fpga-archive-reports-qwen · EAGLE4 FPGA archive
SURVEY: bit precision vs area, record 7-card images vs the Qwen GPTQ-2 (Q5120) images, and the room left to enlarge the 2-bit GEMM (2026-11-16)
Program: the 7-card TARGET-side 4C1 drafter (ATT x4 TP4 + MLPV x2, ONE drafter; Qwen3.8-27B target, drafter = Qwen EAGLE-4 GPTQ v2 2-bit, H 5120). Motto: a research track never ends by denying the idea; a wall is a measured constraint to engineer around, toward the fastest flashable PDI.
Agent: survey census for box1-qwen-junior. User question (verbatim, 2026-11-16): "others did get larger... so I want to see if we can use the extra space to enlrge the gemm for 2 bit... can you do a survay on the current bit preicions - area trade versus the new image".
USER RULING relayed by the coordinator (2026-11-16): activations are A16 on the Qwen GPTQ-2 images, for accuracy. Every headroom row in §1 is priced at A16.
Rules kept: read-only on every fork and build; nothing killed; nothing deleted; no build launched; Vivado runs only on my VMs (buildx04 / buildxd00 / buildy00); bulk scratch /mnt/hdd_data/eagle4_scratch/qwen_gptq2_20261116/survey/.
Layout: §0 answers; §1 holds every calculation, projection and hypothesis; §2 to §6 hold only measured values with their sources.
0. Answers
- Binding resource per card (measured, placed).
- Record ATT 0x13BE: SLICE in SLR0, at 98.12 %. SLR0 also carries DSP at 89.0 %.
- Record MLPV 0x1361 (M1361 final): SLICE in SLR2 (94.55 %) and SLR1 (93.44 %). Next come SLR1 DSP at 84.5 % and SLL SLR1<->SLR0 at 78.2 %.
- Q5120 ATT 0x1346 / 0x134A: SLICE in SLR1, at 93.3 %. SLR0 is almost empty: SLICE 10 %, LUT 3.8 %, URAM 0, DSP 0.5 %. The device as a whole is at SLICE 55 %, LUT 26 %.
- Q5120 MLPV 0x1349: SLICE everywhere: device 94.45 %, SLR 92.8 / 93.4 / 97.2 %.
- 0x135E: close to the same netlist as 0x1349 (synth u_array 312,320 vs 311,050 LUT), placed with AggressiveExplore. It measures SLICE 78.8 % device / 90.8 % SLR1. So SLICE is partly a placement outcome; LUT is 53 % on both.
- Where the bits went (§3 to §4).
- The 2-bit weights stayed in URAM at 16 codes per 32-bit word on every image.
- The MLPV GEMV changed precision class. The record MLPV computes 2-bit x A16 on DSP58: E3 packed dot, 2 products per DSP58, 5,120 DSP for 256 lanes. The Q5120 MLPV computes 2-bit x A8 in LUTs: LUT-GEMM lane with a per-column zero bank, 311K LUT and 1,889 DSP for 256 lanes.
- The MLPV URAM grew 512 -> 1,280 because the Q5120 v0 split put G7 on the MLPV, with 2.6x larger weights.
- G8 went from fp16 rows (788 MB head) to 2-bit rows (340 MB). The synth u_g8 still grew 314K -> 358K LUT, because H 5120 and 248K vocab offset the measured -29K LUT of the 2-bit dot.
- The fp16 seams grew 2-4x (H 3072 -> 5120 plus the G7 hand-off).
- Headroom to enlarge the 2-bit GEMM at A16 (§1.3):
- ATT card: SLR0 is the space, with about 85K SLICE, 3,965 DSP and 677 URAM free after the base. At A16, neither lane form alone reaches the v0 L512: the LUT lane gives 480 lanes (340 at the 2.1x price) and the DSP58 E3 lane gives 401. A per-die hybrid gives 631 lanes (556 at 2.1x): FITS. The largest products per cycle is LUT K64/K128 at x1.49: 22K to 29K.
- MLPV card: at A16, only the DSP58 E3 lane fits L256 (382 lanes max). The LUT lane at A16 reaches only 215 (151 at 2.1x). The largest products per cycle is E3 K64, 15,040, DSP-bound.
- SLL is not the wall for the array on either card (ATT 17.6 / 11.8 % used; MLPV 65.8 % used, about 6.4K free per cut).
- The URAM read width is the free K lever.
- What a URAM gives: a URAM288 is 4,096 x 72 with 2 ports, 144 b per cycle. The shipped bank packs two 32-bit (16-code) words per 72-bit row and reads 32 codes per lane-cycle, about 44 % of what the URAM offers.
- K64: a 64-bit entry at half the depth needs no extra URAM. It is DSP-free on the LUT lane; on the DSP lane it costs +16 DSP per lane.
- K128: needs the lane's 2 URAMs read side by side. Both v0 cards have 2 per lane.
- URAM per card is set by the weight bits, not by K: about 730 minimum for the ATT and about 442 for the MLPV.
- Instrument limit: the SMBus licence served to the VMs (2101) currently returns
FlexNet -95,378 Connection reset by peer([Netlist 29-57]onaxi_smbus_rpu), so no routed full-chip dcp could be opened today. Two consequences:- Per-SLR numbers come from each build's own placed report (and from the M1361 routed
-slrreport). - Per-instance numbers come from
report_utilization -hierarchicalon each build's drafter IP OOC synth dcp. That dcp holds only our RTL and needs no licence. Its counts are pre-opt: they include logic that opt_design later prunes, such as the legacy fp16 dot in the Q5120 u_g8. - No per-instance per-SLR split was measured.
- Per-SLR numbers come from each build's own placed report (and from the M1361 routed
1. Calculations and hypothesis (every derived number in this document)
1.1 Per-SLR capacity used below [D on the place reports]
The capacity of each SLR is computed as used / % from Q46_place_report_utilization_0.rpt:393-412, cross-checked against the used + free sums in SYNTH_CENSUS_WEEK_20261105_12.md:490-510.
| SLR | SLICE | LUT | DSP58 | URAM | BRAM tiles |
|---|---|---|---|---|---|
| SLR0 | 112,527 | 900,528 | 3,984 | 677 | 1,341 |
| SLR1 | 102,530 | 820,161 | 3,456 | 624 | 1,200 |
| SLR2 | 106,719 | 853,519 | 3,408 | 624 | 1,200 |
1.2 Per-lane prices (inputs are the measured rows of §5; the scaling is mine)
A8 LUT lane, in the image: 277 SLICE, 1,082 LUT, 6.0 DSP, 1.5 BRAM tiles, 2 URAM per lane.
- Source: the placed 0x1346 SLR2, which holds exactly the 256-lane array: all 512 array URAM, 256 RAMB36 of scales and 256 RAMB18 of zeros sit there. That die measures SLICE 70,856, LUT 276,918, DSP 1,540, BRAM 384, each divided by 256.
- The record ATT lane is cheaper. 0x13BE SLR2 holds 341 lanes in 82,885 SLICE, so at most 243 SLICE per lane. The LUT rows below are therefore slightly pessimistic.
A16 LUT lane = the A8 price x 1.49, or x 2.1.
- x 1.49 is the measured ratio of the narrow-tree (H2) lane: 1,673 vs 1,122 LUT (
LUTGEMM_LANE_E1_20261102.md:26-27). The shipped lanes are narrow-tree (CSA_NARROW 1,verilog/clean7c/docs/ARRAY.md:225). - x 2.1 is the coordinator's figure. It equals the A16 multiply lane against the A8 LUT lane (2,701 / 1,261,
DESIGN_FABLE_FULL2BIT_LUT_ARRAY_20261102.md:31), so read it as the conservative bound. - At A16 the A8Q quantizer goes away: -5.3K LUT, -96 DSP (§5.1). This is not credited below.
- ADDENDUM 2026-11-16 (later the same day): use the x 2.1 column as the central estimate. The Fable GEMM agent routed the same 8-lane datapath harness at AW=16 (buildxd00;
DESIGN_FABLE_GPTQ2_GEMM_20261116.md§1.8, commit8d6a51568). It measures 1,339 LUT / 642 FF / 206 SLICE per lane at K32, against 667 / 443 / 110 at A8 (§2.3 of that doc). That is 2.0x LUT and 1.87x SLICE, much closer to x 2.1 than to x 1.49. The 1.49 ratio is a synth-only number from the full lane (LUTGEMM_LANE_E1:26-27).- Same run: A16 K64 = 2,541 LUT / 388 SLICE per lane, and A16 W=2 = 2,671 LUT / 406 SLICE per lane.
- At A16 a DSP58 carries 2 products, not the 3 to 4 of the A8 packing, so the E3 rows above (2 per DSP58) are the right DSP-form price.
- What this changes: ATT L512 at A16 fits only as the hybrid (556 lanes at x 2.1). The MLPV conclusion is unchanged: the E3 DSP lane fits L256, and the LUT lane does not.
DSP58 E3 lane at A16, 2 products per DSP58: about 206 SLICE, 20 DSP58, 1.6 BRAM tiles, 2 URAM per lane.
- Source: the record 0x1361 synth hier gives 575 LUT, 539 FF and 20 DSP58 per lane. The 20 DSP58 are 17 in the packed pair dot, 1 in the lane, 1 in the bank and 1 in the scale (§5.3).
- To that I add +45 LUT and one RAMB18 per lane for the per-column zero, which the record lane lacks (
LUTGEMM_LANE_E1:19). - SLICE is estimated as the larger of LUT / 3.91 and FF / 2.61, using the packing of the 0x1346 array die: about 206 per lane.
K scaling.
- LUT lane: +105 SLICE per lane per doubling of codes per lane-cycle at A8 (Fable harness K64,
DESIGN_FABLE_GPTQ2_GEMM_20261116.md§2.3), x 1.49 at A16. - E3 lane: +16 DSP58 per doubling (32 more products at 2 per DSP58), and +25 % SLICE (a guess).
Non-array bases.
- ATT = 0x1346 SLR0 + SLR1, plus +16K SLICE on SLR0 for G7 on the ATT and u_tp at L512. That delta comes from synth: 0x13BE u_tp 104,881 vs 0x1346 50,679, plus SiLU 3,010 and the r4 seams 7,565, about +62K LUT at 3.91 LUT per SLICE.
- MLPV (Q5120 base) = 0x1349 per SLR minus its own 256 lanes. The lanes are placed by URAM as 102 / 53 / 101, at 5 URAM per lane.
- MLPV (record base) = M1361 per SLR minus its E3 array of 142 / 35 / 79 lanes at 2 URAM per lane.
- The record ATT base goes negative in SLR2: its lane is about 243 SLICE, not 277. So the ATT is priced on the Q5120 base only.
- Script:
survey_evidence/a16_fit.py(md5924d1163a76392c1a31cc300911a1f4b); outputsurvey_evidence/a16_fit.out(md55a69a9fd68bef8eff858a97623261d96).
1.3 Headroom table at A16: v0 geometry (ATT LANES 512, MLPV LANES 256), per card and per SLR
Each SLR is listed as SLR0 / SLR1 / SLR2. Lane counts are at 100 % of each die, with the 90 %-SLICE practical bar in brackets. The binding resource for each row is given in words.
Free after the base
| card (base) | SLICE free | DSP free | URAM free |
|---|---|---|---|
| ATT (0x1346 + G7 delta) | 85.2K / 6.8K / 106.7K | 3,965 / 979 / 3,408 | 677 / 479 / 624 |
| MLPV (0x1349 minus array) | 36.3K / 21.5K / 30.9K | 3,966 / 1,161 / 3,354 | 677 / 624 / 624 |
| MLPV (0x1361 minus array) | 50.3K / 13.9K / 22.1K | 3,933 / 1,237 / 3,294 | 677 / 624 / 624 |
ATT card, v0 needs 512 lanes
| lane form | max lanes per SLR | total [90 %] | binding | v0 fits? | products per cycle at max |
|---|---|---|---|---|---|
| LUT A8 K32 (shipped, reference) | 307 / 24 / 312 | 643 [579] | SLICE in SLR0/SLR1, URAM in SLR2 | yes | 20,576 |
| LUT A16 K32 x1.49 | 206 / 16 / 258 | 480 [411] | SLICE | no | 15,360 |
| LUT A16 K32 x2.1 | 146 / 11 / 183 | 340 [292] | SLICE | no | 10,880 |
| LUT A16 K64 x1.49 | 149 / 12 / 187 | 348 [298] | SLICE | no | 22,272 |
| LUT A16 K128 x1.49 | 96 / 7 / 121 | 224 [191] | SLICE | no | 28,672 |
| DSP E3 A16 K32 | 198 / 33 / 170 | 401 [368] | DSP in SLR0/SLR2, SLICE in SLR1 | no | 12,832 |
| DSP E3 A16 K64 | 110 / 26 / 94 | 230 [204] | DSP | no | 14,720 |
| HYBRID A16 K32 (x1.49) | (127 LUT + 159 E3) / (0 + 33) / (205 + 107) | 631 | SLICE + DSP together | yes | 20,192 |
| HYBRID A16 K32 (x2.1) | (85 + 172) / (0 + 33) / (138 + 128) | 556 | SLICE + DSP together | yes | 17,792 |
MLPV card, v0 needs 256 lanes
| lane form | max lanes per SLR | total [90 %] | binding | v0 fits? | products per cycle at max |
|---|---|---|---|---|---|
| LUT A8 K32 (shipped, reference) | 131 / 77 / 111 | 319 [203] | SLICE | at 100 % only | 10,208 |
| LUT A16 K32 x1.49 | 88 / 52 / 75 | 215 [136] | SLICE | no | 6,880 |
| LUT A16 K32 x2.1 | 62 / 36 / 53 | 151 [96] | SLICE | no | 4,832 |
| LUT A16 K128 x1.49 | 41 / 24 / 35 | 100 [62] | SLICE | no | 12,800 |
| DSP E3 A16 K32 | 175 / 58 / 149 | 382 [273] | SLICE in SLR0/SLR2, DSP in SLR1 | yes | 12,224 |
| DSP E3 A16 K64 | 110 / 32 / 93 | 235 [207] | DSP | no (at 256 lanes) | 15,040 |
| HYBRID A16 K32 (x1.49) | (0 + 175) / (28 + 48) / (0 + 149) | 400 | SLICE + DSP together | yes | 12,800 |
| record base, DSP E3 A16 K32 | 196 / 61 / 107 | 364 [261] | DSP in SLR0/SLR1, SLICE in SLR2 | yes | 11,648 |
What the table says:
- v0 products per cycle at K32: ATT 16,384; MLPV 8,192.
- MLPV, the A16 answer is the record's DSP58 lane, not the LUT lane. It fits with 126 lanes to spare; at K64 it is DSP-bound at 235 lanes.
- E3 K64 x L256 would need 9,216 DSP58 against 8,481 free (Q5120 base). The GLM change measured in
reports/results/DRAFTER_SPEED_RESEARCH_20261107/MLPV_G9_KPAR32_20261112.md(64 lanes on the fast clock, -1,024 DSP58) makes 9,505 free, so it fits device-wide. Per die it is still short: SLR1 has 1,161 free, about 32 lanes. - ATT, both pure forms fall short of 512 at A16. The options are:
- a hybrid array (E3 lanes on the DSP-rich SLR0/SLR2, LUT lanes in the remaining SLICE);
- fewer, wider lanes: about 350 LUT lanes at K64 = 1.36x the v0 products, or 224 at K128 = 1.75x.
- Every ATT row leaves SLR1 out. Attention and control hold SLR1 at 93 %, so its 979 DSP and 6.8K SLICE buy almost nothing.
1.4 URAM and K (the read-width fact) [D on R + M]
- What a URAM delivers: URAM288 is 4,096 rows x 72 b, 2 ports, so 144 b per cycle. The bank stores DATA_W 32 words (
verilog/clean7c/rtl/common/weight_bank_uram.sv:30-31, md5059f61de). The synthesis packs two 32-bit words per 72-bit row: DEPTH 8,192 x 32 b = 1 URAM per lane, measured 512 for 512 lanes on 0x13BE. - What the shipped lane uses: 2 ports x 16 codes = 32 codes per lane-cycle = 64 of 144 b. A 72-bit port can carry 36 codes at most.
- K64 (K_PAR 32): reads the whole 64-bit pair per port, 128 of 144 b. Same URAM count, because a 64-bit entry at half the depth holds the same bits (
BANK64_ENTRY_OOC_20260824.md, quoted in the Fable doc §1.4). The 128-code group is not a multiple of 36, so 64 is the practical width per port. - K128 (K_PAR 64): needs 256 b per cycle, which is 2 URAMs read side by side. Both v0 cards carry 2 per lane: ATT L512 has 11,680 of 16,384 entries used; MLPV L256 has 14,120 of 16,384.
- URAM per card does not move with K. It moves with the weight bits / lanes:
- ATT TP4 v0 = (4,608·10,240 + 5,120·1,280 + 2·4,096·5,120) / 16 = 5,980,160 32-bit entries. That is a minimum of 730 URAM; 1,024 at L512 (2 per lane); 768 at L384 or L768.
- MLPV v0 = (5,120·8,192 + 124,160·128) / 16 = 3,614,720 entries. That is a minimum of 442; 512 at L256.
1.5 Precision ledger bytes [D on R]
Weight bytes = N·K / 4 for 2-bit codes, N·K / 64 for int16 scales per 128-group, N·K / 512 for 2-bit zeros. Stage shapes come from ARRAY.md §4 (record) and QWEN5120_P1_20261108.md §1.2 (Q5120).
| card | codes (URAM) | scales int16 (BRAM) | zeros 2-bit (BRAM18) |
|---|---|---|---|
| record ATT 0x13BE (G4 3,072x6,144 + G6 3,072x768 + G7g/u 2,048x3,072) | 8.45 MB (= 4,128 words x 512 lanes x 4 B, ARRAY.md:117) | 0.53 MB | 66 KB |
| record MLPV 0x1361 (G9 6,144x4,096 + G12 64,128x256) | 10.4 MB (10,160 words x 256 x 4 B) | 0.65 MB | none (zero broadcast per group) |
| Q5120 ATT 0x1346 (G4 3,840x10,240 + G6 5,120x1,280) | 11.5 MB (11,200 entries per lane x 256) | 0.72 MB | 90 KB |
| Q5120 MLPV 0x1349 (G7g + G7u 8,192x5,120 + G9 5,120x8,192 + G12 124,160x128) | 35.4 MB | 2.2 MB | 0.28 MB |
| v0 ATT (TP4 v1 split) | 23.9 MB | 1.5 MB | 0.19 MB |
| v0 MLPV | 14.5 MB | 0.90 MB | 0.11 MB |
1.6 Hypotheses
- H1: the A16 LUT lane prices at 1.49x the A8 lane in the image, as it does in the lane harness. Not measured in an image; the 2.1x column is the bound if it does not.
- H2: a mixed LUT/E3 array per die can share one stage walk. The two lane forms have different latencies; the E3 lane is a pair unit. No RTL exists for a mixed array; this is an RTL question for the RTL agent.
- H3: SLICE % is partly a placer outcome (0x1349 94.45 % vs 0x135E 78.81 % on near-identical netlists), so the 100 % column is reachable only with the AggressiveExplore-class recipe; the 90 % column is the safer plan.
- H4: A16 doubles the activation broadcast into each die. About +256 nets per cut at K32 (feed 493 nets per cut at A8,
LUTGEMM_LANE_E3_20261102.md). This is not binding on either card's SLL (§2), but it is the known H5120 timing wall, so it needs the ACT_XPIPE form.
2. Device and per-SLR utilisation per image (measured, placed / routed)
Each per-SLR column reads SLR0 / SLR1 / SLR2. SLL figures are used / 18,870 per cut. The V80 totals are LUT 2,574,208, FF 5,148,416, DSP58 10,848, BRAM36 tiles 3,741, URAM 1,925, SLICE 321,776.
| image (role, build host, report, md5) | SLICE | LUT | FF | DSP58 | BRAM tiles | URAM | SLL 1<->0 / 2<->1 | binding |
|---|---|---|---|---|---|---|---|---|
0x13BE record ATT HCUT, Llama H3072, L512 LUTG A8; buildx04 s3LAtp4BE placed, BE_place_report_utilization_0.rpt 03fdd895 |
263,389 (81.85 %); 110,410 / 70,094 / 82,885 = 98.12 / 68.37 / 77.67 % | 1,023,252 (39.75 %); 45.69 / 31.00 / 41.90 % | 932,324 (18.11 %) | 5,267 (48.55 %); 3,546 / 694 / 1,027 = 89.0 / 20.1 / 30.1 % | 1,783 (47.66 %) | 744 (38.65 %); 232 / 171 / 341 | 9,501 (50.4 %) / 1,880 (10.0 %) | SLICE in SLR0 |
0x1361 record MLPV G8_OWN, L256 E3 A16 (M1361 routed -slr, M1361_final_slr_util.rpt edfa7302; device totals from build placed M61build_… bf16ecff) |
288,227 per-SLR sum (89.6 %); 91,520 / 95,807 / 100,900 = 81.33 / 93.44 / 94.55 % | 1,185,094; 35.70 / 50.31 / 52.83 % | 1,018,982 (19.8 %) | 7,504 (69.17 %); 2,891 / 2,919 / 1,694 = 72.6 / 84.5 / 49.7 % | 1,035.5 (27.7 %) | 512 (26.6 %); 284 / 70 / 158 | 14,764 (78.2 %) / 12,533 (66.4 %) | SLICE in SLR2/SLR1, then SLR1 DSP, then SLL 1<->0 |
0x1346 Q5120 ATT TP4, L256 LUTG A8, 10x128 heads HP5/HPS2; buildx04 ~/build4c1/s3LAq5120t4c placed, Q46_place_… 0670e6d2; route WNS -2.565 (failed setup: area only) |
177,862 (55.28 %); 11,306 / 95,700 / 70,856 = 10.05 / 93.34 / 66.39 % | 661,503 (25.70 %); 3.79 / 42.73 / 32.44 % | 693,438 (13.47 %) | 4,036 (37.21 %); 19 / 2,477 / 1,540 | 1,367.5 (36.55 %); 24 / 959.5 / 384 | 657 (34.13 %); 0 / 145 / 512 | 3,328 (17.6 %) / 2,233 (11.8 %) | SLICE in SLR1 (SLR0 empty) |
0x134A = 0x1346 + G6_EARLY_ISSUE + KVF_TFIX, stopped copy on buildx00 ~/build4c1/s3LAq5120t4k placed, Q4A_place_… e40e6d2e (closed copy +0.047/+0.011 is on buildxd05, not opened) |
177,186 (55.07 %); 9.81 / 93.06 / 66.28 % | 658,621 (25.59 %) | 693,717 (13.47 %) | 4,036 (37.21 %) | 1,367.5 | 657; 0 / 145 / 512 | 3,243 / 2,277 | SLICE in SLR1 |
0x1349 Q5120 MLPV L256 LUTG A8, TREE_P 32, G8_W2; buildx00 s3LCq5120m49 placed, Q49_place_… 5a21d27e |
303,917 (94.45 %); 104,451 / 95,738 / 103,728 = 92.82 / 93.38 / 97.20 % | 1,376,072 (53.46 %); 46.65 / 59.00 / 55.31 % | 1,285,443 (24.97 %) | 4,256 (39.23 %); 771 / 2,686 / 799 | 2,271.5 (60.72 %); 708 / 953.5 / 610 | 1,280 (66.49 %); 510 / 265 / 505 | 12,413 (65.8 %) / 12,420 (65.8 %) | SLICE, every SLR |
0x135E Q5120 MLPV (+A8Q_INREG; place AggressiveExplore); buildy03 s3LCq5120m5E placed, Q5E_place_… 2f61df52; route +0.013 (CONFIG_MANIFEST_135E.md) |
253,589 (78.81 %); 72.37 / 90.78 / 74.10 % | 1,367,658 (53.13 %) | 1,286,295 (24.98 %) | 4,256 (39.23 %); 738 / 2,875 / 643 | 2,271.5 | 1,280; 500 / 350 / 430 | 12,559 (66.6 %) / 11,011 (58.4 %) | SLICE in SLR1 |
| 0x135D Q5120 MLPV L256/TP64/THARD + A8Q_INREG; buildxd05 (not opened: box1-two-v2 partition) | 96.30 % / worst SLR 99.58 % | 64.39 % | 39.23 % | 60.72 % | 66.49 % | 13,269 / 12,011 | SLICE; route -2.862 |
The 0x135D row is the census line SYNTH_CENSUS_WEEK_20261105_12.md:269.
3. Per-instance hierarchy (measured, drafter IP OOC synth dcp, report_utilization -hierarchical, pre-opt)
Each row reads LUT / FF / RAMB36 / RAMB18 / URAM / DSP. The line number is that of the instance row in the image's *_synth_hier.rpt, and the md5 of each report is in the source line at the end of this section. Each dcp was copied from .../prj.runs/top_eagle4_drafter_lb_0_0_synth_1/; the copy md5s are in survey_evidence/*_ooc_dcp.md5.txt.
| block | record ATT 0x13BE | Q5120 ATT 0x1346 | record MLPV 0x1361 | Q5120 MLPV 0x1349 |
|---|---|---|---|---|
| drafter total | 1,046,766 / 940,325 / 1,362 / 610 / 744 / 5,459 (:25) | 645,796 / 653,474 / 1,091 / 321 / 657 / 4,100 (:25) | 1,574,493 / 935,306 / 931 / 26 / 512 / 7,602 (:25) | 1,465,370 / 1,230,084 / 2,187 / 26 / 1,280 / 4,419 (:25) |
| GEMV array u_array | 597,623 / 466,864 / 512 / 512 / 512 / 1,635 (:32) | 311,016 / 241,632 / 256 / 256 / 512 / 1,633 (:30) | 147,305 / 137,895 / 280 / 0 / 512 / 5,120 (:31) | 311,050 / 236,414 / 1,536 / 0 / 1,280 / 1,889 (:30) |
| array lanes | 512 x lutgemm_lane_w2 988 / 628 / DSP 1 (:39) |
256 x 986 / 627 / DSP 1 (:37) | 256 x E3 lane 348 / 366 / DSP 1 (:37) + 256 x pair dot 125 / 53 / DSP 17 (:33,:35) | 256 x 988 / 628 / DSP 1 (:37); u_bank 54 LUT / 5 URAM / 2 DSP per lane (:41) |
A8 quantizer u_a8q |
5,356 / 707 / DSP 96 (:34) | 5,282 / 715 / DSP 96 (:32) | none (A16) | 5,260 / 704 / DSP 96 (:32) |
attention u_attn_full |
164,151 / 169,464 / 672 / 64 / 232 / 3,416 (:2181); 8 engines x 10,974 LUT / 395 DSP | 127,275 / 123,022 / 673 / 47 / 145 / 2,221 (:1092); 5 engines x 11,470 / 393 DSP | none | none |
KV u_kv (K URAM, V BRAM, 16-bit) |
10,392 LUT / 448 RAMB36 / 64 RAMB18 / 232 URAM (:2319) | 6,326 / 560 / 40 / 145 URAM (:1179) | none | none |
RoPE u_rope |
5,818 / 216 RAMB36 / 256 DSP (:2321) | 2,177 / 108 RAMB36 / 256 DSP (:1181) | none | none |
G6 TP exchange u_tp (48-b partials) |
104,881 / 45,059 / 11 RAMB36 / 32 DSP, LUTRAM 14,056 (:29) | 50,679 / 21,879 / 11, LUTRAM 7,036 (:29) | none | none |
G8 exact rescore u_g8 |
none | none | 314,123 / 71,687 / DSP 84 (:821); 4 fp16 dots x 22,751 LUT / 16 DSP (:825) | 358,081 / 77,246 / DSP 100, LUTRAM 10,688 (:1337); 4 W2 dots x 23,634 LUT / 4 DSP (:1341), plus 4 legacy fp16 dots x 23,429 / 16 DSP still present pre-opt (:1342) |
GLM g_glm.u_lmdown (bf16 down-proj x 16-bit act) |
none | none | 30,240 / 79,439 / DSP 2,048 (:29) | 30,543 / 79,436 / DSP 2,048 (:29) |
TREE u_tree / u_cheap_ser / u_tree_tiler |
none | none | 75,099 / 46,280 (LUTRAM 15,360) / 13,743 (:854,:852,:853) | 75,099 / 46,292 / 13,800 (:1427,:1425,:1426) |
| tree DMA / out-walk / bridge | none | none | 45,675 / 13,523 / 21,445 (:808,:810,:807) | same values (:1324,:1326,:1323) |
| fp16/int16 activation seams (FF) | r4g / r4u 3,697 / 3,868 (SDP BRAM 4 each); g5g6 657 / 8 RAMB36 | r1_e 796 / 8 RAMB36; g5g6 7,439 / 8 RAMB36 | r1_e / r1_h 17,460 / 17,769, 50,182 FF each; r5 23,075 / 66,824; r8 17,308 / 49,427; r6 16,397 / 49,161 | r1_e / r1_h 56,615 / 56,320, 83,512 FF each; r5 89,822 / 133,436; r8 28,700 / 82,197; r6 27,406 / 81,930 |
| RMSNorm (each of g1, g1_e, premlp, g11, g1_e_d) | ~2,100-2,450 / ~975 / DSP 48 | ~2,060-2,340 / ~980 / DSP 48 | ~2,050-2,250 / ~970 / DSP 48 | ~2,040-2,360 / ~975 / DSP 48 |
SiLU u_silu |
3,010 / 545 / DSP 32 | none (G7 not on this ATT) | none | 3,010 / 545 / DSP 32 |
u_ctrl |
28,926 | n/a (< 10K row) | 48,100 | 50,840 |
drafter own logic (inst) |
68,776 / 202,087 | 102,170 / 227,659 | 618,175 / 163,190 | 126,842 / 171,957 |
Sources:
survey_evidence/BE_synth_hier.rpt(2d30e2dea45b19a6e4ac9813a14bda20)Q46_synth_hier.rpt(40a9594fb6d41191ffcc1c11f3e2399d)M61_synth_hier.rpt(bd5ad4a233569d4c3274dc5c1659f7e3)Q49_synth_hier.rpt(263db434979bd53207b962fca90b6984)Q5E_synth_hier.rpt(30c9b4a8f3d9fe7daf65cc53c5896388), which matches 0x1349 within 0.3 % per block- Depth-2 digest of all five:
hier_depth2_summary.txt
Notes:
- The array memories are not attributed to any child row. Inside u_array, the bank / scale / zero memories (URAM, RAMB36, RAMB18) and some DSP and LUT (0x1346: 23,808 LUT, 768 DSP) show in the u_array total but in no child row, except 0x1349's
u_bank. Per-lane memory counts above are the u_array totals divided by the lanes. - Synth vs placed: synth totals are above the placed totals (e.g. 0x1349 1,465,370 synth vs 1,376,072 placed, which also includes the shell). opt_design prunes logic such as the legacy fp16 dots.
4. Precision ledger (measured placement of each precision class; bytes in §1.5)
4.1 Storage
| precision class | record ATT 0x13BE | record MLPV 0x1361 | Q5120 ATT 0x1346 | Q5120 MLPV 0x1349 |
|---|---|---|---|---|
| 2-bit weight codes, 16 per 32-b word | URAM 512, 1 per lane, DEPTH 8,192 (ARRAY.md:117; u_array :32) | URAM 512, 2 per lane, DEPTH 16,384 (:117; u_array :31) | URAM 512, 2 per lane (u_array :30; all in SLR2, placed :411) |
URAM 1,280, 5 per lane (u_bank :41) |
GPTQ scales (Q3.13, int16; Qwen F 16, GEMV_SCALE_FRAC 16) |
RAMB36 512, 1 per lane (GEMV_SCALE_FRAC 13, verilog/clean7c/ref/att_13be/CONFIG_MANIFEST.md:111) |
RAMB36 256 (+ 24 LCACHE activation BRAM = 280) | RAMB36 256 | RAMB36 1,536 (6 per lane: scale + zero banks for 5 URAM of weights) |
| GPTQ-v2 2-bit zero | RAMB18 512, per (lane, group) | none: one zero per group, broadcast (qz_raw, ARRAY.md:119) |
RAMB18 256 | within the 1,536 RAMB36 |
| activations inside the GEMV | int8 (A8Q), FF per 8-lane cluster | int16 (A16), BRAM per-die cache, then FF | int8 (A8Q) | int8 (A8Q) |
| fp16/int16 activation seams | SDP BRAM (SEAM_BRAM_W 1) and FF: small (§3 seam row) | FF: r1 / r5 / r6 / r8 ~92K LUT, ~266K FF | SDP BRAM, small | FF: ~259K LUT, ~465K FF |
| KV cache fp16 | K URAM 232 (8 heads x 96), V 448 RAMB36 + 64 RAMB18 (u_kv :2319) | none | K URAM 145 (10 heads x 128, HPS 2), V 560 RAMB36 + 40 RAMB18 (:1179) | none |
| RoPE tables | 216 RAMB36 | none | 108 RAMB36 | none |
| norm gamma | inside rmsnorm / drafter own logic: no separate RAM row | same | same | same |
| cheap lm-head (G12) 2-bit | none | in the array URAM (4,016 words per lane of 10,160) | none | in the array URAM |
| exact lm-head rows (G8) | none | fp16 in HBM: 128,256 x 3,072 x 2 B = 788 MB (DESIGN_DUAL_MODEL_4C1.md:144) |
none | 2-bit in HBM: codes 317.8 + scales 19.9 + zeros 2.5 = 340.2 MB, 1,408 B per row vs 6,144 B fp16 at H3072 / 10,240 B at H5120 (DESIGN_FABLE_FULL2BIT_LUT_ARRAY_20261102.md:89; P2_G8W2_20261108.md:7) |
| G8 hidden-state store | (FF row buffers in u_g8 own logic) | LUTRAM 10,688 in the W2 dots (:1341) |
4.2 Compute
| precision class | record ATT 0x13BE | record MLPV 0x1361 | Q5120 ATT 0x1346 | Q5120 MLPV 0x1349 |
|---|---|---|---|---|
| 2-bit x activation GEMV | 2b x A8 in LUT: 512 lanes x 988 LUT; u_array 597,623 LUT / 1,635 DSP | 2b x A16 on DSP58: 256 lanes x (348 + 125) LUT / 18 DSP plus bank / scale DSP; u_array 147,305 LUT / 5,120 DSP | 2b x A8 in LUT: 256 x 986; u_array 311,016 / 1,633 DSP | 2b x A8 in LUT: 256 x 988; u_array 311,050 / 1,889 DSP |
| lane accumulators | 48-b (result_q, ARRAY.md:135), inside the lane rows |
same | same | same |
| attention QK / softmax / PV (16-bit) | 8 engines, 87,792 LUT / 3,160 DSP | none | 5 engines, 57,350 LUT / 1,965 DSP | none |
| RMSNorm (fp/int 16-b, 48 DSP each) | 4 instances | 4 instances (incl. g11) | 4 | 4 |
| GLM low-rank (bf16 x 16-b) | none | 2,048 DSP | none | 2,048 DSP |
| G8 rescore dot | none | fp16 x fp16 wide dot, 4 x (22,751 LUT, 16 DSP) | none | exact 2-bit x int dot, 4 x (23,634 LUT, 4 DSP); P2_G8W2_20261108.md:108-113: u_g8 at H5120 -29,291 LUT, -3,463 FF, -48 DSP vs fp16 |
| TREE top-K (16-bit scores) | none | 75,099 LUT u_tree | none | 75,099 LUT |
5. Record -> Q5120 delta per card (measured, same instrument)
5.1 ATT (0x13BE -> 0x1346; different lane counts by design: L512 1 URAM per lane vs L256 2 URAM per lane; 0x1346 has no G7)
- Device (placed): SLICE 81.85 -> 55.28 %, LUT 39.75 -> 25.70 %, DSP 48.55 -> 37.21 %, URAM 744 -> 657, BRAM 1,783 -> 1,367.5.
- Array: 597,623 -> 311,016 LUT, which is the lane count (512 -> 256). The lane itself is unchanged at 988 / 986 LUT.
- Attention: 8 heads x 96 -> 10 heads x 128 per card, with 8 -> 5 engines. LUT 164K -> 127K, DSP 3,416 -> 2,221, K URAM 232 -> 145, V RAMB36 448 -> 560.
- G6 exchange u_tp: 104,881 -> 50,679 LUT, which is the lane count again.
- SLR placement: the 0x1346 array sits wholly in SLR2 (URAM 512 = 82 %), and attention plus control sit in SLR1 (93.3 % SLICE). SLR0 is nearly empty.
5.2 MLPV (0x1361 -> 0x1349)
What grew:
- URAM 512 -> 1,280, because G7g/G7u moved onto the MLPV in the Q5120 v0 split.
- BRAM 1,035.5 -> 2,271.5 tiles, from the scale banks for those weights.
- GEMV array LUT 147K -> 311K (A16 DSP -> A8 LUT form).
- G8 u_g8 314K -> 358K LUT pre-opt.
- Seams r1 / r5 / r6 / r8 about 92K -> 259K LUT and about 266K -> 465K FF: H 5120 and the G7 hand-off.
- Vocab: TREE_V_MAX 128,256 -> 248,320 (CONFIG_MANIFEST_1349.md:185). The u_tree / u_cheap_ser rows are identical (75,099 / 46,292), so the vocabulary grew in the G12 bank and the tiling, not in the merge.
What shrank:
- DSP 7,504 -> 4,256: array 5,120 -> 1,889.
- Drafter own logic 618K -> 127K LUT.
- The 2-bit G8 dot: -29,291 LUT / -3,463 FF / -48 DSP at H5120 against fp16 (
P2_G8W2_20261108.md:108-113). - The HBM head: 788 MB fp16 -> 340 MB 2-bit.
Net: device SLICE 89.6 -> 94.45 %, LUT 46 -> 53 %, DSP 69 -> 39 %.
6. Run log and instrument record
- Routed / postroute dcp surveys:
survey_util.tcl(76676fd0) viarun_survey.sh(999ba2ae) were run on buildx04 (0x13BE, 0x1346, 0x1349 copy md5071dd632= the adjudicated final dcp), buildxd00 (M1361_final.dcp) and buildy00 (0x135E copy md5cdcf1631= source).- All runs stopped at open_checkpoint:
ERROR: [Netlist 29-57] IP License required ... top_i/base_logic/axi_smbus_rpu/inst,FlexNet Licensing error:-95,378. System Error: 104 "Connection reset by peer"(Q46_run.out:26-34,BE_run.out:26-34). That is withssshowing 127.0.0.1:2101 listening andauthorized_keyscarrying the tunnel line on every VM tried. - The tunnel port is up and the server behind it is not. This is the same state as the 11-16 post-reboot census commit
49d64afde, "licence 2101 dead on all VMs".
- All runs stopped at open_checkpoint:
- Synth hierarchies:
synth_hier.tcl(dd975bbf) viarun_synth_hier.sh(bb3eb805), on buildx04/shared/scratch/qsurvey_20261116/sh/<img>/, all exit 0. - Not opened: buildxd05 (0x134A closed copy, 0x135D), which is box1-two-v2's partition.
- Evidence:
reports/results/QWEN_GPTQ2_20261116/survey_evidence/(MD5SUMS.txt); parsersparse_util.py,parse_hier.py.