Reply

When will SQ add GPU thread work?

7 replies

gusyoan

Customer, bbp_participant, community, 21 replies.

Visit profile

4 years ago #254816

It seems Xeon can not match professional needs enough, does SQ team plan to add more functionalities part like GPU? even an evolution road map like Bitcoin miner will be better, CPU-GPU-FPGA-ASIC.

0

ivan

Subscriber, bbp_participant, community, 236 replies.

Visit profile

4 years ago #254818

regarding using GPU in SQ, you can read here a recent post from the administrator

https://strategyquant.com/forum/topic/parallel-processing-for-genetic-training-and-generation/#post-247279

The administrator, Mark Fric, has also explained in more older posts these principles like in this one

https://strategyquant.com/forum/topic/1774-what-type-of-computer-specifications-do-you-all-have/page/2/#post-126669

Regarding Xeon and performance, in the present, the latest AMD Ryzen 9 and Threadrippers are among the best that world technology has now.

Timisoara, Romania
3900X 3.8 Ghz 12 cores, 64GB RAM DDR4 3000Mhz, Samsung 970 EVO Plus M.2 NVMe

0

Gianfranco

Subscriber, bbp_participant, customer, community, 114 replies.

Visit profile

4 years ago #254822

in an old post of mine I expressed if it was possible to use gpu (nvidia) cuda cores for parallel computing
and they said that this is not possible ….
so for those who have money to spend … xeon platinum, gold (two) amd TR amd EPYC but a lot of power without knowing what to do .. is useless … I mean that many focus only on computing power. .hope to pull out super strategies .. I know algotraders who work with mid-level computers and ‘make working strategies … I think that running after all the latest CPU is of little use

it’s just my observation

I have changed the computer 3 times, always faster, the strategies have not improved with the speed of the computer

0

ivan

Subscriber, bbp_participant, community, 236 replies.

Visit profile

4 years ago #254824

Mark Fric had a post, a few years back when he explained at the most basic level that the rules in algotrading and the rules on which a strategy is based, “what if” is simply not compatible with how GPU architecture is designed. Or something similar, i cant find the post and i can only remember pieces from that post. In any case, at least for me and for many others, it was convincing enough.

Another argument but based on logic deductions: today, there are several other software similar to SQX but all designed to be run on CPU. If a working software on GPU was possible, most likely it was already functional and public by now. If such a software doesn’t exist, this in itself is a statement.

Regarding the actual performance of SXQ on different processors, its a related but different subject. I started the thread on PC setups and benchmarks for exactly that purpose, for us all to have a technical, exact view on the situation. In theory, 2 things should happen if you run the software on 2 different processors. Either the same strategy is created faster, in a few days instead of weeks, or in the same unit of time, you get a more complex (and by deduction, a more profitable) strategy.

Timisoara, Romania
3900X 3.8 Ghz 12 cores, 64GB RAM DDR4 3000Mhz, Samsung 970 EVO Plus M.2 NVMe

0

Gianfranco

Subscriber, bbp_participant, customer, community, 114 replies.

Visit profile

4 years ago #254826

more complex is not always better…. eventually I can do more strategies in less time, so that I can have more chances to find some profitable strategy and just matter of numbers… because if you calculate how many combinations the building blocks have …. xeon epyc etc will never be enough because the number is so large that with a current computer you will never be able to do all of them so a good mix with good yield price is fine… great research work on the great opportunities you find about price and performance ratio

0

ivan

Subscriber, bbp_participant, community, 236 replies.

Visit profile

4 years ago #254827

if i may ask, what configuration do you have ? you can post in the dedicated thread, the PC setups

Timisoara, Romania
3900X 3.8 Ghz 12 cores, 64GB RAM DDR4 3000Mhz, Samsung 970 EVO Plus M.2 NVMe

0

gusyoan

Customer, bbp_participant, community, 21 replies.

Visit profile

4 years ago #254840

I have double Xeon  64 threads and 128G LR DIMM in my working station, just expect to get more efficient way in order to improve. I have checked previous posts, seems GPU does not work 🙁

0

ivan

Subscriber, bbp_participant, community, 236 replies.

Visit profile

4 years ago #254842

Xeons were (expressed in the past tense) very capable and today still have some decent punch, but in technology, like phones, things are moving very rapidly and by 2020 standards (because we are entering shortly in 2020), the power consumption and heat emissions of older Xeons are (or have become) unacceptable. This is caused by architecture and process fabrication.

For example a dual CPU configuration based on a Haswell Xeon from 2013, E5-2698 v3 2,3 Ghz 16 cores each, has around 400 – 450W (entire system with 2 CPU). 7 years ago this was top of the top.

https://www.tomshardware.com/reviews/intel-xeon-e5-2600-v3-haswell-ep,3932-9.html

today, with modern architecture which has the so called chiplet design and 7 nm fabrication instead of 22 or 32 nm, you get the same performance at 130W, or at 400-450W you get 5 times the performance with a modern Threadripper.

AMD has practically ended the era of 2 CPU servers and now, all modern servers only have one CPU

Timisoara, Romania
3900X 3.8 Ghz 12 cores, 64GB RAM DDR4 3000Mhz, Samsung 970 EVO Plus M.2 NVMe

0

Viewing 7 replies - 1 through 7 (of 7 total)