Why people are getting mad in comment section.😂 He is telling us about Hardware requirements for HFT. Its obvious HFT is used by big players there is no room for retailers.
Hft is not feasible for a retail traders, the brokerage alone will make you broke... And apart from server there are so many overheads that are not talked about in this video.
@@themvp_in if you are not experts in trading how can you recommend servers for that? lol HFTs are using FPGAs and they are not collocating at some random DCs but at the exchanges
It’s just another video. Someone trying to sound like they know something. If you knew how to do this and make it worth it, would you really be creating competitors? No. There is no way this guy even verified that many transactions a second. It would be A few hours to bench mark and due to his focus on hardware which is trivial … he’s a lying dipshit.
This is misleading as the accuracy of a model is influenced by factors like architecture, data quality, hyperparameters, and training methodology, not the hardware used. GPU's have a speed advantage and when training on a large financial dataset spanning across multiple years, CPU's do not make the cut. Better phrasing would be to simply say that smaller models can run on CPU and GPU may not be worth it for this specific use case. But then for smaller models both CPU and GPU's are a bad choice and this is where you shift to a reconfigurable ic.
@@nasserakbar8282 The chart presented in the image is misleading because it does not actually compare CPU versus GPU performance, as one might initially assume from the context. Instead, it focuses on comparing the performance of different versions of the LLaMA-2 model, specifically the 70B, 13B, and 7B parameter versions, across various benchmarks. The chart appears to highlight the accuracy differences between these model sizes rather than the impact of different hardware (CPU vs. GPU) on performance. Moreover, the chart itself seems incomplete, as it does not provide a full set of data, making it challenging to fully comprehend the results and draw clear conclusions. And instances where it can work better because of memory limit, you would still be better off with quantization techniques to fit the model in smaller vram vs trying to run it on a cpu because of speed issues you will face. 189% more accurate in comparison to what ? Which GPU, this info is nowhere to be found, nor do we have a method to do the tests ourselves, and Llama 2 is literally last gen at this point of time, there are better models released with similar weights with better support for GPU.
Guys , stop chasing rainbows. I would advise you to use your money wisely and have a bigger account to trade. You can build a low latency trading trading computer that will cost between $2500- $5,000. Learn to trade using support/resistance and few indicators. They key is to know what you are doing an to be consistent. The more money you have, the better. Been trading for 15 years and it's hard work .
Why people are getting mad in comment section.😂
He is telling us about Hardware requirements for HFT.
Its obvious HFT is used by big players there is no room for retailers.
after 10 years experience in finance i can confidently say the mouse and keyboard rgb lighting is the most important in HFT
Meaning
Sir I need ur help
For hft the minimum capital should be 60cr -100cr
Dont be fool guys
As retailer there is no need HFT trading
itna brokrage aur tax jayega ki capital hi khtm ho jayega
this video isnt exactly for retail traders
Haan bhai hedge fund wale dekhte hain apki video @@themvp_in
@@themvp_in bold of you to assume that institutions would come on youtube for "How to build an Algo Trading PC?" Dayum.
@@KrsnaRahegaTumhareBaadBhi😂😂
Then you picked the wrong segment. @@themvp_in
From which website should we buy accessories... As you mentioned you sell accessories too
Hft is not feasible for a retail traders, the brokerage alone will make you broke... And apart from server there are so many overheads that are not talked about in this video.
We’re no experts in Trading - this video is primarily for giving hardware recommendations for people interested in building an HFT server
@@themvp_in oh I got it now, didn't see the channel before..👍
@@themvp_in if you are not experts in trading how can you recommend servers for that? lol HFTs are using FPGAs and they are not collocating at some random DCs but at the exchanges
What about taxes on transactions, how they manage it
It’s just another video. Someone trying to sound like they know something. If you knew how to do this and make it worth it, would you really be creating competitors? No. There is no way this guy even verified that many transactions a second. It would be A few hours to bench mark and due to his focus on hardware which is trivial … he’s a lying dipshit.
Minimum capital required 😂
if you start with billon you will be left with a million in no time 😅
And how much it cost you to buy a NSE server?
u rent space there
@@themvp_inwhat is the process for renting a space in NSE
Moosabhai kidhar hai ?
in next video : )
Can I use this cpu for mining crypto ?
buy from nse and sell on bse on intraday is not possible
It will take 12 lakhs to build it now
This is misleading as the accuracy of a model is influenced by factors like architecture, data quality, hyperparameters, and training methodology, not the hardware used. GPU's have a speed advantage and when training on a large financial dataset spanning across multiple years, CPU's do not make the cut. Better phrasing would be to simply say that smaller models can run on CPU and GPU may not be worth it for this specific use case. But then for smaller models both CPU and GPU's are a bad choice and this is where you shift to a reconfigurable ic.
2:55 then what’s with this AMD benchmark saying threadripperPRO is more accurate
@@nasserakbar8282 The chart presented in the image is misleading because it does not actually compare CPU versus GPU performance, as one might initially assume from the context. Instead, it focuses on comparing the performance of different versions of the LLaMA-2 model, specifically the 70B, 13B, and 7B parameter versions, across various benchmarks. The chart appears to highlight the accuracy differences between these model sizes rather than the impact of different hardware (CPU vs. GPU) on performance.
Moreover, the chart itself seems incomplete, as it does not provide a full set of data, making it challenging to fully comprehend the results and draw clear conclusions. And instances where it can work better because of memory limit, you would still be better off with quantization techniques to fit the model in smaller vram vs trying to run it on a cpu because of speed issues you will face. 189% more accurate in comparison to what ? Which GPU, this info is nowhere to be found, nor do we have a method to do the tests ourselves, and Llama 2 is literally last gen at this point of time, there are better models released with similar weights with better support for GPU.
And not to mention, you won't be using Llama 2 for HFT, for literally any machine learning operation, you are better off using a GPU.
Is it profitable though?
Nope, Unless Done at Scale(Large Capital) Deployed by you.
Is hft firm like tower research capital Jane Street looting retail trader @@mihirmewada2085
Yes, most of the profitable traders in f&o are hft traders. Retailers can't afford. You can try trading bot with python.
@@vinukn4468what’s the capital needed ?
😂😂😂😂😂😂 joker stt and brokerage laughing at corners 😂😂😂😂
spread is a joke for u guys haha
Just buy a high end mac
Fake video
Is it even possible 😂
if you have at least 50 rs in your stock broker account you will not ask this question 😅😅
Guys , stop chasing rainbows. I would advise you to use your money wisely and have a bigger account to trade. You can build a low latency trading trading computer that will cost between $2500- $5,000. Learn to trade using support/resistance and few indicators. They key is to know what you are doing an to be consistent. The more money you have, the better. Been trading for 15 years and it's hard work .