The spec table
One row per buyable configuration family. Specs checked 2026-10-15; prices carry their own dates, re-checked 2026-10-20. Memory prices moved every few weeks through 2026, so a price is only true on its date.
Apple Silicon Macs
| Machine | Memory | Usable for the model | Bandwidth | Published compute | Power | Prices |
|---|---|---|---|---|---|---|
| Mac Studio, M5 Ultra (30-core CPU / 64-core GPU, or 36 / 80) Metal, MLX; no CUDA |
96, 256 or 512 GB; Apple lists 256 and 512 GB with the 36/80-core chip; the 512 GB option ships in 10-12 weeks (2026-10-30) spec [S1] [S12] [S62] | 75% of memory assumption | 1,200 GB/s spec [S1] | not published; Apple: 'up to 4.3x the peak AI compute' of M3 Ultra (Apple's test), relative only spec [S2] | 9 W idle / 385 W max for 36/80-core, 512 GB, 16 TB; 480 W enclosure max continuous spec [S1] [S7] whole computer |
$10,799: 36/80-core, 256 GB / 1 TB (Apple configurator: 512 GB/16 TB $29,799 minus the 256 GB and 1 TB steps) (2026-10-30) spec [S62] not checked; checked 2026-10-30 $22,299: 36/80-core, 512 GB / 1 TB (Apple configurator: $29,799 minus the 1 TB step of $7,500) (2026-10-30) spec [S62] ships in 10-12 weeks; checked 2026-10-30 $29,799: 36/80-core, 512 GB / 16 TB (top configuration) (2026-10-30) spec [S62] ships in 10-12 weeks; checked 2026-10-30 |
| Mac Studio, M5 Max 18-core CPU / 40-core GPU Metal, MLX; no CUDA |
48, 64 or 128 GB spec [S1] | 75% of memory assumption | 614 GB/s spec [S1] | not published; Apple: LLM prompt processing in LM Studio '3.9x faster than M4 Max' (Apple's test), relative only spec [S2] | 480 W enclosure max continuous; Apple's power page lists only the 32-core configuration spec [S1] [S7] whole computer |
|
| MacBook Pro 14/16, M5 Max 40-core GPU Metal, MLX; no CUDA |
48, 64 or 128 GB (128 GB only with the 40-core GPU) spec [S4] | 75% of memory assumption | 614 GB/s spec [S4] | as Mac Studio M5 Max spec [S2] | 96 W (14-inch) / 140 W (16-inch) adapter spec [S4] adapter rating |
not verified for 128 GB configurations; check apple.com |
| Mac Studio, M5 Max 18-core CPU / 32-core GPU Metal, MLX; no CUDA |
36 GB; the 32-core GPU is sold only with 36 GB spec [S1] | 75% of memory assumption | 460 GB/s spec [S1] | not published spec [S2] | 7 W idle / 200 W max spec [S7] whole computer |
|
| Mac mini, M5 Pro Metal, MLX; no CUDA |
24, 48 or 64 GB spec [S5] | 75% of memory assumption | 307 GB/s spec [S5] | not published spec [S5] | 155 W max continuous spec [S5] whole computer |
NVIDIA GB10 machines (DGX Spark and partner versions)
| Machine | Memory | Usable for the model | Bandwidth | Published compute | Power | Prices |
|---|---|---|---|---|---|---|
| DGX Spark, 128 GB (NVIDIA and OEM GB10 versions) CUDA on Arm (aarch64 packages) |
128 GB LPDDR5x unified, 256-bit spec [S14] | 75% of memory assumption | 273 GB/s spec [S14] | 'Up to 1 PFLOP FP4' (sparse FP4 per The Register); no FP16/BF16 peak published. StorageReview measured 99.8 TFLOPS BF16 on a matmul test (2025-10-14) - an achieved rate, not a peak spec [S14] [S16] [S19] | GB10 TDP 140 W; 240 W external supply spec [S14] [S15] whole computer |
$6,950: NVIDIA's own unit (Founders Edition) (2026-10-02) cited [S17] not checked; checked 2026-10-20 |
| DGX Spark, 64 GB (partners only) CUDA on Arm (aarch64 packages) |
64 GB spec [S18] | 75% of memory assumption | 273 GB/s cited [S17] | same chip as the 128 GB unit spec [S18] | as the 128 GB unit spec [S15] whole computer |
AMD Ryzen AI Max mini PCs ("Strix Halo")
| Machine | Memory | Usable for the model | Bandwidth | Published compute | Power | Prices |
|---|---|---|---|---|---|---|
| Ryzen AI Max+ PRO 495 systems (AMD Ryzen AI Halo next-gen, Framework Desktop 192 GB) ROCm (see rule 4) |
up to 192 GB LPDDR5x-8533 spec [S24] | up to 160 GB assignable to the GPU (AMD footnote GRHP-01) spec [S25] | 273 GB/s spec [S24] | 61 TFLOPS: GPU FP16 'up to 61 TFLOPS'; NPU 55 TOPS spec [S24] | cTDP 45-120 W; Halo TDP 120 W spec [S24] chip only; add board, SSD, fans |
$6,799: Framework Desktop DIY, pre-order (2026-09-30) spec [S27] [S59] pre-order; Batch 1 sold out, Batch 2 ships November; checked 2026-10-20 $7,449: Framework Desktop pre-built, 2 TB SSD, Fedora (2026-09-30) cited [S52] [S29] pre-order, ships November $7,299.99: AMD Ryzen AI Halo, 192 GB, 2 TB SSD, Linux (Micro Center, in store since 2026-10-12) (2026-10-14) cited [S60] in store at Micro Center (reported) |
| Ryzen AI Max+ 395 systems (Framework Desktop, AMD Ryzen AI Halo, others) ROCm (see rule 4) |
up to 128 GB LPDDR5x-8000, 256-bit spec [S22] | up to 96 GB of 128 assignable to the GPU (Variable Graphics Memory) spec [S23] | 256 GB/s spec [S22] | 60 TFLOPS: GPU FP16 'up to 60 TFLOPS'; 126 TOPS total; NPU 50 TOPS spec [S22] | cTDP 45-120 W spec [S22] chip only; add board, SSD, fans |
RTX cards (add a host PC)
| Machine | Memory | Usable for the model | Bandwidth | Published compute | Power | Prices |
|---|---|---|---|---|---|---|
| GeForce RTX 5090 CUDA |
32 GB GDDR7, 512-bit spec [S32] [S33] | 90% of memory assumption | 1,792 GB/s spec [S32] [S33] | 209.5 TFLOPS: FP16/BF16 tensor, FP32 accumulate, dense (419 with FP16 accumulate; FP4 1,676 dense) spec [S32] | 575 W card; 1,000 W PSU recommended spec [S33] card only; add the host PC |
$5,581: card, eBay: average of the 3 lowest listings per day, Sept 2026, new or used not stated (2026-09) cited [S37] eBay listings |
| RTX PRO 6000 Blackwell CUDA |
96 GB GDDR7 ECC spec [S40] | 90% of memory assumption | 1,792 GB/s spec [S40] | 4,000 AI TOPS (FP4, sparse); no dense FP16/BF16 figure on the product page spec [S40] | 600 W spec [S40] card only |
$16,000: Workstation Edition, NVIDIA US Marketplace (2026-08-12) cited [S49] out of stock at NVIDIA (2026-08-12) |
| GeForce RTX 4090 (previous generation) CUDA |
24 GB GDDR6X, 384-bit spec [S32] [S34] | 90% of memory assumption | 1,008 GB/s spec [S32] | 165.2 TFLOPS: FP16/BF16 tensor, FP32 accumulate, dense; no FP4 spec [S32] | 450 W card; 850 W PSU minimum spec [S34] card only |
|
| RTX PRO 4000 Blackwell CUDA |
24 GB GDDR7 ECC spec [S39] | 90% of memory assumption | 672 GB/s spec [S39] | FP4 1,290 TOPS with sparsity; FP16 not on the product page spec [S39] | 145 W, single slot spec [S39] card only |
$3,134: lowest new US listing (tracker of Newegg / Best Buy / B&H) (2026-10-15) cited [S54] not checked |
Related, not in the tables: NVIDIA RTX Spark (N1X) laptops. Windows-on-Arm chip of the GB10 family, up to 128 GB LPDDR5X on a 256-bit bus at about 300 GB/s per NVIDIA's porting guide [spec, S20]. Pre-orders opened 2026-10-07, shipping from about 2026-10-16; launch prices from $2,599 (Surface Laptop Ultra; $5,899.99 with 128 GB / 1 TB) to about $7,000 (ASUS ProArt P16, top configuration) [cited, S61]. No review yet confirms units have arrived; not in the tables until one does. (checked 2026-10-20)
Card prices do not include the host PC. "Usable for the model" is the memory the GPU can use for weights and cache; how we get it is in rule 2.
Sources
- [S1] Apple, Mac Studio tech specs
- [S2] Apple newsroom, new Mac Studio with M5 Max and M5 Ultra (2026-08-25)
- [S3] Apple, Mac Studio buy page
- [S4] Apple, MacBook Pro tech specs
- [S5] Apple, Mac mini tech specs
- [S7] Apple Support, Mac Studio power consumption and thermal output
- [S9] Daring Fireball, M5 Max Mac Studio base prices (2026-08-25)
- [S10] EveryMac, Mac Studio M5 Max 18-core CPU / 40-core GPU
- [S12] MacRumors, 512 GB M5 Ultra in October (2026-08-25)
- [S13] Macworld, M6 and M5 Pro Mac mini prices (2026-08-25)
- [S14] NVIDIA, DGX Spark product page
- [S15] NVIDIA docs, DGX Spark hardware overview
- [S16] The Register, GB10 workstations arrive (2025-10-13)
- [S17] The Register, $4,999 64 GB DGX Spark; 128 GB to $6,950 (2026-10-02)
- [S18] NVIDIA blog, DGX Spark 64 GB (2026-10-02)
- [S19] StorageReview, DGX Spark review (2025-10-14)
- [S21] Tom's Hardware, DGX Spark $3,999 to $4,699 (Feb 2026; headline read)
- [S22] AMD, Ryzen AI Max+ 395 specifications
- [S23] AMD blog, Variable Graphics Memory FAQ (2025-07-29)
- [S24] AMD, Ryzen AI Halo with Ryzen AI Max+ PRO 495
- [S25] AMD blog, Ryzen AI Halo and Ryzen AI Max PRO 400 series (2026-05-20)
- [S26] AMD ROCm on Radeon and Ryzen, Linux support matrices (7.2.1)
- [S27] Framework, Desktop DIY Edition (prices, stock)
- [S28] Framework, Desktop overview
- [S29] Framework blog, 192 GB Framework Desktop pre-order (2026-09-30)
- [S31] Tom's Hardware, $3,999 Ryzen AI Halo (headline read)
- [S32] NVIDIA, RTX Blackwell GPU architecture whitepaper (Table 3)
- [S33] NVIDIA, GeForce RTX 5090 product page
- [S34] NVIDIA, GeForce RTX 4090 product page
- [S35] NVIDIA, GeForce compare page
- [S36] NVIDIA newsroom, RTX 50 series prices (2025-01-06)
- [S37] GPU Poet, RTX 5090 prices September 2026 (eBay listings)
- [S38] Wikipedia, GeForce RTX 40 series
- [S39] NVIDIA, RTX PRO 4000 Blackwell
- [S40] NVIDIA, RTX PRO 6000 Blackwell
- [S42] Llama 3 paper, Table 3
- [S43] Qwen3-30B-A3B model card
- [S44] Qwen3-30B-A3B config.json
- [S45] Qwen3-235B-A22B model card
- [S46] MLX docs, mlx.core.quantize
- [S47] mlx-lm README
- [S48] Rajbhandari et al., ZeRO
- [S49] Let's Data Science, RTX PRO 6000 price (2026-08-12)
- [S50] PCMag via Yahoo, $3,999 DGX Spark (2025-10-13)
- [S51] NVIDIA Developer Forums, 2/23/2026 price change (search result)
- [S52] Phoronix, Framework Desktop PRO 495 from $6,799 (2026-09-30)
- [S53] Let's Data Science, Strix Halo mini PC for $3,999 (2026-07-06)
- [S54] gpuprix, RTX PRO 4000 Blackwell US prices
- [S55] SHI, PNY RTX PRO 4000 Blackwell (search result)
- [S56] Intel, Thunderbolt 5 technology brief
- [S57] EveryMac, Mac Studio M5 Ultra 30-core CPU / 64-core GPU
- [S58] Daring Fireball, configurations and pricing for new Mac minis and Mac Studios (2026-08)
- [S59] Framework, Desktop DIY (Max+ PRO 495) configurator: SSD extra or bring your own; Batch 2 ships November (read 2026-10-20)
- [S60] Borncity, Ryzen AI Halo 192 GB at Micro Center for $7,299 (2026-10-14)
- [S61] PCWorld, the RTX Spark laptops available at launch (2026-10-08)
- [S20] NVIDIA docs, RTX Spark porting guide, system overview
- [S62] Apple, Mac Studio configurator (36/80-core M5 Ultra, 512 GB, 16 TB: $29,799; option steps shown; ships 10-12 weeks; read 2026-10-30)
Accessed 2026-10-15; volatile prices re-checked 2026-10-20.