Sunk Cost sunkcost.ai Data checked 2026-09-03

Corsair AI Workstation 300, 64GB vs 128GB for local AI

The Corsair AI Workstation 300, 128GB holds 33 of the 39 open models here, and the Corsair AI Workstation 300, 64GB holds 27. The Corsair AI Workstation 300, 64GB costs $3,000 less. On Qwen3.8 27B, the strongest model both hold, they run at much the same speed: 10 and 10 tok/s, both estimated from memory bandwidth. The Corsair AI Workstation 300, 64GB pays for itself sooner, in 22 years against 62 years at 500k tokens a day.

Strix Halo Corsair AI Workstation 300, 64GBStrix Halo Corsair AI Workstation 300, 128GB
Price$1,700$4,700
Memory64 GB128 GB
Usable by the GPU48 GB96 GB
Memory bandwidth256 GB/s256 GB/s
Power under load133 Wstand-in133 Wstand-in
Models that fit2733
Best model it runsQwen3.8 27BQwen3.8 27B
Speed on that model10 tok/s estimated10 tok/s estimated
Pay-back on that modelPays back in 22 yearsPays back in 62 years

Neither power figure above is measured on the machine beside it. Both are stand-ins borrowed from the nearest hardware the data does have, so what the row shows is one borrowed figure printed twice. Each machine's page names the figure it borrows and why. Every pay-back figure on this page prices its electricity from these numbers.

Run the numbers on the Strix Halo Corsair AI Workstation 300, 64GB · or the Strix Halo Corsair AI Workstation 300, 128GB

The step up is a different chip, not just more memory

Corsair AI Workstation 300 is one name for two machines. The data lists the 64GB as Ryzen AI Max 385 · Radeon 8050S, 32 CU and the 128GB as Ryzen AI Max+ 395 · Radeon 8060S, 40 CU. The step adds 8 compute units to the graphics part, 40 against 32. That is why this pair is not one of the site's memory-size comparisons: those hold the silicon equal so the memory is the whole of the difference, and Corsair cut both to reach the $1,700.

The bigger graphics part is not what sets the speeds on this page. Writing a token means reading the whole model out of memory, and both chips read it at 256 GB/s, so the figures follow the memory rather than the chip. That is why the table gives them the same speed. What the extra compute units do is read a long prompt before the first token comes back, and that is not something this site measures or prices. The 128GB box holds 33 of the 39 models the calculator counts against 27 on the 64GB. Price the Corsair AI Workstation 300, 64GB on its own before paying for the step: if the models you want fit the cheaper box, the chip above it is not buying you tokens.

How much use it takes to pay back

Everything above is at 500k tokens a day. Pay-back moves with how much you actually run, so here are both machines on Qwen3.8 27B, the strongest model both hold, at the five levels of use the calculator names. The Strix Halo Corsair AI Workstation 300, 64GB pays back sooner at every level of use, so this is not a choice that turns on how hard you work it.

A day's useStrix Halo Corsair AI Workstation 300, 64GBStrix Halo Corsair AI Workstation 300, 128GB
50ka few chats a day223 years615 years
200klight assistant use56 years154 years
1Ma moderate coding-assistant day11 years31 years
4Mheavy coding with an agent2.8 years7.7 years
20Magents running most of the day9.4 monthsits ceiling2.2 yearsits ceiling

On Qwen3.8 27B neither machine can generate 20M tokens a day: the Strix Halo Corsair AI Workstation 300, 64GB manages at most 14.3M and the Strix Halo Corsair AI Workstation 300, 128GB at most 14.3M. Both figures on that row are for the most each can do.

Run the Strix Halo Corsair AI Workstation 300, 64GB at 20M tokens a day · or the Strix Halo Corsair AI Workstation 300, 128GB

What the extra memory buys

The Strix Halo Corsair AI Workstation 300, 128GB holds 6 models the Strix Halo Corsair AI Workstation 300, 64GB cannot at 32k of context. The strongest of them are what the difference in memory actually buys.

ModelWeightsNeeds at 32kOn the Strix Halo Corsair AI Workstation 300, 128GB
Ling 3.0 flashHaiku-class78 GB82 GB20 tok/s estimated
Qwen3.5 122B-A10BHaiku-class78 GB79 GB20 tok/s estimated
gpt-oss-120bBelow every hosted tier63 GB65 GB39 tok/s measured
Mistral Small 4 (119B-2603)Below every hosted tier74 GB75 GB32 tok/s estimated
Qwen3-Coder NextBelow every hosted tier48 GB49 GB54 tok/s estimated
Devstral 2 123BBelow every hosted tier75 GB87 GB2.2 tok/s estimated

The extra memory buys context as well. Of the 27 models both machines hold at 32k, 4 run to a longer window on the Strix Halo Corsair AI Workstation 300, 128GB: the weights are a fixed size and the KV cache is not, so what the weights leave spare is what a longer context grows into. Gemma 4 31B it reaches 256k there against 128k on the Strix Halo Corsair AI Workstation 300, 64GB, each the longest window the calculator offers that the machine still holds it at.

The assumptions behind both columns

Both columns use the same usage: 500k tokens a day at 15:1 input to output, 32k of context, $0.17 per kWh, and today's API prices held flat. Speeds marked estimated are worked out from memory bandwidth rather than measured, and pay-back scales with them. Change any of it in the calculator.

More head to head: everything the Strix Halo Corsair AI Workstation 300, 64GB runs · everything the Strix Halo Corsair AI Workstation 300, 128GB runs · every other match-up · the quickest pay-back at each level of use · every model against the frontier