Will TPU be reviewing cryptocurrency mining graphics cards?

[FordGT90Concept]

GPGPU is probably still most appropriate. These are still GPUs but being used for general purpose computing. Just because it lacks some external connectors doesn't change what it is. A car without wheels installed is still a car.

 

[CAPSLOCKSTUCK]

A car without wheels installed is still a car.

round these parts it is known as a "spare bedroom"

 

[FordGT90Concept]

That's just depressing.

 

[Ahhzz]

round these parts it is known as a "spare bedroom"

 

[dorsetknob]

. A car without wheels installed is still a car.

round these parts it is known as a "spare bedroom"

Also known as sheds

 

[CAPSLOCKSTUCK]

If it has a sunroof and you live in Dorset it is called a conservatory.

 

[Steevo]

It would be difficult to do as different places have different cost power, and ambient temps could very well affect the performance, and how much crap is running.

 

[CJCerny]

The cryptocurrency "fad" is not a fad. It's gained legitimacy when it is discussed on professional money advice programs on television and radio. It's here to stay, I believe.

Not a fad, I suspect. I think there is a problem with currencies being too fleeting, though. All of a sudden, Ethereum is the hot currency, and all those folks who sank all kinds of real $$ into mining Bitcoin better spend it fast before its value sinks like a stone. And then, in 6 months, when someone else cooks up the next hot currency, Ethereum miners/owners will find themselves in the same boat.

 

[Sasqui]

Case in point: compare RX 580 8 GiB (6.2 TFLOP) to GTX 1060 6 GiB (4.4 TFLOP). Similar price and similar gaming performance but 41% more compute performance.

That baffles me, I'd like someone to explain how NVidia does that. Seems to explain the lower power consumption in games at least.

 

[notb]

I'd argue running an international payment network is it's own benefit, but I am a much stronger supporter of coins that do something with your gpu muscle, and think the future truly belongs to them.

Not a great place for this question, but you've summoned the topic... Why exactly is this new "international payment network" important? Think about it.
What does it give you over currently existing ideas (bank accounts, credit cards, online payment services etc.)?

There are 2 things that made bitcoin fairly popular: anonymity (IMO more problems than advantages) and geek-friendliness (bitcoin is clearly better in every way - it's about "hashing". "cryptography" etc. - not some weird financial stuff).
And of course there are all those people that hate "corrupt banks"

There isn't much to review other than output vs power consumption, aka efficiency.

Depends what we're interested in. This used to be a gaming-oriented site and so where the reviews. But people active here are more geeks than gamers.
We already see changes. Like you know... Ryzen came out and it was better for "productivity" than for gaming, so now a lot more "productivity" tests emerged to show that Ryzen is in fact very fast - just like the community wanted.
What stops us from wanting "productivity" GPU tests? Nothing.
In fact it's obvious that AMD cards - while slightly behind NVIDIA in gaming - have more grunt in simple arithmetic operations (mining is a good example).

That baffles me, I'd like someone to explain how NVidia does that. Seems to explain the lower power consumption in games at least.

Efficiency, optimization, better instruction set... etc.
It's not about how many TFLOPS are available. It's about how many are actually utilized in a particular task and how many FLOPs are needed to perform a typical gaming-type job.

 

[FordGT90Concept]

That baffles me, I'd like someone to explain how NVidia does that. Seems to explain the lower power consumption in games at least.

Rendering a frame is a series of tasks that have to be carried out sequentially before the next frame can render. Put simply, AMD GPUs spend more time idle than NVIDIA's GPUs when gaming.

 

[MrGenius]

GPGPU is probably still most appropriate. These are still GPUs but being used for general purpose computing. Just because it lacks some external connectors doesn't change what it is. A car without wheels installed is still a car.

That makes sense. I want to argue with it. But I can't really find a good enough reason to. The term GPGPU came to mind briefly. I just failed to think it through. I was thinking a GPGPU is still a graphics card, since it still has display connectivity. But some don't(even aside from these I mean). So that argument fails. Anyway...I get what you're saying.

 

[Sasqui]

Efficiency, optimization, better instruction set... etc.
It's not about how many TFLOPS are available. It's about how many are actually utilized in a particular task and how many FLOPs are needed to perform a typical gaming-type jo

Yes, that's obvious. The question is how can AMD be so far behind in optimization.

 

[R-T-B]

What's starting to annoy me is people still referring to these as graphics cards or GPUs. Which, if they can't be used for graphics, they clearly aren't. Maybe calling them ASIC miners isn't totally accurate either. Since the weren't originally designed for mining. Which leaves me somewhat at a loss as to what they might be called instead. Possibly something that implies crypto mining optimization. Like CMOGPU or MOGPU. Better yet IMO leave the G out and call them CMOPU or just MOPU. Or maybe go with the dedicated mining terminology. DMGPU, DMPU, or something.

I dunno. But calling them crypto mining graphics cards/GPUs just seems wrong to me. There's got to be a better way.

I like GPCU

General Purpose Compute Unit.

 

[FordGT90Concept]

Polaris has 36 compute units, for example. Phrase is therefore confusing.

Xeon Phi is a "manycore processor." If there was absolutely no graphics hardware in the Polaris/Pascal chips then I think they would fit the definition of manycore processors as well. Remove the graphics functionality of the GPU and you have yourself a completely different animal though. The way the entire GPU is structured to pump millions of rasterized pixels out as fast as possible.

 

[notb]

Rendering a frame is a series of tasks that have to be carried out sequentially before the next frame can render. Put simply, AMD GPUs spend more time idle than NVIDIA's GPUs when gaming.

IMO it's even worse. Gaming - unlike "productivity" tasks - is interactive. So the next frame also depends on what happened lately in the system.
As such, once a GPU (or CPU) can do things "fast enough", further gains in raw power make little to no difference. Low latency in processing becomes a lot more important.
This has been happening in AMD vs NVIDIA for years and lately became visible in Ryzen as well.

Yes, that's obvious. The question is how can AMD be so far behind in optimization.

They don't care? They prioritize processing power and flexibility?
Think Zen with the ability to build CPUs from CCX-modules (not their first attempt at this idea).
Think CrossFire that can handle 4 different GPUs.

They're aiming at scalable, powerful, relatively simple solutions. But "scalable" and "powerful" often implies "laggy" - there's always an additional piece of hardware that has to "think" what to do with a job.