Documents & knowledge
Find an answer in your materials and show where it came from.
Check: a correct answer with the relevant source passage.
READY SYSTEMS FOR LOCAL AI MODELS
Local AI.
For your work.
We specialise in local AI for documents, coding and teams. We help choose the model, hardware and a clear way to check the result.
Find a system for my workSee systems and pricesCore 72 personal workstation1,390,000–1,650,000 RUBPreliminary price for the complete system
Runs on your premises
Interface example
Your AILocal
Summarise the key points in this document
Here is a brief summary
Hardware visualisation
Data under your control
We’ll set up document processing on your hardware.
Easy from the first launch
Talk to AI in a familiar chat interface.
Suited to your work
We’ll select the model, hardware and software you need.
START WITH THE WORK YOU NEED DONE
One AI. Different kinds of work.
You do not have to guess which GPU to buy. Start with a useful result, then choose the model and system.
Code & development
Understand code, prepare a change and verify it with tests.
Check: the selected change passes the agreed tests.
An assistant for your team
One local service for employees, with the access rights you need.
Check: response time with concurrent requests.
COMPLETE SYSTEMS FROM CHINA
Servers for Qwen3.8-Flash-Next-FP8.
See the configuration and price upfront.
Model, assembly and setup in one package.
Selected model: Qwen3.8-Flash-Next-FP8 · FP8Show all systems
For DeepSeek and large modelsBuilt to order
Atlas 336
D1Four professional GPUs for development, agents and complex tasks.
Preliminary complete-system price5,710,000–6,400,000 RUBHardware, assembly and initial setup
- Graphics cards
- 4 × RTX PRO 6000D 84 GB
- Memory / storage
- 256 GB DDR5 ECC RDIMM / 2 + 4 TB
- DeepSeek V4.1 Flash profile
- One server with four GPUs
Full configuration and terms
- Processor / motherboard
- Threadripper PRO 9965WX · 24 cores / ASUS WRX90
- Graphics cards
- 4 × RTX PRO 6000D 84 GB
- System memory
- 256 GB DDR5 ECC RDIMM · 8 × 32 GB
- Drives
- 2 TB for system + 4 TB for models
- Power and cooling
- At least 3000 W usable power · server airflow or factory liquid cooling
- Model placement
- For DeepSeek: four GPUs in one group, with part of the model in RAM
We test speed and your workloads before handover. Shipping, import and complex integrations are charged separately.
For Qwen Flash and MiniMaxBuilt to order
Studio 288
N1A four-GPU system for one large model or several separate Qwen 27B replicas.
Preliminary complete-system price5,640,000–6,330,000 RUBHardware, assembly and initial setup
- Graphics cards
- 4 × RTX PRO 5000 Blackwell 72 GB
- Memory / storage
- 256 GB DDR5 ECC RDIMM / 2 + 4 TB
- Qwen Flash or MiniMax M2.7: one model
- Qwen 27B: four separate replicas
Full configuration and terms
- Processor / motherboard
- Threadripper PRO 9965WX · 24 cores / ASUS WRX90
- Graphics cards
- 4 × RTX PRO 5000 Blackwell 72 GB
- System memory
- 256 GB DDR5 ECC RDIMM · 8 × 32 GB
- Drives
- 2 TB for system + 4 TB for models
- Power and cooling
- 2050–2200 W · case for four GPUs
- Model placement
- Large model: one 288 GB group. Qwen 27B: four 72 GB replicas
We test speed and your workloads before handover. Shipping, import and complex integrations are charged separately.
Two 96 GB GPUsBuilt to order
Flow 192
N2For Qwen Flash, subject to memory-headroom checks, or two separate Qwen 27B replicas.
Preliminary complete-system price5,610,000–6,300,000 RUBHardware, assembly and initial setup
- Graphics cards
- 2 × RTX PRO 6000 Blackwell Workstation 96 GB
- Memory / storage
- 256 GB DDR5 ECC RDIMM / 2 + 4 TB
- Qwen Flash: one GPU group
- Qwen 27B: two independent replicas
Full configuration and terms
- Processor / motherboard
- Threadripper PRO 9965WX · 24 cores / ASUS WRX90
- Graphics cards
- 2 × RTX PRO 6000 Blackwell Workstation 96 GB
- System memory
- 256 GB DDR5 ECC RDIMM · 8 × 32 GB
- Drives
- 2 TB for system + 4 TB for models
- Power and cooling
- 2050–2200 W · case for two GPUs
- Model placement
- Qwen Flash: a 192 GB group with limited headroom. Qwen 27B: two 96 GB replicas
We test speed and your workloads before handover. Shipping, import and complex integrations are charged separately.
What is included in the price and how is it calculated?
Estimate dated 12 October 2026: the complete hardware system plus allowances for assembly, initial setup of the selected model and testing. Handover in China. Shipping, import, ongoing support and complex integrations are calculated separately.
This preliminary range uses public component and base-platform prices, budget allowances and a 10–15% contingency. Before ordering, the quotation will confirm final price, compatibility, availability, lead time and warranty. This range is not a system integrator’s quotation.
View calculations and sourcesNot sure what to choose?Answer two simple questions
A CHOICE YOU CAN VERIFY
Your task first. Confidence in the choice follows.
Before ordering, agree what the system must do and how to assess it. Evaluate speed, quality and workload on the selected model and configuration.
The test setup, access and testing cost are agreed separately.- 01
Choose example tasks
Your document, code excerpt or team workflow.
- 02
Agree the criteria
What counts as a correct answer and an acceptable response time.
- 03
Test the selected system
Record the model, hardware and test conditions together.
LOCAL AI IN PRACTICE
News that helps you choose.
International research, new models and practical implications for your system.
ModelDeepSeek
DeepSeek-V4.1-Flash: preparing for local deployment
DeepSeek introduced a model with a new architecture and visual understanding, alongside weights and a technical report. The announcement describes further work on inference support.
What this means for a buyer
Select hardware around an exact model version, a supported inference engine and your workload. Assess the local deployment separately from the developer’s hosted service.
Source publication date
Read the analysisModelNVIDIA Technical Blog
Nemotron Nano Omni: documents, images and audio
NVIDIA introduced an open model that processes text, images, audio and video together. Weights, evaluation materials and local deployment guides are available.
What this means for a buyer
Consider this model for scans, tables and mixed materials. Before selecting hardware, test whether it extracts the required information from documents in your formats and languages.
Source publication date
Read the analysisResearchSWE-bench authors
AI for coding: verify the change in a real project
SWE-bench evaluates fixes for real repository issues. Models receive a problem description and code; results are checked in the project environment, connecting evaluation to completed software work.
What this means for a buyer
Choose a local coding assistant by checking an actual change and your project’s tests. Results depend on the model, available tools and working environment.
Source publication date
Read the analysisSTART WITH YOUR TASK
Just two
simple questions.
Tell us what AI should help with
and who will use it.
We’ll select the rest.
Your taskReady-to-use solution
Prefer to talk?+7 903 110‑46‑38
QUESTION 1 OF 2
What will AI help you with?
You can select several tasks.
FROM SELECTION TO FIRST LAUNCH
Focus on your work.
We’ll take care of your system.
One clear process.
We agree the configuration, checks and supply terms.
01
Choose a system
Complete configurations and preliminary prices are already in the catalogue.
02
Receive a quotation
We’ll specify the hardware, price and terms.
03
Assembly and testing
We’ll prepare the system in China and test it on agreed examples.
04
Start using it
We’ll agree the handover and help with the first launch.
Checks before handoverThe acceptance programme will cover what the system can do and how it handles your tasks.
A READY RESULT, BEYOND THE COMPONENTS
What your proposal will cover
Agree each item and its cost before ordering. Additional integrations and ongoing support are described separately.
Hardware
Exact specifications, component condition and handover terms.
Model version
Model name, weight format and terms of use.
An easy start
Interface, required software and access settings.
Result verification
Tasks, conditions and a way to assess the system.
Handover & guidance
Startup instructions and the agreed level of assistance.
Warranty & support
Terms, a responsible contact and the service procedure.
What will I receive?
The agreed computer or server, an installed AI model and the required software. We help you get started. The quotation specifies the exact configuration, setup, price and support.
Do I need to understand models and graphics cards?
Just tell us what you want AI to help with. We’ll clarify your tasks, select a model and explain suitable options in plain language.
What is included in the price?
The catalogue shows a preliminary price for the complete system: hardware and allowances for assembly, initial setup and testing. Handover is in China. Shipping, import and complex integrations are calculated separately. Price, lead time and warranty are confirmed in the quotation before ordering.
How quickly will AI respond?
We test the selected system on agreed tasks, measuring time to first response and generation speed. For teams, we test simultaneous requests. Measurements from another system are not presented as the speed of your purchase.
Will my documents stay with me?
A local model can process documents on your own hardware. We agree deployment, staff access and connected services to meet your data requirements.
Can I choose DeepSeek or another model?
Yes. We consider DeepSeek, Qwen, GLM, MiniMax and other models. The exact version and tasks affect the hardware configuration. If you already know what you need, open the technical selector below.
Can I collect the system in China myself?
Self-collection can be arranged. The handover location, configuration, checks, documents and warranty are set out in the quotation. Shipping can also be discussed for your order.
I know the model and specificationsTechnical selection, hardware configurations and benchmark sources
FOR TECHNICAL USERS
Model, workload, hardware.
Preliminary sizing for the exact model version. Final configuration and speed are confirmed by testing.
Weight files: 172.78 GiB Details
Generations at the same time, not the number of staff or agents.
Weight format and context-memory format are separate settings.
Preliminary calculation for the exact checkpoint. Speed, actual memory use and the selected workload have not yet been measured. The native context limit is not confirmed here; the selected length needs testing.
Flow 192
2 × RTX PRO 6000 Blackwell Workstation 96 GB
5,610,000–6,300,000 RUBPreliminary complete-system price
One model replica. Context and the selected workload are subject to testing.
Group: 192 GBGPU weights: 172.78 GiB
Configuration and terms
- CPU
- Threadripper PRO, 24 cores / WRX90, two direct PCIe x16 slots
- RAM
- 256 GB ECC, 8 × 32 GB
- Storage
- NVMe TLC: 2 TB + 4 TB
- Power
- Factory 2050–2200 W power for two 600 W cards
Flash-Next leaves no more than 19.22 GiB nominally before buffers.
Studio 288
4 × RTX PRO 5000 Blackwell 72 GB
5,640,000–6,330,000 RUBPreliminary complete-system price
One model replica. Context and the selected workload are subject to testing.
Group: 288 GBGPU weights: 172.78 GiB
Configuration and terms
- CPU
- Threadripper PRO, 24–32 cores, WRX90, four CPU-connected PCIe slots
- RAM
- 256 GB ECC, 8 × 32 GB
- Storage
- NVMe TLC: 2 TB + 4 TB
- Power
- Factory 2050–2200 W power; four 300 W cards with verified clearance for active cooling
GLM Flash is excluded. Four Qwen replicas require four single-GPU groups; speed must not be multiplied automatically.
Atlas 336
4 × RTX PRO 6000D 84 GB
5,710,000–6,400,000 RUBPreliminary complete-system price
One model replica. Context and the selected workload are subject to testing.
Group: 336 GBGPU weights: 172.78 GiB
Configuration and terms
- CPU
- Threadripper PRO, 24–32 cores, e.g. 9965WX/9975WX; WRX90 with direct PCIe lanes. A factory AMD EPYC server with verified lane layout is also possible.
- RAM
- 256 GB ECC, 8 × 32 GB, 8 channels
- Storage
- TLC NVMe: 2 TB for system + 4 TB for models
- Power
- At least 3000 W usable PSU capacity; server airflow or factory liquid cooling for four cards
DeepSeek requires a specified Engram-to-RAM offload method; peak memory must be tested. GLM Flash leaves limited headroom after loading weights. Workload and placement on each GPU require testing.
MODEL DETAILSDeepSeek-V4.1-Flash · FP4/FP8Qwen3.8-27B · BF16Qwen3.8-27B-FP8 · FP8Qwen3.8-Flash-Next-FP8 · FP8MiniMax-M2.7 · FP8GLM-5.3-Flash · FP8MiniMax-M3-MXFP8 · MXFP8GLM-5.3 · FP8Kimi-K3 · MXFP4
All 8 hardware configurations
Atlas 336
4 × RTX PRO 6000D 84 GB
Full configuration
- Processor
- Threadripper PRO, 24–32 cores, e.g. 9965WX/9975WX; WRX90 with direct PCIe lanes. A factory AMD EPYC server with verified lane layout is also possible.
- RAM
- 256 GB ECC, 8 × 32 GB, 8 channels
- Storage
- TLC NVMe: 2 TB for system + 4 TB for models
- Power
- At least 3000 W usable PSU capacity; server airflow or factory liquid cooling for four cards
DeepSeek requires a specified Engram-to-RAM offload method; peak memory must be tested. GLM Flash leaves limited headroom after loading weights. Workload and placement on each GPU require testing.
Studio 288
4 × RTX PRO 5000 Blackwell 72 GB
Full configuration
- Processor
- Threadripper PRO, 24–32 cores, WRX90, four CPU-connected PCIe slots
- RAM
- 256 GB ECC, 8 × 32 GB
- Storage
- NVMe TLC: 2 TB + 4 TB
- Power
- Factory 2050–2200 W power; four 300 W cards with verified clearance for active cooling
GLM Flash is excluded. Four Qwen replicas require four single-GPU groups; speed must not be multiplied automatically.
Flow 192
2 × RTX PRO 6000 Blackwell Workstation 96 GB
Full configuration
- Processor
- Threadripper PRO, 24 cores / WRX90, two direct PCIe x16 slots
- RAM
- 256 GB ECC, 8 × 32 GB
- Storage
- NVMe TLC: 2 TB + 4 TB
- Power
- Factory 2050–2200 W power for two 600 W cards
Flash-Next leaves no more than 19.22 GiB nominally before buffers.
Core 72
1 × RTX PRO 5000 Blackwell 72 GB
Full configuration
- Processor
- Ryzen 9 9950X, motherboard with a full CPU-connected PCIe x16 slot
- RAM
- 64–128 GB DDR5, motherboard-compatible UDIMM kit
- Storage
- TLC NVMe: 1–2 TB for system + 4 TB for models
- Power
- Quality 1000 W PSU, ventilated case, actively cooled GPU
Qwen BF16 passes the preliminary weight-size check. BF16 speed and C4/C8 modes have not been measured.
Core Duo
2 × RTX PRO 5000 Blackwell 72 GB
Full configuration
- Processor
- Threadripper PRO, 24 cores, WRX90, two direct PCIe x16 slots
- RAM
- 128 GB ECC, 4 × 32 GB; expandable to 256 GB
- Storage
- NVMe TLC: 2 TB + 4 TB
- Power
- Quality 1600 W PSU, two actively cooled GPUs
144 GB is not used as a single pool for Flash-Next.
Core Twin
2 workstations, each with 1 × RTX PRO 5000 72 GB
Full configuration
- Processor
- Ryzen 9 9950X in each, separate OS / containers and code-build resources
- RAM
- 64–128 GB DDR5 UDIMM in each
- Storage
- 1–2 TB for system + 4 TB for models in each
- Power
- 1000 W PSU and verified active cooling in each
Adding the VRAM of two workstations does not permit splitting one model across them without a separate deployment recipe.
Radeon 64
2 × Radeon AI PRO R9700 32 GB
Full configuration
- Processor
- Ryzen 9 9950X; motherboard with two CPU-connected slots and confirmed x8/x8 mode
- RAM
- 128 GB DDR5 UDIMM
- Storage
- NVMe TLC: 1–2 TB + 4 TB
- Power
- 1200–1600 W PSU, cooling for two 300 W cards
Weight placement, ROCm, TP2, P2P and concurrent requests are untested.
Radeon 128
4 × Radeon AI PRO R9700 32 GB
Full configuration
- Processor
- Threadripper PRO, 24 cores, WRX90, four direct PCIe slots, ROCm-approved platform
- RAM
- 128 GB ECC, 4 × 32 GB
- Storage
- NVMe TLC: 2 TB + 4 TB
- Power
- Factory 2050–2200 W power, four 300 W GPUs
Simultaneous operation of both TP2 groups has not been tested.
External benchmarks and test conditions
These are the authors’ results on different hardware. They cannot be applied to catalogue systems.
DeepSeek V4.1 Flash
4 × RTX PRO 6000 Workstation 96 GB, Threadripper PRO 9985WX, approximately 639 GiB RAM, Engram in RAM, patched SGLang and DSpark.
- 1 request
- 365 tokens/s · median for one response
- 8 requests
- 1,194 tokens/s · aggregate
- 32 requests
- 3,081 tokens/s · aggregate
Qwen 3.8 27B · FP8
RTX PRO 5000 72 GB, Core Ultra 7 270K Plus, 270 W, vLLM 0.26, MTP 1, one request.
- Approximately 11K context
- 43.2 tokens/s
- 148–200K context
- 15.3 tokens/s
When usage is unavailable, the author counts SSE chunks, which limits measurement accuracy.
Test conditions and codeLET’S START WITH A CONVERSATION
Tell us about your work.
We’ll select your AI system.
A clear configuration. A complete system. Help getting started.