← Studio Toriumi

Local LLM · RTX PRO 6000 Blackwell · 192GB VRAM · 2026-09

The best local-LLM rig
an individual can buy, under ¥6M.

Go above an RTX 5090 and the field splits into professional and data-center territory. With a budget of ¥6 million (about $41,000), the answer converges on one build: two NVIDIA RTX PRO 6000 Blackwell cards (96GB each) — 192GB of VRAM total, enough to run Qwen3-235B-class MoE models and Llama 405B-class dense models on the desk in your home. This page is the blueprint: parts list, reasoning, and the electrical realities.

For cheaper tiers (dual 5090, dual used 3090), see the used RTX 3090 PC finder. This page covers the ceiling.

Prices surveyed September 2026; they move with FX and AI demand / This page is design reference, not an endorsement of any particular store or product

The verdict (September 2026 prices)

An RTX PRO 6000 Blackwell 96GB runs about ¥2.3–2.5M ($16,000-ish in the US). Two of them is roughly ¥4.8M; a Threadripper platform fills out the rest for a grand total around ¥5.6–5.9M. That buys 192GB of VRAM and 3.6TB/s of combined memory bandwidth — 3× the VRAM of a dual-5090 build, and unlike data-center cards (H100 etc.) it fits in a normal PC case and runs on ordinary GeForce-style drivers. This is the practical ceiling of what an individual can buy.

01The parts list

Estimated from Japanese retail (authorized and parallel imports), including tax and shipping.

PartPickApprox. price
GPUNVIDIA RTX PRO 6000 Blackwell 96GB ×2
(PNY/Leadtek etc. — dual-slot workstation cards, 400W each, one 12V-2x6 cable per card)
¥2.3–2.5M ×2 = ¥4.6–5.0M
CPUAMD Ryzen Threadripper 7970X (32 cores)
128 PCIe 5.0 lanes — gives both GPUs x16/x16
~¥0.5M
MotherboardTRX50 chipset (e.g. ASUS Pro WS TRX50-SAGE)
The SAGE-style 7-slot boards leave a gap between cards
~¥0.12–0.15M
RAMDDR5 RDIMM 128GB (4×32GB)
Spillover target when offloading MoE experts to CPU
~¥0.12–0.15M
PSU1600–1700W ATX 3.1 with 12V-2x6 ports
Two GPUs plus CPU can peak past 1.2kW; 1600W is the floor for sustained load
~¥0.07–0.09M
Storage4TB NVMe SSD (system + models)
A 235B-class GGUF is 130GB+ on its own
~¥0.06–0.08M
Case & coolingSilent full tower (e.g. Fractal Define 7 XL) + 360mm AIO
Airflow-first case selection
~¥0.05–0.07M
Total—~¥5.6–5.9M

The two GPUs are 80% of the body price. Since launch in 2025 (about ¥1.3M/card) they have roughly 1.8×'d in price, so ¥6M is a tight fit; if this tier cools off, the same build lands around ¥4.5M.

02Why the RTX PRO 6000

Above the 5090, your options split into pro-workstation and data-center cards.

CandidateVRAM (each)Verdict for this build
RTX PRO 6000 Blackwell96GBThe pick. GeForce-style drivers, dual-slot 400W cards that fit a normal case, ECC GDDR7 at 1.79TB/s
RTX 6000 Ada / A6000 (used)48GBRunner-up. 96GB for two — but no room for 235B-class
H100 80GB (used)80GBWrong tool. Over ¥4M for two, plus blower noise, exotic power, and awkward form factors
A100 80GB (used)80GBWrong tool. Cheaper per card but everything around it costs you
RTX 5090 ×232GBAlternate. 64GB for well under ¥1M total — one model tier lower

Data-center cards may win on VRAM-per-yen, but they lose on cooling, power, drivers, and installation — all four, for an individual. The PRO 6000 is a workstation card, built on the assumption that it works on a desk, and all four of those are as tame as a GeForce. That's the whole argument.

03What runs on 192GB

Model (examples)Approx. sizeOn this rig
Qwen3-235B-A22B-class (MoE) Q4_K_M~130–140GBFits. Only 22B of experts are active per token, so it's fast for its size — the flagship use of this build
Llama-3.1-405B-class Q3~170GBJust fits. Q4 (~230GB) does not
70B-class dense Q8 / fp16~75GB / ~141GBComfortably. Uncensored (abliterated) 70Bs at high precision
DeepSeek-R1/V3-class (671B MoE) Q2_K200GB+Overflows. Ultra-low-bit quants squeeze in, at a quality cost
27–32B-class (mainstream uncensored)~16–20GBTrivially. Even 100k+ token contexts leave headroom

Token generation speed is bounded by total memory bandwidth ÷ bytes read per token. Two cards give ~3.6TB/s. Qwen3-235B-A22B reads ~13GB per token at Q4 (22B active experts), so the theoretical ceiling is ~270 tokens/s — but only with tensor parallelism (vLLM and similar) keeping both cards busy at once. With llama.cpp's default layer split, the layers run on one card after the other, so a single conversation is capped by one card's 1.79TB/s, about 135 tokens/s. Even 20–40% below those ceilings in practice, it outpaces reading. These are spec-sheet estimates — no hands-on benchmark was done for this page.

04Power, placement, and pitfalls

05The ladder down

BudgetCore of the buildVRAMModel ceiling
~¥5.8MThis page: PRO 6000 96GB ×2192GBQwen3-235B Q4 / 405B Q3
~¥3MPRO 6000 96GB ×196GB70B fp16 / 235B Q2-class
~¥0.9MRTX 5090 32GB ×264GB70B Q4 / ~120B MoE
~¥0.5MFour routes compared: V100 32GB / modded 3080 20GB ×2 / used RTX 3090 ×2 / EVO-X232–48GBFast 27B or 70B Q4 (3090 hunting guide)
~¥0.24MUsed 3090 desktop PC24GB27–32B Q4 (pc3090)

At any tier: compute the Q4_K_M size of the model you actually want plus context plus 20% headroom, and buy one tier above that. That's the whole strategy.

06The best rig under ¥500k: four routes, four use cases

With used 3090s holding at ~¥200k each, dual 3090s (48GB) is now just one option — the budget-maxed generalist. Priced September 2026; speeds are a mix of spec-sheet math and published measurements.

RouteApprox. costMemory27B Q4 speed70B Q4Video genIn one line
① V100 32GB (SXM2 + blower shroud)¥150–350k32GB HBM2
900GB/s
~70 tok/sNo (40GB)NoFastest yen-per-token for chat. Conditions: CUDA 12.x only (CUDA 13 dropped Volta), no surrounding ecosystem. SXM2 needs a converter board + fan shroud; the PCIe version costs twice as much
② Modded 3080 20GB ×2¥250–280k40GB GDDR6X
760GB/s ×2
ComfortableJust fitsTight (20GB×2)Often rebuilt with 4090-class coolers, so condition tends to beat tired mining-era 3090s. But domestic supply is a trickle (a few listings/month), no warranty, 320W each
③ Used RTX 3090 ×2¥500–520k48GB GDDR6X
936GB/s ×2
ComfortableFitsYes (+64GB RAM advised)The generalist: LLM and video. Biggest risk is worn mining-era boards (condition checklist at pc3090)
④ EVO-X2 RAM64 (Ryzen AI Max+ 395)~¥350k64GB unified
256GB/s
10–15 tok/s~5 tok/sWeakA big-MoE appliance: gpt-oss-120b at usable speeds (32 tok/s on the 128GB version, 235B at 11). You give up CUDA (ROCm/Vulkan). Nothing else is this small, quiet, and frugal

How to choose: fastest uncensored-27B chat → ① V100. LLM and video generation (Wan-class) → ③ 3090×2, no real alternative. Big MoE models idling under RAG/agent workloads → ④ EVO-X2. ② is the fallback when you can't win a V100 — spend the change on storage and RAM.

If you buy a V100

SXM2 cards are passively cooled: a blower-fan shroud conversion is mandatory (kits and 3D-printed shrouds are widely available). Run llama.cpp on CUDA 12.x builds; don't count on Flash-Attention-class optimizations (Volta has fp16 tensor cores only). Power is a plain 250W per card on standard PCIe 8-pin.

Update (2026-09-23): sizing for Qwen4-27B

At the Apsara Conference (Sep 22–24, 2026) Alibaba announced the Qwen4 lineup (Max / Plus / Flash / 27B). The 27B isn't released yet and specs are unannounced, but recent 27B-class models (Qwen3.6-27B: ~19GB at Q4, ~30GB at Q8) make the preparation obvious. Abliterated builds lose a little capability to the pruning, so you want to run them at Q5/Q6 or higher — meaning "Q4 fits" is not the bar. Q6 plus long context is: 32GB (V100 / 5090) is the comfortable Q4 line, and 40–48GB (modded 3080 20GB ×2 / 3090×2) is the right size for high-quant + long context. One more reason to buy one rung up the ladder if 27B-class is your main battlefield. And if Qwen4-27B turns out to be MoE (the Flash-Next architecture is the rumored preview), total size could jump — that lands in ④EVO-X2 (128GB) or flagship-rig territory.

07Sources & method

  1. GPU pricing — NVIDIA's official marketplace and US retail tracking (September 2026, ~$16,000; per Thunder Compute and Price History aggregates). Converted at ¥145/USD.
  2. Specifications (VRAM, bandwidth, TDP, slot width) — NVIDIA's official spec sheets (RTX PRO 6000 Blackwell: 96GB GDDR7, 1.79TB/s, 400W, dual-slot).
  3. Model sizes — GGUF listings on Hugging Face; Q4_K_M estimated at ~4.8 bits/parameter. MoE speed estimates assume only active experts are read per token.
  4. Component prices — Japanese retail listings, September 2026. Subject to FX and stock.
  5. ¥500k comparison — V100 32GB from eBay/Yahoo! Auctions used prices (PCIe ~$800, SXM2+converter cheaper) with speed estimated from the 900GB/s HBM2 spec; the ~70 tok/s figure for 27B Q4 follows reports from Japanese operators. Modded 3080 20GB from Yahoo! Auctions sold prices (~¥65k average over 120 days) and Chinese-market reviews. EVO-X2 (Ryzen AI Max+ 395): ~¥350k for 64GB, with ServeTheHome and Impress PC Watch measurements (dense 70B ~5 tok/s; 128GB version: gpt-oss-120b 32 tok/s, Qwen3-235B 11 tok/s).