - "mhs": 9.0, - "w": 160.0, - "mhw": 0.0563, + "limit_w": 120.0, + "mhs": 13.9, + "w": 120.0, + "mhw": 0.1158, "source": "stock", "label": "estimated", - "uj": 17.8, - "note": "modelled" + "uj": 8.63, + "note": "the same Navi 33 die as the measured RX 7600 (13.88 MH/s at 113 W) with 16 GB in clamshell: the same rate, about 120 W (8.6); the card-in queue holds one" } ], "src": "modelled on the 9070 XT rows" @@ -1285,9 +1352,9 @@ "memory_gb": 16, "label": "estimated", "stock": { - "mhs": 9.5, + "mhs": 12.0, "w": 130.0, - "uj": 13.7, + "uj": 10.83, "class": "v4" }, "tiers": [ @@ -1298,13 +1365,13 @@ "mem_mhz": 0, "core_offset_mhz": -500, "power_offset_pct": -30, - "limit_w": 160, - "mhs": 9.5, - "w": 95.0, - "mhw": 0.1, + "limit_w": 130.0, + "mhs": 12.0, + "w": 98.8, + "mhw": 0.1215, "source": "measured", "label": "estimated", - "uj": 10.0, + "uj": 8.23, "note": "8 channels of GDDR6 on RDNA 4; the card-in queue holds one" }, { @@ -1314,13 +1381,13 @@ "mem_mhz": 0, "core_offset_mhz": -300, "power_offset_pct": -15, - "limit_w": 160, - "mhs": 9.5, - "w": 110.0, - "mhw": 0.0864, + "limit_w": 130.0, + "mhs": 12.0, + "w": 111.8, + "mhw": 0.1073, "source": "measured", "label": "estimated", - "uj": 11.6, + "uj": 9.32, "note": "half the offsets" }, { @@ -1330,14 +1397,14 @@ "mem_mhz": 0, "core_offset_mhz": 0, "power_offset_pct": 0, - "limit_w": 160, - "mhs": 9.5, + "limit_w": 130.0, + "mhs": 12.0, "w": 130.0, - "mhw": 0.0731, + "mhw": 0.0923, "source": "stock", "label": "estimated", - "uj": 13.7, - "note": "modelled" + "uj": 10.83, + "note": "8 channels on RDNA 4 between the 9070 XT's 1.18 MH/s per channel and the 7600's 1.74: about 12 MH/s at about 130 W (10.8); the card-in queue holds one" } ], "src": "modelled on the 9070 XT rows" diff --git a/docs/analysis/class-v6/floor/denominator.md b/docs/analysis/class-v6/floor/denominator.md index a3fc4e906..5188f132c 100644 --- a/docs/analysis/class-v6/floor/denominator.md +++ b/docs/analysis/class-v6/floor/denominator.md @@ -118,10 +118,11 @@ Max; the wall more); the NVIDIA and AMD rows are whole-card. | NVIDIA GeForce RTX 3070 | 8 GB | 4.8 | estimated | 5.29 (measured) | 8.3x / 6.1x / 4.3x | 11.1x / 7.4x / 5.0x | 19.1x / 10.4x / 6.1x | 1800 MHz at 80 percent. the Ampere cap lever on the measured stock rows (class v3 142.3 W, class v4 196.4 W); the 3070 Ti 38.98 MH/s at 177.1 / 267.6 W Stock: rented pods 8 October; the class v5 sweep's two 3070 hosts closed the connection mid-bench | | NVIDIA GeForce RTX 3060 | 12 GB | 5.77 | estimated | 6.4 (measured) | 10.0x / 7.3x / 5.2x | 13.3x / 8.9x / 6.0x | 23.0x / 12.4x / 7.3x | 1800 MHz at 80 percent. the Ampere cap lever (10 percent) on the measured class v4 stock row; the 3060 Ti at a 130 W host cap held 33.06 MH/s at 128.7 W under class v4 (3.89): a cap row measured on a sibling Stock: class v5 measured on a Vast 3060 (13:53 UK): 26.53 MH/s at 169.8 W (6.40) at the host's 170 W limit (the limit binding: the class v4 pod read 26.89 at 166.1 W, 6.18); class v3 26.53 at 120.4 W on the same host; stock, lock owed | | AMD Radeon RX 9070 XT | 16 GB | 7.9 | measured | 10.7 (measured) | 13.7x / 10.0x / 7.1x | 18.3x / 12.3x / 8.2x | 31.4x / 17.0x / 10.0x | ADLX -500 MHz, -30 percent. the ADLX grid (24 rows, PC 1): the rate flat at 18.9 MH/s across the grid, the best point -500 MHz core and -30 percent power, 24 percent under stock; class v5 about 8.1 Stock: the app's own power reading at the stock point, 8 October 07:11 UK | -| AMD Radeon RX 7900 XTX | 24 GB | 8.0 | estimated | 10.7 (estimated) | 13.9x / 10.1x / 7.2x | 18.5x / 12.4x / 8.3x | 31.8x / 17.3x / 10.2x | ADLX -500 MHz, -30 percent. not rentable, not owned: the 9070 XT's dependent-read rate per channel (150 M per second per 32-bit channel) on 24 channels; the knob by the 9070 XT grid Stock: modelled | -| AMD Radeon RX 7800 XT | 16 GB | 9.5 | estimated | 12.8 (estimated) | 16.5x / 12.0x / 8.5x | 22.0x / 14.7x / 9.8x | 37.8x / 20.5x / 12.1x | ADLX -500 MHz, -30 percent. 16 channels of GDDR6 at the 9070 XT's per-channel rate; the knob by the 9070 XT grid Stock: modelled | -| AMD Radeon RX 7600 XT | 16 GB | 13.0 | estimated | 17.8 (estimated) | 22.5x / 16.5x / 11.7x | 30.1x / 20.2x / 13.4x | 51.7x / 28.0x / 16.5x | ADLX -500 MHz, -30 percent. 8 channels; the card-in queue holds one for a PC measurement Stock: modelled | -| AMD Radeon RX 9060 XT | 16 GB | 10.0 | estimated | 13.7 (estimated) | 17.3x / 12.7x / 9.0x | 23.1x / 15.5x / 10.3x | 39.8x / 21.6x / 12.7x | ADLX -500 MHz, -30 percent. 8 channels of GDDR6 on RDNA 4; the card-in queue holds one Stock: modelled | +| AMD Radeon RX 7900 XTX | 24 GB | 5.89 | estimated | 7.75 (estimated) | 10.2x / 7.5x / 5.3x | 13.6x / 9.1x / 6.1x | 23.4x / 12.7x / 7.5x | ADLX -500 MHz, -30 percent. not rentable, not owned: the 9070 XT's dependent-read rate per channel (150 M per second per 32-bit channel) on 24 channels; the knob by the 9070 XT grid Stock: 24 channels at the RX 7600's measured per-channel rate: about 40 MH/s at about 310 W (7.8); not rentable, not owned | +| AMD Radeon RX 7800 XT | 16 GB | 6.47 | estimated | 8.52 (estimated) | 11.2x / 8.2x / 5.8x | 15.0x / 10.0x / 6.7x | 25.7x / 14.0x / 8.2x | ADLX -500 MHz, -30 percent. 16 channels of GDDR6 at the 9070 XT's per-channel rate; the knob by the 9070 XT grid Stock: 16 channels at the RX 7600's measured 1.74 MH/s per channel: about 27 MH/s at about 230 W (8.5); not rentable, not owned | +| AMD Radeon RX 7600 | 8 GB | 6.19 | estimated | 8.14 (measured) | 10.7x / 7.8x / 5.6x | 14.3x / 9.6x / 6.4x | 24.6x / 13.3x / 7.9x | ADLX -500 MHz, -30 percent. the 9070 XT's ADLX grid (24 percent of the draw for no rate) applied; this card's own 24-point grid runs on PC 1 by about 16:50 UK and replaces this row Stock: the founder's RX 7600 (8 GB) on PC 1, the card-in job 15:46 to 15:50 UK: class v4 sub-version 3 at stock 13.88 MH/s at 113 W (8.14), the v4 and v5 fingerprints equal to the pinned readings, the app's own row 13.4 at 113 W; class v5 within 2 percent; the 1 GiB dataset fits in 8,176 MB (the 2, 4 and 5.5 GiB rows follow) | +| AMD Radeon RX 7600 XT | 16 GB | 6.56 | estimated | 8.63 (estimated) | 11.4x / 8.3x / 5.9x | 15.2x / 10.2x / 6.8x | 26.1x / 14.1x / 8.3x | ADLX -500 MHz, -30 percent. 8 channels; the card-in queue holds one for a PC measurement Stock: the same Navi 33 die as the measured RX 7600 (13.88 MH/s at 113 W) with 16 GB in clamshell: the same rate, about 120 W (8.6); the card-in queue holds one | +| AMD Radeon RX 9060 XT | 16 GB | 8.23 | estimated | 10.83 (estimated) | 14.3x / 10.4x / 7.4x | 19.0x / 12.8x / 8.5x | 32.8x / 17.7x / 10.5x | ADLX -500 MHz, -30 percent. 8 channels of GDDR6 on RDNA 4; the card-in queue holds one Stock: 8 channels on RDNA 4 between the 9070 XT's 1.18 MH/s per channel and the 7600's 1.74: about 12 MH/s at about 130 W (10.8); the card-in queue holds one | | NVIDIA H100 80GB HBM3 | DC | 2.0 | estimated | 2.58 (measured) | 3.5x / 2.5x / 1.8x | 4.6x / 3.1x / 2.1x | 8.0x / 4.3x / 2.5x | 1400 MHz at 80 percent. the premium 240 to 280 W at stock says half of it is the clock; a lock on an owned host (Hopper at 1,980 MHz boost, HBM3-bound): about 500 W (2.0); band 1.9 to 2.4; rented hosts refuse -lgc Stock: the denominator sweep, RunPod, 13:25 UK: class v3 252.96 MH/s at 382.4 W (1.51), class v5 241.34 at 621.6 W (2.58, 7 samples, the SM throttled to 1,590 MHz under the 700 W cap); the 7 and 8 October pods read class v4 250.6 at 695.1 W (2.77); -lmc answered 'use --lock-memory-clocks-deferred' and the memory clock stayed 2,619; stock, lock owed | | NVIDIA L40S | DC | 3.78 | estimated | 5.29 (measured) | 6.5x / 4.8x / 3.4x | 8.7x / 5.9x / 3.9x | 15.0x / 8.2x / 4.8x | 1860 MHz at 60 percent. the Ada shape on the measured stock rows (class v3 220.7 W, class v4 277.6 W); re-based on the measured class v5 stock row of 13:5x UK (the architecture's shape on the measured v3 draw and the measured premium) Stock: class v5 measured (13:49 UK): 56.49 MH/s at 298.7 W (5.29, 46 samples, SM 2,520); class v3 56.35 at 221.5 W on the same host; the v4watts pod read class v4 56.4 at 277.6 W; stock, lock owed | | NVIDIA A100-SXM4-80GB | DC | 2.89 | estimated | 2.99 (measured) | 5.0x / 3.7x / 2.6x | 6.7x / 4.5x / 3.0x | 11.5x / 6.2x / 3.7x | 1200 MHz at 80 percent. the card already runs at 1,410 MHz and the 400 W host limit binds under class v5 (SM clock gives); a cap under it costs rate on Ampere, so the floor is within 5 percent of the measured stock row Stock: the denominator sweep, RunPod, 13:24 UK: class v3 138.06 MH/s at 294.9 W (2.14), class v5 133.11 at 398.1 W (2.99, 19 samples, at the host's 400 W limit, memory 1,593 MHz); the v4watts pod on a 500 W host read class v4 138.0 at 489.3 W (3.54); -lmc 'not supported' on this card; stock, lock owed | @@ -359,8 +360,10 @@ closed the connection mid-bench). The sweep closed at 14:35 UK: 20 rows landed, - The 5070 Ti and 5070 floors (1.7 to 2.1) are the strongest claims in this file and both are estimated from stock rows; a 5070 Ti ladder on PC 1 or PC 2 (the card-in job's queue) would settle whether the honest NVIDIA floor is a 16 GB card. -- The AMD rows other than the 9070 XT are modelled on one card; the 9060 XT and 7600 XT in the card-in queue will - replace two of them. +- The AMD rows other than the 9070 XT and the RX 7600 are modelled on those two measured points (the 7600's stock row + landed 15:50 UK from the card-in job on PC 1: 13.88 MH/s at 113 W, 8.14 microjoules, twice as good as this file's + morning estimate for the Navi 33 die; its own ADLX grid by about 16:50 and the 9060 XT in the card-in queue replace + the estimated rows). - The Apple M4 Max, M4 Pro and M3 Max rows are modelled on the M5 Max's split; a Metal bench on any of them with the IOReport meter would replace its row in an hour. - Section 4's LPDDR6 rows are modelled on JESD209-6's public summary (the bank count and tFAW are behind the