ftufek

Never thought I'd get a response from Jeremy Howard, now I have to post a well polished article! Thanks for all your teachings btw, I really enjoyed and learned a lot from the fast.ai ML course!

I posted a link just below to twitter with an image of the riser setup. That setup worked well for 4x, but for the 7x we're moving the cards upside down and setting them up like tree branches if you will. So the trunk/floor is the motherboard and you get close to edges, the cards are angled and use longer riser cables.

The issue we had with barebones was cost and cooling, we use 30$ racks from Target and hang the GPUs with metal zip ties and a box fan from below, so they get lots of air and we don't break the bank and can easily roll them around.

Threadrippers are great and I had 4x Threadripper setup for the longest time, but they are a bit more expensive.

The advantage of EPYC is that because it's so common, we can find used cheaper ones on ebay. They are a bit slower I believe, but we can deal with that by using Nvidia's DALI and decoding images on the GPU rather than CPU.

Sure, will do, though it might take some time to finish writing the blog post, you can get a preview of our previous setup with 4x GPUs here: https://twitter.com/ftufek/status/1569367127878139905. For those that are curious, that's running a Threadripper 3970x.

It's not exactly a "clean" one, like a proper 2u/4u chassis and server grade GPUs but it does the job for 70-90% cheaper.

It's not even just about the slots, it's about the PCIe lanes (which is something I never had to worry until now, though I built countless PCs in the past).

We tried bunch of setups with Threadrippers and EPYC, at the end settled for the ROMED8-2T which is a monster motherboard.

RTX 3090, it has 24 gb of memory which is barely enough nowadays. Hoping that 4090s will have 48gb, but we'll know on the 20th.

Looking forward to all the creative AI startups using these cheap cards to build cool products. We've been waiting a long for the merge to finally happen.

I can't believe that you can buy a card that'll do 15+ Tflops for like 500$.

You could, but you'll need server grade parts and a few good electric circuits. Oh and it costs hundreds of dollars in electricity every month.

It's actually not that straightforward to plug in these consumer cards as 4x setup. We spent weeks researching how to achieve up to 7x RTX 3090 setup in a single rig. Could write up our method if anyone is interested.

Not much difference between 3090 or 3090 ti though, while the price is usually 30% higher.

Thanks for the feedback. Yes, that's the right, the main reason is traffic/domain authority. The current site has gathered some backlinks and ranks high for the niche we care about.

The way I look at it, 99.9% of our traffic is consumers and all our paying business customers started out as consumers, since <x>.com can be more brand able and trustworthy, I think it has higher likelihood of ranking even higher eventually which would increase our funnel.

Yeah that was my initial thinking as well. My only concern was discoverability, meaning if we put a huge "Businesses here >" it might be a bit annoying to normal users, but I guess we could design it in a better way.

Out of curiosity, which one is most prominent on the homepage itself?

Only if you exercise the options, you won't exercise the options if it's not profitable to you.

In the same boat, I left 2 months ago to bootstrap my own startup. I'd recommend getting a shared co-working space. It'll give you home/work separation and you'll get to meet others who are in a similar situation and you can bounce ideas off or just socialize. Of course, keeping up with friends, ex-coworkers and getting lunch/coffee every once in a while is nice too. Good luck!

Another thing is, you have to lead other engineers. Not only you shouldn't need constant supervision, you are the supervisor in many cases.

There's a lot of other reasons why it can go down and it wouldn't necessarily go down because of halving. But when a lot of people expect something to go up, it can have reverse effects.

A lot of people have this assumption that it will always go up after a halving, it very well might, but at the current price point, the average person can't even afford a single Bitcoin whereas at previous halving (in 2016) that was still doable (it was on the order of a few hundreds of dollars). At this stage, you would need big investors and probably that's what companies like Bakkt will allow.

The way I look at it, Bitcoin is a historical experiment happening during our lifetimes and mid-May (the time of halving) will be a very interesting milestone in that experiment.

It could go up or down, but either way, it will be interesting to watch.

Having been through that bubble, I'm fairly confident FOMO had a huge effect. Every single person I know was asking how to buy some. You can even see it on Google Trends[1].

It's possible that the very first initial bump was manipulated (and crypto is definitely manipulated each and every day...), but the crazy increase afterwards probably wasn't due to a single entity.

1: https://trends.google.com/trends/explore?date=today%205-y&q=...

Of course you need to have demand for scarcity to be valuable. And I would argue that in the 10 years of it's existence so far, Bitcoin has proven that there is demand for it.

Of course, you can create an identical coin on paper. Many have tried and their attempts have failed because like you said, the human behavior (aka the social aspect, network effects, psychology, etc) also have a big effect.

In my opinion and feel free to disagree, I think Bitcoin has the true potential of being the internet's native currency, especially with many lightning network wallets coming online on the mainnet. It's far from being perfect and unfortunately has attracted many scammers and want-to-get-rich-quick-people over the last few years, but I believe it has a really great potential.

Except that your bitcoin won't have exactly the same properties as bitcoin, mainly the miners which do provide network security.

Isn't the underlying value here mathematically provable scarcity ?

Keep in mind that tutorials will always make it look easy compared to debugging actual production code. If you look through tensorflow tutorials, they also look very easy, especially with TF2.

That said, I've experimented with pytorch and I agree that it is really nice to work with.

Disclaimer: I work at Google and do use tensorflow, though I don't work on the tensorflow team.

You could try Google Photos, it can now search text in images: https://9to5google.com/2019/08/22/google-photos-text-search/.

Disclaimer: I work at Google, though not on this product.