Crowdsupply cell network device based on LimeSDR

Nice looking cases.

It will be interesting to see the pricing on these.

My educated guess is $900 :wink:

Definitely, but seeing as how I have ordered 3 of the USB versions I’ll be holding out for the PCI appliance.

@M0GLO May I politely ask why you need 3 units? What is your project? I’ve got one USB version arriving sometime next week but it’s very much my ‘plan B’ as I’m hoping a slightly simpler and cheaper arrangement will work.

You could ask rudely and it wouldn’t bother me. :slight_smile:
One is for a SDR based DMR repeater system I am building for 70cm, another for a small portable all band ham rig I can pack into an overnight bag with my overnight stuff and the third is for lab work on WiFi and other IT protocol security.

1 Like

For the non Mini it should be nice but big bucks - the cpu ( i7-6950X CPU 10-core 3.0 GHz 25 MB Cache ) is over $1500 and then you have the 64GB memory :smile:

Hopefully there will be enough space in the box to fit a high end GPU card
to do the really processor intensive DSP stuff.

I think they would be better off selling empty boxes and let the end
user spec their own hardware. However in an R&D lab the price is
probably less than an Engineers weekly cost and supplying all the
hardware will reduce tech support costs for Lime.

It looks like neither of the two devices listed are large enough to fit a high end GPU inside. But maybe I’m wrong. High end GPUs are large.

Sell an empty box? You mean sell a system that includes the LimeSDR PCIe with the nice case but no memory or CPU? I thought you can buy a LimeSDR PCIe now.
But more to the target demographic for these devices… The higher end of the two seems to be targeted at towns/villages/encampments that have no infrastructure. I hate to use the word appliance but I believe that is the intended role.

On the topic of GPU what are the calculations in RF that benefit from using the GPU? I am curious because bitcoin miners started out using (abusing) GPUs and then as the difficulty level increased ASICs were needed to mine efficiently. Has anyone started looking at RF ASICs in this context? It seems inevitable.

You mean, RF ASIC as in a digital device that’s optimised for baseband processing? They’ve been around for a long time, in the form of SoCs with hardware accelerated DSP.

In addition to all the configurable RF goodness in the LMS7002M, there is also hardware DSP.

I barely know what I mean, know what I mean? So then GPU use in RF applications is popular because there are simpler APIs to put RF workloads onto GPUs? Or is it simply that GPUs are ubiquitous and mostly idle? A combination of those two?

Are the SoCs with hw accel. DSP expensive? Can you name some of these devices? I am curious of their cost.

I’m not sure quite how popular GPUs are in RF applications, but they do give you a lot of performance for the money, in addition to being commonly available and with reasonably well understood programming models.

Baseband processors are available from Qualcomm, TI and others.

They tend to be sold with support for specific wireless standards and often only to major customers under NDA. So, highly application-specific and not easy to use, but very energy efficient at what they do.

However, there are other much more general purpose and easily obtainable DSPs available too. A topic that is prime for a spot of googling and seeing what is out there :slight_smile:

Note that SIMD extensions in Intel and ARM, e.g. SSE and NEON, are also often used for accelerating DSP (see VOLK as used with GNU Radio). After which GPUs and FPGAs — e.g. the one on the LimeSDR — are probably easier to use for most other offload/acceleration tasks.

I was really talking about using GPUs for R&D, they are easier to program than FPGAs
cheaper than rolling ASICs and come with loads of tools for debugging and optimising.
They are also scalable.

Most signal processing algorithms are fairly simple they just have to happen very fast.
If you can break your problem down into something that can be done in parallel then
GPUs are ideal.

I wouldn’t use them in a commercial product but I would think about using them if I
had the job of building a experimental wireless network for instance.

I did a Google search on GPUs and Wireless and the first item it came up with was
a paper on using GPUs for a massive Wireless MIMO systems. (The GPU used is
really old now compared to what is on the market now).

I am fairly new to GPUs but I have already seen a huge increase in performance by going to
a GPU in comparison with an i7 only solution.

Personally I would rather spend my money on a 4 core i7 and a GPU rather than a 10 core i7.
But I am not in the market for one of these boxes.

1 Like

@andrewback, those NDAs really stifle innovation, you know. Will do some searching, so far have found athletic shoes.

@g4guo, BTW be careful buying used GPUs. Bitcoin miners sell those and many times they have been running fully utilized 24x7 for months and years.

Yes better off buying them from gamers, every time NVIDIA comes out with a new GPU
the market is flooded with last year’s cards. After all, a few extra FPS could mean you
snatch victory from defeat.

There’s an fft display that uses gpu, fosphor: