BenchmarkApple TV 4KA12A15A19

Apple TV 4K benchmark: A12, A15 and A19

The A19 is faster than the A15 on every kind of GPU work measured: 1.3× to 2.2× as much per 60 Hz frame at 1080p on a cool box, 2× or more only on heavy shader effects, least on layers with holes and 2D sprites in one batch.After 20 minutes it is still more than 10 % ahead on 5 of 7 kinds of work (up to 1.6×), and within 10 % of the A15 on layers with holes and 2D sprites in one batch; every box gives one app the same 2,098 MB, whatever its memory.

Budgets a developer can plan with, not a score. Measured 14–15 November 2026 on tvOS 27.2 and 27.2.1, by one developer with his own test app, on one A12, one A15, two A19 units.

The A19 is faster than the A15 on every kind of GPU work measured: 1.3× to 2.2× as much per frame on a cool box, 1.04× to 1.6× after 20 minutes

The A19’s budget per 60 Hz frame at 1080p ÷ the A15’s, one row per test, the largest lead first. Filled dot: on a cool box (the one-minute budget). Ring: after 20 minutes (each budget × the share of speed that test keeps on that chip, at most 100 %). At 1× both fit the same.

  • cool box
  • after 20 minutes
1×1.5×2×2.5×1× = the same as the A15Full-screen shader passesheavy shader effects (lighting, post-processing)2.2× → 1.6×Meshes, instanced3D models, drawn as one batch1.8× → 1.5×Meshes, one draw each3D models, drawn one by one1.6× → 1.5×Sprites, one draw each2D sprites, drawn one by one1.5× → 1.2×Blended full-screen layerssee-through layers (smoke, glass)1.4× → 1.1×Sprites, instanced2D sprites, drawn as one batch1.3× → 1.04×Alpha-tested layerslayers with holes in them (leaves, fences)1.3× → 1.04×
  • No single test says how much faster a chip is: the lead depends on the kind of work, and heat takes away more of it on some kinds than on others.
  • The opaque-layer control is left out: it measures how little hidden work costs, not a load.
  • The table below also gives the A19 ÷ A12 and A15 ÷ A12 ratios.
The numbers as a table
TestA19 ÷ A15, coolafter 20 minA19 ÷ A12, coolafter 20 minA15 ÷ A12, coolafter 20 min
Full-screen shader passes2.2×1.6×3.7×2.8×1.7×1.7×
Blended full-screen layers1.4×1.1×2.0×1.6×1.4×1.4×
Alpha-tested layers1.3×1.04×1.8×1.4×1.4×1.3×
Sprites, instanced1.3×1.04×2.1×1.3×1.6×1.3×
Sprites, one draw each1.5×1.2×2.2×1.5×1.4×1.2×
Meshes, instanced1.8×1.5×2.1×1.9×1.2×1.2×
Meshes, one draw each1.6×1.5×2.1×1.7×1.3×1.2×

What each test is

Full-screen shader passes
heavy shader effects (lighting, post-processing)
Blended full-screen layers
see-through layers (smoke, glass)
Alpha-tested layers
layers with holes in them (leaves, fences)
Sprites, instanced
2D sprites, drawn as one batch
Sprites, one draw each
2D sprites, drawn one by one
Meshes, instanced
3D models, drawn as one batch
Meshes, one draw each
3D models, drawn one by one
Opaque layers (control)
the control: hidden layers the GPU never has to shade

At 1080p, one 60 Hz frame holds 11 full-screen shader passes on the A19: 2.2× the A15 and 3.7× the A12

Passes of the reference shader per frame at 60 Hz, higher is better: at 1080p and at native 4K (3840×2160).

  • A12
  • A15
  • A19
0510151080pA123A155A19112160p (4K)A12nothing fitsA151A192
  • A budget is the largest load whose 99th-percentile frame time stays within one 60 Hz frame, measured off-screen: it is GPU work per frame, not frames on screen.
The numbers as a table
ChipResolutionPasses per 60 Hz frameSpreadStatus
A121080p3all equal over 3 runs (CV 0 %)final
A151080p5all equal over 3 runs (CV 0 %)final
A191080p11all equal over 3 runs (CV 0 %)final
A122160p (4K)nothing fits—final
A152160p (4K)1all equal over 3 runs (CV 0 %)final
A192160p (4K)2all equal over 3 runs (CV 0 %)final

Why it slows down

Under constant load a chip heats up, and the system slows it down to keep it cool (thermal throttling). The A12 box has a fan; the A15 and A19 boxes do not. No temperature was read: what was measured is how much each box slows under constant load, and how it recovers after rest. Rest brings it back: after a 2-minute run, 4 minutes of rest returned the A19 to within 1 % of its rested speed.

Same load for 20 minutes: each chip slows while still reporting nominal (frames per second)

The same fixed load on every box (full-screen shader passes at 4 passes, 1080p), off-screen, from a rested box. Before its first change of thermal state the A15 had lost 17 % (by 8:40) and the A19 had lost 7 % (by 1:50); the A12 lost 23 % and never left nominal.

  • A12
  • A15
  • A19 unit 1
  • A19 unit 2
  • first “fair”
  • first “serious”
  • first “critical”
  • 60 fps: above it is off-screen throughput
0501001502002500 min2468101214161820
  • Each point is the mean of a 10-second window. tvOS reports four thermal states (nominal, fair, serious, critical), not a temperature; the marks show the first second each box reported a hotter state.
  • Frames per second above 60 are off-screen throughput: the work is drawn as fast as the GPU allows, never shown.
  • The dashed line is the second A19 unit, on the same load: the two units track each other.
  • Every 20-minute run here is a single run.
The numbers as a table
BoxFirst 5 s, fpsAt 1 min, fpsLast minute, fpsLast ÷ first 5 sThermal stateRest at nominal before
A1269.866.253.977 %nominal throughout300 s
A15108.489.274.068 %serious at 8:49300 s
A19 unit 1215.0201.5124.358 %serious at 1:59, critical at 5:39300 s
A19 unit 2212.4199.0128.460 %fair at 1:56, serious at 2:01, critical at 5:31300 s

After 20 minutes, pixel work keeps 62–85 % of its one-minute speed; opaque overdraw and meshes keep 83–100 %

Frames per second in the last minute of a 20-minute run ÷ frames per second at the one-minute mark (seconds 45–75), each test at about the chip’s own 1080p budget load. Sorted from the test that loses most.

  • A12
  • A15
  • A19
020406080100100 % = one-minute speedBlended full-screen layersA1278 %A1578 %A1964 %Full-screen shader passesA1282 %A1584 %A1962 %Sprites, one draw eachA1292 %A1577 %A1962 %Sprites, instancedA1294 %A1577 %A1961 %Alpha-tested layersA1285 %A1582 %A1967 %Meshes, one draw eachA1299 %A1591 %A1983 %Meshes, instancedA1296 %A15100 %A1986 %Opaque layers (control)A12100 %A15100 %A19100 %
  • Pixel work here means the shader passes and the blended and alpha-tested layers; the opaque layers are mostly removed before shading, and the meshes are limited by triangle throughput.
  • Each 20-minute budget in the tables is the one-minute budget × this share, for its own test. Other resolutions use the 1080p share.
  • A share above 100 % is run-to-run noise on a test that loses nothing, so it is drawn and counted as 100 %; the table keeps the measured value.
  • Every 20-minute run here is a single run.
The numbers as a table
TestChipLoadAt 1 min, fpsLast minute, fpsKeptThermal state
Blended full-screen layersA1273 layers73.257.478 %nominal throughout
Blended full-screen layersA15108 layers67.052.578 %serious at 9:46
Blended full-screen layersA19152 layers75.448.064 %serious at 1:57, critical at 4:37
Full-screen shader passesA123 passes86.971.682 %nominal throughout
Full-screen shader passesA155 passes71.459.784 %serious at 8:36
Full-screen shader passesA1911 passes73.745.662 %serious at 2:19, critical at 6:44
Sprites, one draw eachA1223,750 sprites69.764.392 %nominal throughout
Sprites, one draw eachA1535,000 sprites69.253.177 %serious at 10:35
Sprites, one draw eachA1958,000 sprites65.941.262 %fair at 1:53, serious at 1:58, critical at 4:23
Sprites, instancedA1223,500 sprites71.967.794 %nominal throughout
Sprites, instancedA1540,000 sprites66.951.677 %serious at 11:48
Sprites, instancedA1956,400 sprites73.944.761 %fair at 1:33, serious at 1:38, critical at 2:58
Alpha-tested layersA1296 layers67.457.585 %nominal throughout
Alpha-tested layersA15128 layers69.957.082 %serious at 7:58
Alpha-tested layersA19166 layers72.748.367 %serious at 1:27, critical at 3:42
Meshes, one draw eachA126,300 meshes87.686.799 %nominal throughout
Meshes, one draw eachA1512,000 meshes67.962.091 %serious at 13:18
Meshes, one draw eachA1920,400 meshes64.153.383 %critical at 7:08
Meshes, instancedA1210,375 meshes64.261.496 %nominal throughout
Meshes, instancedA1511,875 meshes63.365.1103 %serious at 14:13
Meshes, instancedA1921,000 meshes63.955.286 %serious at 9:08, critical at 9:13
Opaque layers (control)A121,040 layers62.962.9100 %nominal throughout
Opaque layers (control)A152,240 layers60.863.3104 %nominal throughout
Opaque layers (control)A193,264 layers63.563.5100 %nominal throughout

On the A19, rendering at 1080p and upscaling with MetalFX spatial fits 4× the shader passes of native 4K (2× on the A15)

Passes of the reference shader per 60 Hz frame when the frame is drawn at 1080p or 1440p and scaled to 3840×2160 by MetalFX, against drawing it at native 4K. Higher is better.

  • A12
  • A15
  • A19
02468A12native 4Knothing fits1440p + spatialnothing fits1080p + spatial11440p + temporalnot supported on this chip1080p + temporalnot supported on this chipA15native 4K11440p + spatial11080p + spatial21440p + temporalnothing fits1080p + temporalnothing fitsA19native 4K21440p + spatial51080p + spatial81440p + temporal11080p + temporal3
  • The scaler’s own cost is inside each budget: a frame is the passes at the input size, one MetalFX scale to 4K and a one-pixel read-back. “Nothing fits” means not even one pass does.
  • Timed alone and settled, from 1080p to 4K, the spatial scaler costs 9.2 ms on the A12 (55 % of a 60 Hz frame), 7.0 ms on the A15 (42 %) and 3.1 ms on the A19 (18 %); the temporal scaler costs 20.1 ms on the A15 (121 % of a 60 Hz frame: more than a whole frame) and 9.9 ms on the A19 (60 %), and is not supported on the A12.
  • Inside a real frame, the spatial scaler from 1080p plus the frame’s fixed cost (the intercept of a straight-line fit of frame time against passes) is 10.0 ms of the 16.7 on the A12, 8.2 ms on the A15 and 3.6 ms on the A19.
  • The temporal scaler was timed with still, unjittered content; its cost with real motion was not checked.
The numbers as a table
ChipDrawn asPasses per 60 Hz frameSpreadStatus
A12native 4Knothing fits—final
A121440p + spatialnothing fits—final
A121080p + spatial1all equal over 3 runs (CV 0 %)final
A121440p + temporalnot supported on this chip—final
A121080p + temporalnot supported on this chip—final
A15native 4K1all equal over 3 runs (CV 0 %)final
A151440p + spatial1all equal over 3 runs (CV 0 %)final
A151080p + spatial2all equal over 3 runs (CV 0 %)final
A151440p + temporalnothing fits—final
A151080p + temporalnothing fits—final
A19native 4K2all equal over 3 runs (CV 0 %)final
A191440p + spatial54–5 over 3 runs (CV 9.4 %)final
A191080p + spatial8all equal over 3 runs (CV 0 %)final
A191440p + temporal1all equal over 3 runs (CV 0 %)final
A191080p + temporal3all equal over 3 runs (CV 0 %)final
The MetalFX scaler timed alone, ms per frame
chipboxscalerinput sizeoutput sizestatusunitcost ms from restcost ms settledgpu cost ms settledwith scaler ms from restwith scaler ms settledwithout ms from restwithout ms settledshare of 60hz frame settledoutput megapixels per sthermalinterruptionsndate
A12A12spatial1920x10803840x2160measuredms per frame9.239.229.2211.1811.181.961.960.55742nominal→nominal012026-11-14
A12A12spatial2560x14403840x2160measuredms per frame9.329.289.2812.7512.753.423.460.56651nominal→nominal012026-11-14
A12A12temporal1920x10803840x2160unsupportedms per frame——————————012026-11-14
A12A12temporal2560x14403840x2160unsupportedms per frame——————————012026-11-14
A15A15spatial1920x10803840x2160measuredms per frame5.736.956.956.898.461.161.520.421,032nominal→nominal012026-11-14
A15A15spatial2560x14403840x2160measuredms per frame5.466.966.967.479.612.012.650.42910nominal→nominal012026-11-14
A15A15temporal1920x10803840x2160measuredms per frame14.5620.0920.115.7221.611.161.521.21409nominal→nominal012026-11-14
A15A15temporal2560x14403840x2160measuredms per frame16.9421.2921.2918.9523.942.012.651.28363nominal→nominal012026-11-14
A19A19 unit 1spatial1920x10803840x2160measuredms per frame2.843.063.063.553.790.70.730.182,253nominal→nominal012026-11-14
A19A19 unit 1spatial2560x14403840x2160measuredms per frame2.833.13.14.084.381.251.280.191,950nominal→nominal012026-11-14
A19A19 unit 1temporal1920x10803840x2160measuredms per frame9.39.929.921010.650.70.730.6793nominal→nominal012026-11-14
A19A19 unit 1temporal2560x14403840x2160measuredms per frame10.6211.4111.4111.8712.691.251.280.69672nominal→nominal012026-11-14

On one thread each chip runs about 1.6× as fast as the one before it (CoreMark, iterations per second)

Settled figures (after the first 30 seconds), higher is better; each panel has its own unit and its own zero-based axis. On all threads, Jolt physics steps the same scene 1.5× as fast on the A15 as on the A12, and 1.4× as fast on the A19 as on the A15.

  • A12
  • A15
  • A19
CoreMark, 1 thread iterations/s
020K40K60KA1220,743A1533,313A1952,273
Jolt physics, 1 thread steps/s
050100150A1252A1580.7A19114
Jolt physics, all threads steps/s
0100200300A12102A15154A19223
Memory bandwidth, 1 thread MB/s
020K40K60KA1220,591A1531,659A1956,297
Memory bandwidth, all threads MB/s
020K40K60KA1219,449A1527,198A1952,854
zstd-3 decompression, 1 thread MB/s
05001,0001,500A12776A151,041A191,269
zstd-3 decompression, all threads MB/s
02,0004,0006,000A122,391A152,866A194,444
lz4 decompression, 1 thread MB/s
02,0004,0006,000A122,789A154,120A195,393
lz4 decompression, all threads MB/s
05K10K15K20KA1210,367A1511,461A1917,781
  • Core counts as each box reports them: A12: 6 cores (2 performance + 4 efficiency); A15: 5 cores (2 performance + 3 efficiency); A19: 5 cores (2 performance + 3 efficiency). All-thread figures are not per core.
  • Memory bandwidth and decompression use MB = 10^6 bytes; decompression is MB/s of output, over a generated 24 MiB corpus of text, JSON, vertex data, texture and audio.
  • Every CPU figure here is a single run.
The numbers as a table
TestA12, settledA15, settledA19, settled
CoreMark, 1 thread (iterations/s)20,74333,31352,273
Jolt physics, 1 thread (steps/s)5280.7114
Jolt physics, all threads (steps/s)102154223
Memory bandwidth, 1 thread (MB/s)20,59131,65956,297
Memory bandwidth, all threads (MB/s)19,44927,19852,854
zstd-3 decompression, 1 thread (MB/s)7761,0411,269
zstd-3 decompression, all threads (MB/s)2,3912,8664,444
lz4 decompression, 1 thread (MB/s)2,7894,1205,393
lz4 decompression, all threads (MB/s)10,36711,46117,781

All-thread physics for 20 minutes: the A12 keeps 97 % of its minute-1 speed, the A15 78 %, the A19 67 % (steps per second)

Jolt physics on every core for 20 minutes from a rested box; each point is a 10-second window. Stability (worst loop ÷ best loop): A12 91.9 %, A15 64.3 % and A19 58.4 %.

  • A12
  • A15
  • A19
  • first “serious”
  • first “critical”
01002003000 min2468101214161820
  • The same work on every chip: Jolt’s result hash was identical in every run.
  • A single 20-minute run per chip.
The numbers as a table
ChipFirst loopMinute 1 medianBest loopWorst loopStabilityLast minute ÷ minute 1Thermal state
A12100.9101.7102.594.291.9 %97 %nominal throughout
A15187.5158.0187.5120.564.3 %78 %serious at 4:09
A19274.2243.9274.2160.158.4 %67 %serious at 1:24, critical at 2:26

Every box lets one app use about 2,098 MB, whether it has 2.9, 3.9 or 7.9 GB of memory

MB = 2^20 bytes. The limit is the app’s footprint plus the memory the system still reports available to it, read at every step of an allocate-until-killed test.

  • A12
  • A15
  • A19
02K4K6K8K10KPhysical memoryA122.94 GBA153.88 GBA197.87 GBWhat one app can use before tvOS ends itA122,098 MBA152,098 MBA192,098 MB
  • The memory warning arrives first: plan on less than the limit. In these runs the first warning came with 1,088–1,664 MB held.
  • Metal textures, buffers and render targets count exactly as heap memory does, and memory compression does not raise the limit.
The numbers as a table
BoxWhat was allocatedHeld at the last step, MBFirst warning, MB4096² RGBA8 textures4K RGBA16F targetsRuns
A12Metal buffers2,0481,34432321
A12heap, random-filled2,0481,088–1,40832323
A12heap, zero-filled2,0481,66432321
A124K rgba16Float render targets2,0481,34432321
A12Metal textures, 4096² RGBA81,9881,283–1,60331313
A15Metal buffers2,0481,66432321
A15heap, random-filled2,0481,66432323
A15heap, zero-filled2,0481,66432321
A154K rgba16Float render targets2,0561,60632321
A15Metal textures, 4096² RGBA82,0001,54831313
A19 unit 1heap, random-filled2,0481,66432321
A19 unit 14K rgba16Float render targets2,0561,60632321
A19 unit 1Metal textures, 4096² RGBA82,0001,54831311

A classic deferred frame at 4K takes 3.2 ms on the A12, 2.6 on the A15 and 1.9 on the A19: at most 19 % of a 60 Hz frame

Apple’s “Rendering a scene with deferred lighting” sample, unmodified shaders, settled milliseconds per frame (lower is better); 67,202 triangles submitted per frame.

  • A12
  • A15
  • A19
012341080pA121.36 msA150.88 msA190.62 ms2160p (4K)A123.24 msA152.59 msA191.86 ms
  • A light frame, useful as what a classic deferred frame with many lights costs, not as a stress test. The rates behind it are far above 60 fps: off-screen throughput, not frames on screen.
  • Every scene figure here is a single run.
The numbers as a table
ChipResolutionFrom rest, msSettled, msSettled, fps (off-screen)CPU encode, ms
A121080p1.371.367350.42
A122160p (4K)3.223.243090.57
A151080p0.670.881,1350.26
A152160p (4K)1.952.593860.27
A191080p0.570.621,6150.15
A192160p (4K)1.721.865370.17

Budgets to plan with

For each chip, test and resolution: the budget at 60 Hz, the largest load that held 60 frames a second for a 40 s hold on a rested box; its spread over repeated launches; and the 20-minute budget, the budget times the share of speed that test keeps after 20 minutes on that chip. Plan against the 20-minute figure. It is measured where the 20-minute run was at that resolution and scaled from 1080p where the 1080p share was applied to another resolution. Rows that say “not measured yet” are still to come from the campaign.

How the chips compare, test by test

One chip’s budget ÷ another’s, on a cool box and after 20 minutes (each budget × the share of speed that test keeps on that chip, at most 100 %). A dash means one of the two chips has no measured budget there yet.

At 1080p

TestA19 ÷ A15, coolafter 20 minA19 ÷ A12, coolafter 20 minA15 ÷ A12, coolafter 20 min
Full-screen shader passes2.2×1.6×3.7×2.8×1.7×1.7×
Blended full-screen layers1.4×1.1×2.0×1.6×1.4×1.4×
Alpha-tested layers1.3×1.04×1.8×1.4×1.4×1.3×
Sprites, instanced1.3×1.04×2.1×1.3×1.6×1.3×
Sprites, one draw each1.5×1.2×2.2×1.5×1.4×1.2×
Meshes, instanced1.8×1.5×2.1×1.9×1.2×1.2×
Meshes, one draw each1.6×1.5×2.1×1.7×1.3×1.2×

At 2160p (4K)

TestA19 ÷ A15, coolafter 20 minA19 ÷ A12, coolafter 20 minA15 ÷ A12, coolafter 20 min
Full-screen shader passes2.0×1.5×————
Blended full-screen layers1.4×1.2×2.3×1.9×1.6×1.6×
Alpha-tested layers1.2×1.01×1.8×1.4×1.5×1.4×
Sprites, instanced1.4×1.1×2.3×1.5×1.6×1.3×
Sprites, one draw each1.4×1.1×2.3×1.6×1.7×1.4×
Meshes, instanced1.7×1.4×2.3×2.0×1.4×1.4×
Meshes, one draw each1.6×1.5×2.2×1.8×1.3×1.2×

Full-screen shader passes

full-screen passes of the reference shader

ChipResolutionBudget at 60 HzSpread20-minute budgetStatus
A121080p3 passesall equal over 3 runs (CV 0 %)2 passesmeasuredfinal, 2026-11-15
A121440p2 passesall equal over 3 runs (CV 0 %)1 passscaled from 1080pfinal, 2026-11-15
A122160p (4K)nothing fits——final, 2026-11-15
A151080p5 passesall equal over 3 runs (CV 0 %)4 passesmeasuredfinal, 2026-11-15
A151440p3 passesall equal over 3 runs (CV 0 %)2 passesscaled from 1080pfinal, 2026-11-15
A152160p (4K)1 passall equal over 3 runs (CV 0 %)0 passesscaled from 1080pfinal, 2026-11-15
A191080p11 passesall equal over 3 runs (CV 0 %)6 passesmeasuredfinal, 2026-11-15
A191440p6 passesall equal over 3 runs (CV 0 %)3 passesscaled from 1080pfinal, 2026-11-15
A192160p (4K)2 passesall equal over 3 runs (CV 0 %)1 passscaled from 1080pfinal, 2026-11-15

Blended full-screen layers

ChipResolutionBudget at 60 HzSpread20-minute budgetStatus
A121080p75 layers72–76 over 3 runs (CV 2.3 %)58 layersmeasuredfinal, 2026-11-15
A121440p41 layersall equal over 3 runs (CV 0 %)32 layersscaled from 1080pfinal, 2026-11-15
A122160p (4K)18 layers18–19 over 3 runs (CV 2.6 %)14 layersscaled from 1080pfinal, 2026-11-15
A151080p108 layersall equal over 3 runs (CV 0 %)84 layersmeasuredfinal, 2026-11-15
A151440p64 layers63–64 over 3 runs (CV 0.7 %)50 layersscaled from 1080pfinal, 2026-11-15
A152160p (4K)29 layersall equal over 3 runs (CV 0 %)22 layersscaled from 1080pfinal, 2026-11-15
A191080p150 layers150–152 over 3 runs (CV 0.6 %)95 layersmeasuredfinal, 2026-11-15
A191440p86 layers86–87 over 3 runs (CV 0.6 %)54 layersscaled from 1080pfinal, 2026-11-15
A192160p (4K)42 layers40–42 over 3 runs (CV 2.2 %)26 layersscaled from 1080pfinal, 2026-11-15

Opaque layers (control)

opaque full-screen layers (control)

ChipResolutionBudget at 60 HzSpread20-minute budgetStatus
A121080p1,024 layers1,024–1,040 over 3 runs (CV 0.7 %)1,024 layersmeasuredfinal, 2026-11-15
A121440p584 layersall equal over 3 runs (CV 0 %)584 layersscaled from 1080pfinal, 2026-11-15
A122160p (4K)260 layersall equal over 3 runs (CV 0 %)260 layersscaled from 1080pfinal, 2026-11-15
A151080p2,240 layersall equal over 3 runs (CV 0 %)2,240 layersmeasuredfinal, 2026-11-15
A151440p1,312 layersall equal over 3 runs (CV 0 %)1,312 layersscaled from 1080pfinal, 2026-11-15
A152160p (4K)544 layers536–552 over 3 runs (CV 1.2 %)544 layersscaled from 1080pfinal, 2026-11-15
A191080p3,264 layers3,200–3,264 over 3 runs (CV 0.9 %)3,264 layersmeasuredfinal, 2026-11-15
A191440p1,952 layersall equal over 3 runs (CV 0 %)1,952 layersscaled from 1080pfinal, 2026-11-15
A192160p (4K)848 layersall equal over 3 runs (CV 0 %)848 layersscaled from 1080pfinal, 2026-11-15

Alpha-tested layers

ChipResolutionBudget at 60 HzSpread20-minute budgetStatus
A121080p96 layers93–96 over 3 runs (CV 1.5 %)81 layersmeasuredfinal, 2026-11-15
A121440p56 layers54–56 over 3 runs (CV 1.7 %)47 layersscaled from 1080pfinal, 2026-11-15
A122160p (4K)25 layersall equal over 3 runs (CV 0 %)21 layersscaled from 1080pfinal, 2026-11-15
A151080p132 layersall equal over 3 runs (CV 0 %)107 layersmeasuredfinal, 2026-11-15
A151440p79 layers79–80 over 3 runs (CV 0.6 %)64 layersscaled from 1080pfinal, 2026-11-15
A152160p (4K)37 layers35–37 over 3 runs (CV 2.6 %)30 layersscaled from 1080pfinal, 2026-11-15
A191080p168 layers167–172 over 3 runs (CV 1.3 %)111 layersmeasuredfinal, 2026-11-15
A191440p100 layers100–102 over 3 runs (CV 0.9 %)66 layersscaled from 1080pfinal, 2026-11-15
A192160p (4K)46 layers44–46 over 3 runs (CV 2.1 %)30 layersscaled from 1080pfinal, 2026-11-15

Sprites, instanced

ChipResolutionBudget at 60 HzSpread20-minute budgetStatus
A121080p27,000 sprites24,761–27,000 over 5 runs (CV 3.3 %)25,411 spritesmeasuredfinal, 2026-11-15
A121440p12,000 spritesall equal over 3 runs (CV 0 %)11,293 spritesscaled from 1080pfinal, 2026-11-15
A122160p (4K)6,625 sprites6,500–6,625 over 3 runs (CV 0.9 %)6,235 spritesscaled from 1080pfinal, 2026-11-15
A151080p42,000 spritesall equal over 3 runs (CV 0 %)32,400 spritesmeasuredfinal, 2026-11-15
A151440p22,000 sprites20,672–22,000 over 3 runs (CV 2.9 %)16,971 spritesscaled from 1080pfinal, 2026-11-15
A152160p (4K)10,750 spritesall equal over 3 runs (CV 0 %)8,292 spritesscaled from 1080pfinal, 2026-11-15
A191080p55,885 sprites55,849–56,429 over 3 runs (CV 0.5 %)33,831 spritesmeasuredfinal, 2026-11-15
A191440p31,239 sprites31,212–31,622 over 3 runs (CV 0.6 %)18,911 spritesscaled from 1080pfinal, 2026-11-15
A192160p (4K)15,392 sprites15,360–15,531 over 3 runs (CV 0.5 %)9,317 spritesscaled from 1080pfinal, 2026-11-15

Sprites, one draw each

ChipResolutionBudget at 60 HzSpread20-minute budgetStatus
A121080p25,500 sprites24,750–25,500 over 3 runs (CV 1.4 %)23,547 spritesmeasuredfinal, 2026-11-15
A121440p11,875 sprites11,750–11,875 over 3 runs (CV 0.5 %)10,965 spritesscaled from 1080pfinal, 2026-11-15
A122160p (4K)6,500 sprites6,187–6,500 over 3 runs (CV 2.3 %)6,002 spritesscaled from 1080pfinal, 2026-11-15
A151080p36,500 sprites36,500–37,000 over 3 runs (CV 0.7 %)28,003 spritesmeasuredfinal, 2026-11-15
A151440p22,000 spritesall equal over 3 runs (CV 0 %)16,879 spritesscaled from 1080pfinal, 2026-11-15
A152160p (4K)10,750 spritesall equal over 3 runs (CV 0 %)8,247 spritesscaled from 1080pfinal, 2026-11-15
A191080p55,612 sprites54,000–56,247 over 3 runs (CV 1.7 %)34,750 spritesmeasuredfinal, 2026-11-15
A191440p30,890 sprites30,818–30,916 over 3 runs (CV 0.1 %)19,302 spritesscaled from 1080pfinal, 2026-11-15
A192160p (4K)15,023 sprites15,017–15,028 over 3 runs (CV 0 %)9,387 spritesscaled from 1080pfinal, 2026-11-15

Meshes, instanced

meshes of ~1,000 triangles, instanced

ChipResolutionBudget at 60 HzSpread20-minute budgetStatus
A121080p10,000 meshes9,875–10,375 over 3 runs (CV 2.1 %)9,563 meshesmeasuredfinal, 2026-11-15
A121440p9,500 meshes9,375–9,500 over 3 runs (CV 0.6 %)9,085 meshesscaled from 1080pfinal, 2026-11-15
A122160p (4K)7,875 meshes7,875–8,000 over 3 runs (CV 0.8 %)7,531 meshesscaled from 1080pfinal, 2026-11-15
A151080p11,875 meshesall equal over 3 runs (CV 0 %)11,875 meshesmeasuredfinal, 2026-11-15
A151440p11,375 meshes11,375–11,500 over 3 runs (CV 0.5 %)11,375 meshesscaled from 1080pfinal, 2026-11-15
A152160p (4K)10,750 meshes10,750–11,000 over 3 runs (CV 1.1 %)10,750 meshesscaled from 1080pfinal, 2026-11-15
A191080p21,000 meshesall equal over 3 runs (CV 0 %)18,120 meshesmeasuredfinal, 2026-11-15
A191440p19,750 meshesall equal over 3 runs (CV 0 %)17,041 meshesscaled from 1080pfinal, 2026-11-15
A192160p (4K)17,750 meshesall equal over 3 runs (CV 0 %)15,316 meshesscaled from 1080pfinal, 2026-11-15

Meshes, one draw each

meshes of ~1,000 triangles, one draw each

ChipResolutionBudget at 60 HzSpread20-minute budgetStatus
A121080p9,800 meshes9,800–9,900 over 3 runs (CV 0.5 %)9,701 meshesmeasuredfinal, 2026-11-15
A121440p9,300 meshes9,200–9,300 over 3 runs (CV 0.5 %)9,206 meshesscaled from 1080pfinal, 2026-11-15
A122160p (4K)7,900 meshes7,900–8,100 over 3 runs (CV 1.2 %)7,820 meshesscaled from 1080pfinal, 2026-11-15
A151080p12,431 meshes12,000–12,907 over 3 runs (CV 3 %)11,350 meshesmeasuredfinal, 2026-11-15
A151440p11,400 meshesall equal over 3 runs (CV 0 %)10,409 meshesscaled from 1080pfinal, 2026-11-15
A152160p (4K)10,600 meshes10,400–10,922 over 3 runs (CV 2 %)9,678 meshesscaled from 1080pfinal, 2026-11-15
A191080p20,400 meshesall equal over 3 runs (CV 0 %)16,966 meshesmeasuredfinal, 2026-11-15
A191440p19,200 meshesall equal over 3 runs (CV 0 %)15,968 meshesscaled from 1080pfinal, 2026-11-15
A192160p (4K)17,200 meshesall equal over 3 runs (CV 0 %)14,304 meshesscaled from 1080pfinal, 2026-11-15

Rendering lower and upscaling to 4K

Shader passes per 60 Hz frame when the frame is drawn at 1080p or 1440p and scaled to 3840×2160 by MetalFX in the same frame.

Shader passes + MetalFX spatial to 4K

passes of the reference shader at the input size + MetalFX spatial to 3840x2160

ChipDrawn at, then to 4KBudget at 60 HzSpreadStatus
A121080p1 passall equal over 3 runs (CV 0 %)final, 2026-11-15
A121440pnothing fits—final, 2026-11-15
A151080p2 passesall equal over 3 runs (CV 0 %)final, 2026-11-15
A151440p1 passall equal over 3 runs (CV 0 %)final, 2026-11-15
A191080p8 passesall equal over 3 runs (CV 0 %)final, 2026-11-15
A191440p5 passes4–5 over 3 runs (CV 9.4 %)final, 2026-11-15

Shader passes + MetalFX temporal to 4K

passes of the reference shader at the input size + MetalFX temporal to 3840x2160

ChipDrawn at, then to 4KBudget at 60 HzSpreadStatus
A121080pnot supported on this chip—final, 2026-11-15
A121440pnot supported on this chip—final, 2026-11-15
A151080pnothing fits—final, 2026-11-15
A151440pnothing fits—final, 2026-11-15
A191080p3 passesall equal over 3 runs (CV 0 %)final, 2026-11-15
A191440p1 passall equal over 3 runs (CV 0 %)final, 2026-11-15

Method

Who measured, on what

One A12, one A15, two A19 units, measured by one developer with his own test app: native Metal tests written for this, with no code from a game or from another benchmark’s renderer. Every A19 figure is from A19 unit 1 unless it names A19 unit 2. Unit 2 ran the early tests only, and is there to show whether two units of the same box agree.

What each test does

  • Full-screen shader passes. Each pass is its own render pass over the whole screen that reads the previous pass and stores its result, so the GPU cannot skip any. The reference shader does, per pixel, 8 bilinear samples of the previous pass, then 32 loop iterations of two 4-wide multiply-adds and one 4-wide fract.
  • Blended layers. Alpha-blended full-screen layers in one render pass. Opaque layers are the control: the GPU removes hidden surfaces before shading, so they cost little. Alpha-tested layers discard some pixels.
  • Sprites. Textured quads, drawn instanced or with one draw call each; their size scales with the screen.
  • Meshes. Meshes of ~1,000 triangles, instanced, and the same with one draw call each. They are limited by triangle throughput, not by pixels.
  • Upscaling. Shader passes at 1080p or 1440p, one MetalFX scale (spatial or temporal) to 3840×2160, and a one-pixel read-back, all in the frame.
  • A real frame. Apple’s “Rendering a Scene with Deferred Lighting” sample inside the same off-screen loop, with its shaders unchanged.
  • CPU. CoreMark 1.0 under EEMBC’s run rules (one thread); Jolt Physics’ performance test (its ragdoll scene) on one thread and on all; a STREAM-style triad for memory bandwidth; and decompression with zstd, lz4 and Apple’s Compression framework over a generated 24 MiB corpus. One thread runs at user-interactive priority; all threads means one per logical core, with work handed out as threads free up.
  • Memory. Allocate in 64 MB steps, writing to every page and rendering into every texture so the memory is really in use, until tvOS ends the app. Every step is written to disk first, so the last good step survives the kill.

How a budget is found

Every GPU test runs off-screen with a fixed time step, so every run draws the same frames, with at most 3 frames in flight, timed by when each frame finishes. The load doubles until a step fails, then a bisection finds the edge. A step passes when the 99th-percentile frame interval of a 3 s probe is at most 16.0 ms (4 % headroom under one 60 Hz frame). The largest passing load must then hold 16.67 ms for one 40 s hold. There is one hold per launch: if it misses, the load is cut by the size of the miss and held again in a new launch, after another rest, because on a warming box a second hold in the same launch only chases the temperature. The figure is the median of the repeats; the campaign takes three launches per figure, and the tables say how many each figure has so far.

On a real 60 Hz display

Budgets are measured off-screen. To check that they hold on a screen, the same work was drawn at loads around the budget, then copied or scaled into the 3840×2160 drawable once per display refresh for a minute, counting the refreshes that showed no new frame. At each chip’s 1080p budget, 6 of 7 one-minute runs on a 60 Hz display missed no refresh at all; the worst missed 0.17 % of refreshes (A12, blended layers). About 10 % above the budget, 5 of 6 runs missed refreshes, up to 5.9 %. With CAMetalDisplayLink instead, runs at or below the budget missed up to 6 % of refreshes, so the runs that count use CADisplayLink.

The on-screen runs (49)
BoxTestDrawn atLoad× budgetDisplay linkRefreshesMissedMissed shareLongest run of misses
A15Full-screen shader passes1920×108040.8CAMetalDisplayLink899121.34 %3
A12Full-screen shader passes1920×108031CAMetalDisplayLink719243.34 %2
A12Full-screen shader passes1920×108031CAMetalDisplayLink3,596711.97 %3
A15Full-screen shader passes1920×108051CAMetalDisplayLink3,596581.61 %3
A19 unit 1Full-screen shader passes1920×1080111CAMetalDisplayLink3,59600 %0
A12Blended full-screen layers1920×1080730.97CAMetalDisplayLink3,5962145.95 %3
A15Blended full-screen layers1920×10801081CAMetalDisplayLink3,596391.09 %2
A12Full-screen shader passes2112×108031.1CAMetalDisplayLink3,5961825.06 %3
A15Full-screen shader passes2112×108051.1CAMetalDisplayLink3,5961624.51 %4
A19 unit 1Blended full-screen layers1920×10801521.01CAMetalDisplayLink3,59600 %0
A12Blended full-screen layers1920×1080801.07CAMetalDisplayLink3,59648913.6 %3
A15Blended full-screen layers1920×10801191.1CAMetalDisplayLink3,5962346.51 %4
A12Full-screen shader passes1728×108030.9CAMetalDisplayLink3,596661.84 %2
A15Full-screen shader passes1728×108050.9CAMetalDisplayLink3,596381.06 %3
A19 unit 1Full-screen shader passes2112×1080111.1CAMetalDisplayLink3,596942.61 %1
A12Blended full-screen layers1920×1080660.88CAMetalDisplayLink3,5961694.7 %3
A15Blended full-screen layers1920×1080970.9CAMetalDisplayLink3,596330.92 %2
A19 unit 1Blended full-screen layers1920×10801671.11CAMetalDisplayLink3,59650.14 %1
A15Full-screen shader passes1920×108010.2CAMetalDisplayLink3,596471.31 %2
A12Full-screen shader passes1920×108010.33CAMetalDisplayLink3,5961514.2 %3
A15Blended full-screen layers1920×1080860.8CAMetalDisplayLink3,596641.78 %2
A12Blended full-screen layers1920×1080580.77CAMetalDisplayLink3,596551.53 %2
A19 unit 1Full-screen shader passes1728×1080110.9CAMetalDisplayLink3,59600 %0
A15Full-screen shader passes1920×108040.8CAMetalDisplayLink3,596371.03 %3
A12Full-screen shader passes1920×108020.67CAMetalDisplayLink3,5961183.28 %3
A15Blended full-screen layers1920×1080650.6CAMetalDisplayLink3,596571.59 %3
A12Blended full-screen layers1920×1080440.59CAMetalDisplayLink3,5961032.86 %3
A19 unit 1Blended full-screen layers1920×10801370.91CAMetalDisplayLink3,59610.03 %1
A12Full-screen shader passes1920×108010.33CADisplayLink3,59600 %0
A12Full-screen shader passes1920×108010.33CAMetalDisplayLink3,5961042.89 %2
A15Full-screen shader passes1920×108010.2CADisplayLink3,59600 %0
A12Full-screen shader passes1920×108031CADisplayLink3,59600 %0
A15Full-screen shader passes1920×108051CADisplayLink3,59600 %0
A12Blended full-screen layers1920×1080730.97CADisplayLink3,59660.17 %1
A15Blended full-screen layers1920×10801081CADisplayLink3,59600 %0
A12Full-screen shader passes2112×108031.1CADisplayLink3,59600 %0
A15Full-screen shader passes2112×108051.1CADisplayLink3,5961143.17 %1
A12Blended full-screen layers1920×1080801.07CADisplayLink3,5961062.95 %1
A15Blended full-screen layers1920×10801191.1CADisplayLink3,5962115.87 %1
A12Full-screen shader passes1728×108030.9CADisplayLink3,59600 %0
A15Full-screen shader passes1728×108050.9CADisplayLink3,59600 %0
A12Blended full-screen layers1920×1080660.88CADisplayLink3,59600 %0
A15Blended full-screen layers1920×1080970.9CADisplayLink3,59600 %0
A12Full-screen shader passes1920×108031CADisplayLink89900 %0
A15Full-screen shader passes1920×108040.8CADisplayLink89900 %0
A19 unit 1Full-screen shader passes1920×1080111CADisplayLink3,59600 %0
A19 unit 1Blended full-screen layers1920×10801521.01CADisplayLink3,59600 %0
A19 unit 1Full-screen shader passes2112×1080111.1CADisplayLink3,596772.14 %1
A19 unit 1Blended full-screen layers1920×10801671.11CADisplayLink3,59640.11 %1

Two figures matter, because no chip holds its first seconds: from rest (the first seconds on a rested box, what short benchmarks elsewhere report) and settled (after the first half-minute). The budgets are settled figures; the 20-minute figures go further.

Temperature

Heat is the easiest way to get a wrong number out of these boxes, so this is what was done about it.

  • What was controlled. Before any warm-up or timing, the app sat idle, drawing nothing but a label, until tvOS had reported its normal thermal state for a set time. A box that would not settle was not measured: the run was refused, not taken hot. Every budget launch followed 240 s at nominal. Before the 20-minute runs, CPU tests and scene runs the waits actually were 120–900 s on the A12, 120–989 s on the A15, 120–1,070 s on the A19.
  • What the signal is, and is not. tvOS gives an app four thermal states, nominal, fair, serious and critical, and no temperature. “Nominal” is a range. That the range was narrow enough here rests on the two A19 units agreeing. On the 16 budgets both units ran, the median difference was 0.4 % and the largest 9.8 % (sprites, one draw each at 4K). In the same 20-minute run they were 1.2 % apart at one minute and 3.3 % apart in the last minute, and first reported “serious” at 1:59 and 2:01. The three repeats of every budget are whole passes over the test list, taken hours apart, so drift between them shows up as spread between repeats: Over the 78 budgets with repeats so far, the spread between repeats (coefficient of variation) is 0.1 % at the median and 9.4 % at most. And nominal is not rested: after 2 minutes of full load (it reported serious at 1:54), A19 unit 2 took 69 s to read nominal again; just back at nominal it ran 2.7 % slow against its own rested run, and after 240 s at nominal it was still 0.9 % slow (one unit, one series of 6 runs, a Debug build). That is why every launch rests first, and longer after a long run.
    The rest measurements
    chipboxroletestloadsizesecondsstate at launchseconds to reach nominalrest at nominal sfpsfps first secondfps last secondvs rested referencefirst serious sthermal endbuildinterruptionsdatetime
    A19A19 unit 2rested referencepass41920x108010nominal0202122132111—nominalDebug02026-11-140832
    A19A19 unit 2heating runpass41920x1080120nominal0202002141900.95114seriousDebug02026-11-140838
    A19A19 unit 2recoverypass41920x108010serious68.6202062091990.97—nominalDebug02026-11-140842
    A19A19 unit 2recoverypass41920x108010nominal0602072092000.98—nominalDebug02026-11-140844
    A19A19 unit 2recoverypass41920x108010nominal01202092102070.99—nominalDebug02026-11-140845
    A19A19 unit 2recoverypass41920x108010nominal02402102112080.99—nominalDebug02026-11-140848
  • Each chip slows while still reporting nominal. On the same fixed load, frames per second fell 5 % below the first ten seconds for the A12 within 60 s, the A15 within 20 s and the A19 within 30 s of starting, still reporting nominal. Over the run, the loss was the A12 23 % over the whole run without leaving nominal, the A15 17 % before it reported anything hotter (at 8:40) and the A19 7 % before it reported anything hotter (at 1:50). An app cannot wait for the thermal state to change before shedding load. This is why the results are given from rest and settled, and why the chart of frames per second against time is shown.
  • What heat does. The short-test numbers are a cool box’s numbers, taken about a minute into the load. The 20-minute runs show what is left once the box is warm, as their own charts and their own column in the tables, never mixed into the short-test charts.
  • What differs between the boxes. The A12 model has a fan; the A15 and A19 models do not. One of the two A19 units, A19 unit 1, the one every A19 figure comes from, reported its highest thermal state while idle where it sits, before any test. Placement and ventilation matter as much as the chip.
  • The raw data. Every run records the thermal state for every second. The downloadable series gives it for each 10-second window (the hottest state reported in it), and each 20-minute run lists when it first reported each state.

What was left out, and why

  • Runs in which the app lost the screen (a system prompt or the screen saver on top of it) are not used.
  • A hold that missed is not a budget; a load that never confirmed shows as “not confirmed”.
  • Launches that tvOS ended before anything was measured were run again; nothing from them is used.
  • Only Release builds count: a Debug build changes the CPU-bound tests a lot.
  • A19 unit 2 was unavailable after the early tests, so the later A19 figures are unit 1 alone.

What was not measured

  • A heavy scene (a port of Apple’s “Modern Rendering with Metal” sample) is planned and has not been run.
  • Ray tracing and mesh shaders: the boxes report whether they have them, but their speed was not measured.
  • The MetalFX temporal scaler with real motion; it was timed with still, unjittered content.
  • Box2D physics: the results are in the CPU download but are not charted here.
  • Other streaming boxes and phones: they are not compared.

The three boxes

What each box reports to an app at run time, read by the test app itself (except cooling, which comes from a teardown and press reviews). MB = 2^20 bytes. The A19 column is A19 unit 1; unit 2 reports the same figures.

A12A15A19
tvOS (build)27.2.1 (24K157)27.2 (24K122)A19 unit 1: 27.2.1 (24K157); A19 unit 2: 27.2.1 (24K157)
Model identifierAppleTV11,1AppleTV14,1AppleTV18,1
Memory2.94 GB3.88 GB7.87 GB
What one app can use2,098 MB2,098 MB2,098 MB
Available to the app at launch2,089 MB2,089 MB2,091 MB
CPU cores6 (2 performance + 4 efficiency)5 (2 performance + 3 efficiency)5 (2 performance + 3 efficiency)
L2 cache8 MiB performance, 2 MiB efficiency12 MiB performance, 4 MiB efficiency8 MiB performance, 4 MiB efficiency
Coolingfan (iFixit teardown)no fan (press reviews; no teardown found)no fan (press reviews; no teardown found)
GPUApple A12 GPUApple A15 GPUApple A19 GPU
Highest Apple GPU familyapple5apple8apple10
Apple GPU families supportedapple1, apple2, apple3, apple4, apple5apple1, apple2, apple3, apple4, apple5, apple6, apple7, apple8apple1, apple2, apple3, apple4, apple5, apple6, apple7, apple8, apple9, apple10
Metal 3noyesyes
Metal 4noyesyes
Ray tracingnoyesyes
Unified memoryyesyesyes
MetalFX spatial scaleryesyesyes
MetalFX temporal scalernoyesyes
MetalFX framework loadsyesyesyes
Max buffer length, MB1,5362,0484,096
Recommended max working set, MB2,0482,7315,461
GPU counter setstimestamptimestamptimestamp
Counter sampling at blit boundarynonono
Counter sampling at dispatch boundarynonono
Counter sampling at draw boundarynonono
Counter sampling at stage boundaryyesyesyes
Counter sampling at tile dispatch boundarynonono
Native screen3840×21603840×21603840×2160
Max frames per second606060

Raw data

The files behind every chart and table on this page, as published on 16 November 2026.

Schema tvbench-data-1. Every row carries how it was measured, its date and its number of runs (n). MB: 2^20 bytes, except STREAM and compression (10^6 bytes). Frames per second above 60: off-screen throughput, not frames on screen. Provisional: a budget row not yet measured in the requested campaign: the latest line outside it is shown instead.

For coding agents

The same findings, written as rules an AI coding agent can load while it builds or tunes a tvOS app or game: what fits in a frame on each chip, what heat takes away, the memory limit, what each box supports and how to tell the boxes apart at run time, with each rule’s caveat. It is for developers who work with an agent such as Claude Code and want it to plan against these budgets instead of guessing. Pack version 1.0, from the data of 16 November 2026.

  • SKILL.mdThe Claude Code skill: the rules, and when to use them, 15 KB
  • apple-tv-4k-budgets.mdThe same rules as plain Markdown, for other tools (AGENTS.md style), 14 KB
  • budgets.jsonThe full tables the skill points to, 38 KB

To install the skill in Claude Code, save SKILL.md and budgets.json into the folder .claude/skills/apple-tv-4k-budgets/ in a project (or ~/.claude/skills/apple-tv-4k-budgets/ to have it in every project):

mkdir -p .claude/skills/apple-tv-4k-budgets
curl -o .claude/skills/apple-tv-4k-budgets/SKILL.md https://icallwindowseat.com/tvbench/pack/SKILL.md
curl -o .claude/skills/apple-tv-4k-budgets/budgets.json https://icallwindowseat.com/tvbench/pack/budgets.json

Claude Code can then load it on its own whenever a task matches the skill’s description, such as building or tuning a tvOS app or choosing quality tiers per Apple TV model.

Changelog

  1. Data file regenerated (campaign final-1115b, budget runs through 16 November 2026).
  2. First version of the page, built while the final measurement campaign is still running: figures marked provisional are from earlier single runs and are being replaced.
  3. The final measurement campaign finished on all three chips. Every budget is now the median of three or more confirmed repeats, and no figure is provisional.

Measured by Josh, who makes I Call Window Seat and RevealHaven. Neither app was used as a test.