You know how to build a computer. You've been building computers for years.
What you might not know is that MatLab isn't fully supported on the GPU you've chosen. Or that you don't have enough bandwidth to support the number of GPUs you want to run.
That's where we come in.
Tailor made for your project
Our specialists can create a custom specification that matches your exact requirements and use cases.
Discounts, grants and rebates
We can help you find savings and discounts through government schemes or direct from our suppliers that you may not be aware of.
Ongoing support
We offer warranties and technical support on all our products so you can be confident that your build will last and that any issues will be quickly dealt with.
Have total confidence in your build
We can cross-reference every component with your use cases to ensure total hardware and software compatibility.
Benefit from our relationships
We can use the roadmaps provided by vendors to flag up new components that are about to hit the market that may have an impact on your price or performance.
Ready to go
All Novatech Deep Learning systems can be pre-installed with Ubuntu 16.04 server LTS and your choice of frameworks, so you can get to work right out of the box.
What are you working on?
Prostation DL9-2R
The ProStation DL9-2R is optimised for deep-learning development, where models can be created and optimised at an affordable workstation level.
Configure & BuyHyperStation DLX-4R
The HyperStation DLX-4R allows developers and researchers to experiment and train on models of a more complicated nature, using larger data sets.
Configure & BuyHyperServe RME2
A high density 2U server, designed to train large models, requiring massive GPU compute power.
Configure & BuyCan't see what you need? Got a build you want to discuss?
View our entire range, or contact our specialists today.
Make the best hardware decisions for your research budget
We're not Data Scientists, but we've worked with enough to know exactly what makes for a zero compromise Workstation.
Of course, all Workstations are designed relative to the level of performance required for any given task or workload, and rarely are two ever identical - such is the nature of efficiency and purpose.
So to help simplify the process of choosing your configuration, we've hand-picked some of the most popular components that make their way through our production line.
Popular Graphics Cards
NVIDIA GeForce RTX 2080Ti
- Memory 11GB GDDR6
- CUDA Cores 4,352
- Tensor Cores 544
- RT Cores 68
- Single-Precision Performance 13.4 TFLOPS
- Power Consumption 250 W
The RTX 2080Ti is possibly the most popular choice for development workstations due to the price/performance ratio; the blower-style design also allows for up to 4x GPUs within a single desktop solution (although they can be noisy under load). The only drawback is the relatively small memory size, which will require the use of smaller batch sizes and limit model size.
View NVIDIA RTX WorkstationsNVIDIA TITAN RTX
- Memory 24GB GDDR6
- CUDA Cores 4,608
- Tensor Cores 576
- RT Cores 72
- Single-Precision Performance 16.3 TFLOPS
- Power Consumption 280 W
The Titan RTX offers amazing performance in terms of pure horsepower and in some instances is a 'quicker' card than the Quadro options. NVLink also gives you the ability to link two cards, effectively giving you 48GB of memory. The main drawback is the cooling, as there is no blower design available, so you're limited to only a pair of Titan's per system. Any more and they will massively thermal throttle, and simply stop working.
View NVIDIA RTX WorkstationsNVIDIA Quadro RTX 6000
- Memory 24GB GDDR6 ECC
- CUDA Cores 4,608
- Tensor Cores 576
- RT Cores 72
- Single-Precision Performance 16.3 TFLOPS
- Power Consumption 295 W
Although much more expensive than the 2080Ti, the RTX 6000 shares a similar blower design, allowing for densely populated systems. They also have the same memory capacity as the TITAN RTX (24GB per GPU, 48GB with NVLink) enabling you to to work with large batch sizes and train large models. Although the benefits may be slight, the Quadro RTX 6000 and 8000 also have ECC memory, offering additional stability.
View NVIDIA Quadro WorkstationsNVIDIA Quadro RTX 8000
- Memory 48GB GDDR6 ECC
- CUDA Cores 4,608
- Tensor Cores 576
- RT Cores 72
- Single-Precision Performance 16.3 TFLOPS
- Power Consumption 295 W
The RTX 8000 has the largest memory capacity of any NVIDIA GPU, with 48GB per card, connect a pair with NVLink and you'll have a massive 96GB of GDDR6 memory. The blower design also allows for densely populated systems (if you can afford it!). This is the ultimate card for any task that requires extremely large models. The only downside is the price tag.
View NVIDIA Quadro WorkstationsPopular Processors
Intel Core i9-10900X
- Base Clock (at TDP) 3.70 GHz
- Cores/Threads 10/20
- PCIe lane support 48 @ PCIe 3.0
The i9-10900X is one of the fastest CPU's in the HEDT space. It may still be running on Intel's 'mature' 14nm process, but to compete with AMD, Intel are really pushing these chips to their limit, giving you very impressive clock speeds (these usually run at well above 'base clock' with sufficient cooling). Pair this with a motherboard that supports PLX switching, and you can happily run 4x GPUs at PCIe x16. It's a solid all-rounder for a desktop workstation.
View Intel Core WorkstationsAMD Ryzen Threadripper 3960X
- Base Clock (at TDP) 3.80 GHz
- Cores/Threads 24/48
- PCIe lane support 88 @ PCIe 4.0
Although the 3960X runs a little slower than the i9, the 24 cores more than makes up for any single core speed loss, making it an excellent choice for heavily multi-threaded processes such as data preprocessing and some visualisation workloads. The 88 PCIe 4.0 lanes (double the bandwidth of PCIe 3.0) provide vast I/O throughput, allowing for multiple GPUs and NVME hard drives, without the bottlenecks.
View AMD Threadripper WorkstationsAMD EPYC 7452
- Base Clock (at TDP) 2.35 GHz
- Cores/Threads 32/64
- PCIe lane support 128 @ PCIe 4.0
With 32 cores, 128 PCIe lanes and 8-channel memory support, the AMD EPYC platform offers superior memory, and I/O throughput allowing for flexibility in extending storage, networking or compute to a level previously unattainable, and only seen on dual-processor platforms.
View AMD EPYC WorkstationsIntel Xeon Gold 5120
- Base Clock (at TDP) 2.2 GHz
- Cores/Threads 14/28
- PCIe lane support 48 @ PCIe 3.0
Although on the surface the Xeon may be a little underpowered in comparison to the AMD EPYC, the Xeon has propriatary technologies specifically designed for ML/AI workloads. For example, Intel's DL Boost significantly increases performance within deep learning inference use cases such as image recognition, object detection, speech recognition, language translation, and others. And of course, there's guaranteed support for the Intel MKL to take into consideration.
View Intel Xeon Workstations