开云体育

The other argument against buying the absolute high end CPU (more CPU
than necessary for the purpose) is that Moore's Law still holds when
it comes to CPU performance vs. price.? By the time the user's needs
even come close to the capability of the CPU the price of that CPU
will drop significantly or newer CPUs will be available with major
advances in capability.

A typical lifetime of computer technology is something on the order
of six years - or perhaps two major versions of Windows.? Purchasing
capability beyond that lifetime is a waste of time.? At every 6-8
year point I have replaced computers with new systems having four
times the performance (MOps/Sec, RAM, etc.) the previous system at
half the cost of the "obsolete" system.

Today, with systems featuring CPUs at the median ~24,000 CPU Mark
level available for between $350 and $500, it makes no sense to
pay three to four times as much for the highest performance CPUs
that show CPU Marks of ~28,000-29,000.

73,

? ? ... Joe, W4TV

On 2025-03-26 5:57 PM, Dave AA6YQ wrote:
> + AA6YQ comments below
>
>>
>> I've followed this thread with some interest. It seems to me that an
>> important feature of computer, monitor, and graphics cards purchases is
>> missing.
>>
>> I believe one should attempt purchases that reflect more what demands on
>> these devices are likely to be in the future than what they might be in
>> the present. While the future is unknown, it seems very likely that
>> demands for more memory, more sophisticated and faster processors will be
>> the future as they have been in the past. High-end graphics cards are a
>> requirement in the AI world we'll be living in well before you will want
>> to retire today's computer purchase.
>>
>> Even if you can afford a computer used exclusively for amateur radio you
>> are likely to benefit from buying the? most powerful hardware you can
>> afford rather than buying what's needed for using the software hams are
>> using today.
>
> + That's a slippery slope:
>
> 1. It's not clear that high-end graphics cards will be a requirement for using AI. Yes, the development and training of Large Language Models (LLMs) currently involves the heavy use of graphics processing units (GPUs) because GPUs can rapidly execute the matrix multiplication operations employed in neural networks. However, exploiting LLMs - which is what hams would likely do - does not require such computation. Furthermore, the DeepSeek LLM demonstrates that the race for market share led many LLM developers to forgo optimizations that enable training with significantly less computation.
>
> 2. With respect to CPUs, plotting available CPU models on a price vs. performance graph generally reveals a suite spot. At the high end, paying an extra 25% to get 10% more performance makes little sense.? Larger CPU caches are worthwhile. More CPU cores will definitely make DXLab applications run faster. As described in the "Hardware Capabilities" section of
>
>
>
> SpotCollector alone can make good use of 3 cores. One core for each additional application you expect to consume CPU cycles in parallel is a reasonable rule of thumb. Don't count the DXLab Launcher, as it only consumes significant cycles during upgrades. You won't need 256 cores unless you're going into weather forecasting.
>
> 3. With respect to RAM, more is always better. While 32 GB may be more than sufficient today, a motherboard with 32GB that is expandible to 64GB when you need it would be preferable.
>
> 4. Secondary storage - whether solid state or rotating - is easy to expand as your needs evolve. The 4 terabyte rotating drive I use for local backup connects to my laptop via USB 3.2 and cost $112! No, it won't meet my needs if I start DXing for dark matter and begin recording and analyzing data from radio telescopes, but likely other upgrades will be required if I head in that direction.
>
> 73,
>
> Dave, AA6YQ
>