Jonathan Lei
@jonathanlei0
Followers
699
Following
586
Media
48
Statuses
382
credit card connoisseur // airline miles collector; and also cloud architect @voltagepark
Massachusetts, USA
Joined April 2016
Don't wanna be here?
Send us removal request.
Explore trending content on Musk Viewer
and more
• 1829182 Tweets
LINGORM PANTENE TRIP
• 643934 Tweets
Corinthians
• 129786 Tweets
Flamengo
• 129372 Tweets
Al Smith
• 119816 Tweets
Yankees
• 89765 Tweets
Saints
• 87062 Tweets
Schumer
• 66754 Tweets
CONGRATS CL IG1M400K
• 63049 Tweets
Broncos
• 41035 Tweets
Cleveland
• 39792 Tweets
Guardians
• 37030 Tweets
Saint Dr MSG
• 36249 Tweets
Allan
• 35833 Tweets
APT OUT NOW
• 31937 Tweets
depay
• 25706 Tweets
対象作品
• 22886 Tweets
Stanton
• 19077 Tweets
Ganso
• 18647 Tweets
David Fry
• 17499 Tweets
ROSSI
• 17416 Tweets
Yuri Alberto
• 16643 Tweets
Patrick New Single
• 16600 Tweets
#FridayFitness
• 15415 Tweets
Big Christmas
• 15071 Tweets
Bo Nix
• 14685 Tweets
#EliminaçãoAFazenda
• 14541 Tweets
Athletico
• 13794 Tweets
Mookie
• 12936 Tweets
Rattler
• 12901 Tweets
Fernanda Montenegro
• 12209 Tweets
集英社秋マンガチャ開催
• 11297 Tweets
Gabigol
• 10696 Tweets
Drew Brees
• 10096 Tweets
Dennis Allen
• 10009 Tweets
5:24 AM
@hackmit
. Have 3 psets due at night, but I will not sleep yet- my team’s project is not done!!
11
1
68
Personal website drop! 🎉
7
0
46
Tmr at HackHarvard, TensorDock will be giving 300 limited edition miniature petite glassware. For water and Celsius consumption only, of course
Come meet our team, learn about our GPU cloud, and win some free credits!
(This could either be the best or worst marketing expense…)
5
1
27
My first hackathon project in 2 years: vision OS text to 3D -
@scale_AI
Generative AI hackathon with
@0xSigil
@rrchalamala
@0xSigil
Made in less than 8 hrs :))
Dream with Apple's Vision Pro 🪄
Transform your imagination into a stunning immersive reality 💭
Simply type in your vision, and voila - the objects appear right before your eyes! 🐘🗡️🔮
Built with 💪
@0xSigil
@rrchalamala
@jonathanlei0
at Scale’s Generative AI Hackathon.
12
8
111
1
1
25
Astoundingly true lesson: if someone repeatedly talks about how good they are, they aren’t.
Applies to advice, assistance, money, etc
“My offer is so generous” when repeatedly mentioned, is such a red flag for me
4
1
22
Literally half of my comp is the free travel that comes with the miles we earn… 4x back on cloud service providers with Amex Gold card is insane
Too bad that’s capped at $150k :))
Very interesting hack I just learned.
Put GCP on company credit card, and use earned flyer miles so everyone flies free ...
21
6
292
1
1
23
Today I learned I have a Pitchbook account through Harvard SSO. Huge W
2
0
21
From my calc partner yesterday: “humans are like vectors. We need direction for the magnitude to mean something”
Well said, Marko, well said.
1
0
18
I feel like I’ve been really calm and happy lately. I think that this is because I have a lot of agency over my own life: whatever I want, I will try to find a way to get.
Today I realized that this is something I subconsciously look for in other people as well
2
0
17
This is one of the coolest things I’ve seen in a while.
Big hats off to the team, and proud to see it runs on TensorDock infrastructure!
0
1
14
@G413N
Rent it to me! We pay ~$200/mo/4090 and you can use the money you earn to rent beefy A100s and H100s on demand :)
This looks like a gigabyte mobo btw?
1
0
15
Yesterday, I ran the same ML workloads across a dozen GPUs.
H100 and A100 performance is unbeatable, but the price-to-performance of lower-end cards is much better. L40 and RTX 6000 Ada outperform the A100 at some tasks, as they are 1 gen newer
(thread)
6
1
13
amazing people, amazing program! defo consider applying if you're into human augmentation
We are hosting a residency on Wearable AI & Personal Agents. 16 hackers, 8 weeks, at Harvard & MIT. Join to work on the HARD problems in building deeply personal, privacy-conserving agents. More here
7
16
105
0
0
12
(1/n) what did i do why is everyone running ads on tensordock's name now
is this war? do i retaliate and run ads on their keywords?
3
0
12
Small piece of wisdom through experience: relationships are more important than sales. If a big customer is leaving, do not panic. help them through the process. Graciousness is rewarded in the long run.
0
0
9
Just added!
Carbon neutral NVIDIA A100 80GBs from $1.32/hr
(vs $5.12/hr on GCP/AWS!)
0
1
9
Taking on funding is associated with adding obligations [to generate returns for investors], which in turn decreases your agency. In that sense, I'm really lucky that the investors I work with intentionally do not encroach but somehow remain super helpful
0
0
7
Built using good old plain HTML and hosted on Cloudflare pages. 97/100 on PageSpeed Insights
1
0
7
@itsandrewgao
@TensorDock
Yes! 3090 is goated (and so is the 4090, A5000, 5000 Ada etc)... if you don't need more than 24 GB of VRAM
0
1
5
the actual video has like 5 straight mins of this, Silicon Valley HBO show writers could not even make this up
the greatness starts at -2:09:41
0
0
7
Our H100s are available in full 8x node configurations with full 7.2TB/s NVSwitch and NUMA topologies passed into virtual machines.
Further, preallocated RAM, RAM-based VM filesystem caching, and further optimizations make for near, or even better than bare metal performance :)
1
0
5
TensorDock Marketplace will be going public next week - so excited! Amazing
Preview some of the lowest costs on the private marketplace right now. Prices include CPU/RAM/NVMe SSD:
3090: $0.30/hr - vs $3.60/hr @ iRender
3080: $0.20/hr - vs $0.90/hr @ Genesis Cloud
1650: $0.10/hr
0
1
6
Meta to make an open source Sora alternative?
1
0
6
Update: I know I should not be this happy, but I ended up with the only empty row!
Window seat pax must have decided to move to a front window seat, leaving me with this bed :)
0
0
5
The one thing that remains relatively constant across time seems to be culture.
Competitors come and go, customers come and go, team members come and (eventually) go — but culture remains constant. A strong team hires strong teammates, and the cycle repeats…
1
0
7
Air France roulette - will I get an empty row?
(Flight is in 10 hrs. Normally I’m a front cabin window person, but I figured I have a higher chance of getting a row to myself if I select a back aisle seat… hoping some tour group doesn’t claim all seats and make me regret this)
1
0
5
(5/n) PRICE LOW --> LOW STOCK, NO AD SPEND
Some clouds price low. Longer than 18 month ROI even at 100% utilization. And they get that utilization. They don't spend on ads, and they don't need to—stock is always rented! But it also means you can't scale if you use them.
1
0
3
Up to 8x NVIDIA A6000s available on TensorDock Marketplace for $0.415/hr ea
2
2
5
From what I can tell, these are THE world's cheapest on-demand H100s. Sub-$2/hr pricing is typically what you'd need a 3 year commitment for.
We've priced these in staggered so that you always know you're getting the best price you can while having availability to scale 🎉
0
0
4
by now you definitely could have deployed your own 3 70B on a <$3/hr H100 instead of waiting for chatGPT to come back online 😛
1
0
5
I find tech bros (err, maybe just myself 😅) are drawn to artificial imitations of real life.
Indoor rock climbing, silent discos, DoorDash/Uber Eats, soylent.
(when I suggested to walk on treadmills with VR goggles, a friend didn’t realize it was a joke!)
0
0
5
@CalebDenio
$96 for priority boarding is objectively absurd
And it's not like boarding early for economy gets you anything - who wants to pay to get into the metal sardine can first?
1
0
3
TensorDock: 100+ globally distributed GPUs. Deploy on-demand or bid from just $0.05/GPU/hr.
A4000s from $0.17/hr
V100s from $0.22/hr
3090s from $0.22/hr
4090s from $0.37/hr
A100 80GBs from $1.22/hr
1
2
4
First picture of TensorDock's rackspace at Singapore at Ascenix! Many more servers to come :)
0
2
4
Demand exceeding supply on our marketplace, so having to reserve specific hostnodes to high spending customers so that the machines don't get taken up by other customers first😮💨
0
0
4
TensorDock is hiring!
Why join?
- Build cloud infra that researchers, startups, and hobbyists use to build products of tomorrow on
- Quickly-growing reseller business
- Strong balance sheet, with physical assets spanning BOS to SIN
- Net profitability
0
1
4
Just exited my website hosting business () to
@servarica
- they are an absolute pleasure to work with!🎉
0
0
4
Each one is a TensorDock virtual machine! :)
In a single day, more than 1/10 of our virtual servers are created/deleted. Yet volume is so high that actual usage ends up being rather consistent - generating consistent revenue for hosts!
0
0
4
@ashleyyjoelle
yes i gotchu, shoot me a DM :)
for everyone else - i run a cloud that has them for $0.4-$0.6/hr here!
0
0
3
(6/n) There isn't really a good middle ground; no traditional cloud is both cost-effective AND scalable.
At TensorDock, we're trying to commoditize compute into something that can be traded like grain. Whenever inventory is rented out, suppliers add more—albeit at higher prices.
1
0
3
Our first ad!
0
1
3
@kdrag0n
Rite of passage indeed! During Thanksgiving I finally got around to moving everything (projects, TD microservices, S3 backups, etc) off AWS. Now everything costs <$100/mo combined
Taking the AWS activate credits was a huge mistake in hindsight 😅
egress fees be a cash cow
2
0
2
Peak usage from earlier month: 400 concurrent GPU virtual machines!
0
0
3
I visited
@aaronchewbani
/
@calebsirak
from
@FUN
in July- they’re building infrastructure for crypto authentication. excited to see them launch!
Introducing 🔥🔑🏦
The easiest way to build crypto accounts without compromising usability, trust, or security.
Fully non-custodial. Powered by ERC-4337.
Built on
@BuildOnBase
🚀
/1
98
157
571
0
0
3
(9/9) So... if you need GPUs, try us out! Hundreds of GPUs released and claimed daily
--> 3090s from $0.21/hr
--> A6000s from $0.47/hr
And if you believe in our mission, shoot me a DM! I'm looking to add 1-2 opportunistic full-time hires to our team :)
1
1
3
@amasad
@0interestrates
Holy shit. I run the high school hackathon Hack3 of the T-shirt
@0interestrates
is sporting — keep it strong!
Hope the team you met + workshops you hopefully attended will help you find a job after being fired from twitter🤣
0
0
2
(4/n) and these are the ones that you should only ever use for bursts. In general, if you need X thousand servers, and someone has them in stock available tomorrow, they're going to be overpriced... because that's why nobody else is renting them
1
0
2
@kdrag0n
@willdepue
@LambdaAPI
@runpod_io
LMAOOO true true,
@willdepue
when you need more GPUs, I gotchu as well
0
0
2
(7/n) There's always an equilibrium that guarantees both inventory availability and cost-effectiveness. When you rent servers from us, you get the best deal available. You can scale up/down without needing to integrate multiple cloud providers' APIs, albeit at higher unit prices
1
0
3
@developedbytoby
A welcome change! Though wholesale bandwidth costs almost always average to less than <$0.01/GB at data centers :)
0
0
3
Lambda Labs did a similar GPU benchmark a while ago, and I just wanted to point out that their measured H100 PCIE performance is only 20% better than A100 PCIE performance as well
1
0
2
Just reduced our database usage by 50%+ by merging together individual transactions & creating monthly spend ones 🎉
(Backups made of course, but the primary db should be much snappier now)
1
0
2
@stephenbalaban
@MasterScrat
@stephenbalaban
Just to understand, $100k leads to ONLY $35k spending at reduced costs right?
H100 SXM: $2.8 -> $1.98, $14.5k/yr saved
So $100k upfront would take ~7 years to generate a ROI on for H100 SXMs. For A100 40G, ~11 yrs
2
0
2
Congrats to
@max_elbo
/ (a TensorDock customer) for being featured in The Spectator! Honored to have provided you with the infrastructure to create these wonders :)
Also,
@spectator
, thanks for mentioning us in the image credits!
0
1
2
is one of the companies that never ceases to amaze me. I used to never use Twitter; now, I log on once a day and try to create something new with
@max_elbo
. I feel like I'm becoming a bit of an art connoisseur trying different art styles!
We just cross 2000+ images generated 🥳!
Thank you all for being our early adopters and trying us out 🙏.
If you are thinking of signing up for our creative plan, please use the coupon PICASSO to get 20% off FOREVER! (Only 30 coupons available).
3
0
6
1
1
2
@NothotdogCorp
@skr_elbo
@danielgross
@TensorDock
Thanks for the mention
@skr_elbo
!
@NothotdogCorp
- happy to give you some free credits if you'd like. Shoot me an email :)
0
0
2
@stanleyzhaoo
@pitdesi
@vladtenev
Interchange (what merchant pays so that the card is accepted) after network fees will probably be only ~2% for RH.
So the rest has to come from enough people moving over balances from month to month on their high APR. they’re basically subsidizing people who pay on time
1
0
2
(8/n) It's a win-win-win:
- you get compute at the market's best prices
- suppliers serve more workloads than they would without us (customers use us and pay a premium because they can scale across our entire network of hosts rather than a single one)
- we take a cut.
1
0
1
@stephenbalaban
@MasterScrat
Gotcha - this would probs be the more significant benefit than pure discounts :)
(ROI on money in 7-ish yrs = ~10% YoY return which while good, is hard to justify from a cash allocation standpoint purely from a risk/security view for a startup like mine)
1
0
2
Launching long-term cloud servers for 10 GPU models at 11 locations! More options, lower prices!
NVIDIA A100 servers for only $950/month!
0
0
2
When grouping by VRAM, we see a clear trend. Lower VRAM GPUs consistently provide better value... but that's only useful if your model fits on the GPU.
So rule of thumb: always use the smallest GPU your model can fit on if you want the best value.
1
0
2
(3/n) PRICE HIGH --> PLENTIFUL INVENTORY.
H100s should not cost $5/hr+ on-demand, but some clouds charge that. When they're rented, great! But much of the time, they aren't. Expensive clouds are also the ones dumping money on ads to try to fill them up....
2
0
2
An inside glimpse at the hardware we've been testing over the past month at
#TensorDock
. We rigorously test each hardware configuration before deploying it.
AMD Ryzens coming soon?😉
We have over 1k GPUs in-stock, ready for deployment:
1
1
1
The generational speedups from NVIDIA's newest chips come more from the interconnect that allow you to train or run inference more efficiently across a number of nodes and not the unit performance from individual GPUs :)
0
0
2
@Suhail
For training clusters is probably the move.
If you ever need non-interconnected nodes, we have H100s from $2.49/hr :)
0
0
2
Sneak preview of upcoming TensorDock compute marketplace...
In dark, with organizations so you'll be able to invite your coworkers!
0
0
2
(2/n) In all seriousness, this is symbolic of the widely-differing pricing/business strategies cloud providers employ.
I'm leaving out reserved, prepaid, and lease-to-own arrangements for this short thread.
1
0
1
18 GPU servers created today on the TensorDock Marketplace!🎉
At the lowest stock during past two months
Still available: 9 A6000s from $0.42/hr, 7 3090s from $0.22/hr, etc.
Adding a boatloads more A4000s at $0.14/hr next month along with more 3090s
0
0
2
0
0
2
Run Stable Diffusion under 45 seconds on TensorDock Marketplace by Yusuf Sarıgöz in
@TDataScience
🎉🎉🎉
1
0
2