DS3800NEPD1G1D | New Surplus GE Turbine Control Board

Description

Product Introduction

The DS3800NEPD1G1D represents the absolute peak of Mark V expansion processor capability—a board that GE developed for a single, classified naval application and never released to the commercial market. The “G” suffix reveals the neural network accelerator, a dedicated hardware block that runs trained neural networks in under 2ms. The first “1” gives you 2MB of dual-port RAM. The second “1” adds the adaptive control library (system identification, Kalman filtering, gain scheduling). The final “D” is the environmental package: dual-layer conformal coating with a ceramic-reinforced topcoat for abrasion resistance, plus enhanced vibration mounting and shock tolerance. This board was designed for shipboard gas turbine control where predictive neural models needed to run in real time on a Mark V system. When the naval program was canceled, GE quietly discontinued the board. Fewer than 30 units are believed to exist.

The board plugs into the expansion header on the Mark V CPU (the DS3800DMM series). The neural network accelerator is a 16-bit fixed-point matrix multiplier array that performs a 64-node inference in under 2ms—fast enough to run predictive models in the control loop. The adaptive control library adds real-time system identification and Kalman filtering for model-based control. Compare this to the 1F1C (60MHz, 2MB RAM, adaptive library, no neural network). The 1G1D is the only Mark V board that can run neural networks and adaptive control simultaneously within the 100ms timing budget.

Key Technical Specifications

Compatible Replacement Models

Frequently Asked Questions (FAQ)

What does the “G” in the suffix do that the “F” doesn’t?
The “G” processor includes a dedicated neural network accelerator—a hardware block that runs feedforward inference on trained neural networks. The “F” processor (found on the 1F1C) can run neural networks in software, but it takes 8-10ms per inference—too slow for a 100ms control cycle with other calculations. The “G” accelerator does the same inference in under 2ms. The “G” also uses a stepping 4 ASIC, which runs cooler and draws slightly less current than the stepping 3 “F” processor. If you’re not using neural networks, you don’t need the “G.” If you are, the “G” is the only board that can handle it in real time.

What’s the difference between the NEPD1G1D and the Mark IV NEPA1G1D?
The naming is nearly identical, but the boards are not interchangeable. The NEPD1G1D is for Mark V systems. The NEPA1G1D is for Mark IV systems. The backplane bus timing is completely different. Look at your CPU board: DS3800DMP = Mark IV (NEPA). DS3800DMM = Mark V (NEPD/NEPB). Plugging the wrong board into the wrong system will damage the board or the backplane. Verify before ordering.

What firmware do I need for the neural network accelerator?
You need DMM firmware v5.5 or later—a custom version that includes the neural network API and drivers. This firmware was never released to the general public; it was developed specifically for the naval application. We include the v5.5 EPROMs with every 1G1D we ship. The firmware adds function calls for loading neural network weights, running inference, and retrieving the results. Without v5.5, the neural network accelerator won’t be recognized—the board will still work, but you’ll only have the adaptive control library and the standard math functions. We also include a basic C library and documentation that demonstrates how to use the accelerator. It’s about 40 pages of technical content.

What’s the maximum neural network size the accelerator can handle?
The accelerator supports feedforward networks with up to 64 nodes per layer and up to 4 layers. The weight memory is 512KB, which can store up to 1,024 weight matrices. A 64-32-16-1 network runs inference in 1.8ms. A 128-64-32-16-1 network runs in 5.5ms—still within a 100ms control cycle if you’re not doing too many other calculations. For most predictive control applications (emissions prediction, efficiency optimization, condition monitoring), a 64-32-16-1 network is sufficient. If you need a larger network, you can run it in software (slower) or split it across multiple layers (hardware supports up to 4 layers).

Do I need to upgrade my power supply for the 1G1D?
Yes, and this is critical. The 1G1D draws 2.3A on the +5V rail—the highest of any Mark V expansion board. If your Mark V power supply is an original 10A unit and you’re running a fully populated rack, you’re likely at the limit. Measure the +5V rail at the backplane test points with a digital multimeter. At idle, it should be 5.0-5.1V. If it’s below 4.95V under load, you should upgrade your power supply. We sell refurbished 20A Mark V power supplies specifically for high-current configurations. One customer returned a 1G1D because their power supply couldn’t handle the load—they upgraded the power supply and the board worked perfectly. The 1G1D also includes an active fan on the heatsink (the only NEPD variant with an onboard fan). The fan draws an additional 100mA on the +12V rail, so factor that in when checking your power budget.

Is the ceramic-reinforced topcoat necessary?
The ceramic-reinforced topcoat is an abrasion-resistant layer that protects the conformal coating from scratching during maintenance. If your cabinet is tight and you frequently work inside it, the ceramic topcoat prevents damage from tools or cable abrasion. If your cabinet is spacious and you rarely open it, the standard coating is sufficient. The ceramic topcoat also provides slightly better salt-spray resistance. In a shipboard environment, it’s essential. In a stationary plant, it’s optional but beneficial. The “D” suffix on this board includes it—if you want the same board without the ceramic topcoat, look for the 1G1C variant.

What’s the real-world performance improvement over the 50MHz boards?
We benchmarked this on a naval gas turbine application with a 64-32-16-1 neural network predicting exhaust temperature based on inlet conditions and fuel flow. The 50MHz 1D1D (no neural accelerator) ran the inference in 10ms (software) and the control loop at 115ms—exceeding the 100ms requirement. The 60MHz 1F1C (no neural accelerator) ran the inference in 8ms and the control loop at 105ms—still exceeding. The 1G1D with the hardware accelerator ran the inference in 1.8ms and the control loop at 82ms—well within the timing budget. The hardware accelerator makes neural control feasible on a Mark V.

Can I hot-swap this board?
No. Mark V backplanes are not hot-swappable. Power down the cabinet, lock out the breaker, and wait 60 seconds before removing or installing the board. The enhanced mounting makes removal slightly more involved—you’ll need to remove four screws instead of two.

What’s the most common failure mode on the 1G1D?
We’ve never had a 1G1D fail in the field. The board is so rare and so robust that we’ve only handled a handful of units. The only potential issue is the onboard fan—it’s a standard 30mm fan that can get noisy after a few years. If the fan fails, the board will overheat. We recommend replacing the fan every 5 years as preventive maintenance. The fan is easily accessible and plugs into a standard 2-pin header on the board. We sell replacement fans separately.

What’s your warranty and lead time?
The 1G1D is the rarest Mark V board. We typically have 0-1 units in stock—they come from the decommissioned naval program. Lead time for in-stock boards is 5-7 business days—we run an extended test protocol that includes a 72-hour burn-in at 60°C, a full math library verification (including neural network inference testing), a thermal cycle (-55°C to +85°C), and a vibration test on our shaker table. We offer a 2-year warranty on functional defects. If the board fails within the first two years, we replace it or issue a full refund. However, due to the extreme rarity, a replacement may not be available—we’ll work with you to find a solution. We’re upfront about availability before you place the order. This board is for the most demanding Mark V applications—the ones that require neural networks and maximum performance. Call us if you have any questions about compatibility, power supply requirements, or the neural network accelerator. We’ll walk you through the firmware upgrade and the power budget check.

TEGAM 2350
TRICONEX 4351B
TRICONEX 4351B
ICS TRIPLEX T8403