Posts by David Riese

1) Message boards : Number crunching : After 48 hours Boinc not crediting collatz credits, any reason? (Message 2710)
Posted 9 days ago by David Riese
Post:
You can run older Mac laptops without a battery. Not practical to do so with more recent models, as the batteries cannot be removed without disassembling the unit. Yet another drawback of these units. :-(
2) Message boards : Number crunching : After 48 hours Boinc not crediting collatz credits, any reason? (Message 2705)
Posted 11 days ago by David Riese
Post:
I am on my third battery on my 2012 13" MacBook Pro and on my second battery on my 2015 13" MacBook Pro. In each case, the battery swelled to the point that the case warped and I was unable to depress the trackpad. Have also had a couple of ancient iPhones fail due to swollen batteries. I wasn't crunching with any of these units at the time the battery failed. However, all of these units were connected 24/7 to a charger. So, you might think about taking your unit off of the charger on a periodic basis in order to discharge and recharge the battery.
3) Questions and Answers : Macintosh : MacOS High Sierra - (17G12034) - Web Driver 387.10.10.10.40.135 (Message 2663)
Posted 18 days ago by David Riese
Post:
i am so delighted that you successfully installed the GTX 770 in Daedalus and congratulation on your plan to use your GTX 970 in Andromeda!

Email me when you are ready to update the firmware on Daedalus (in preparation for the upgrade to Mojave). I'll definitely find a time that is convenient for the two of us to work on this.

Just finished a little "hacking" of my own. Thanks to the wonderful folks at egpu.io, I connected AMD RX 5700XT GPUs in Akitio eGPU enclosures to an older MacBook Pro and an even older Mac Mini. They shared some essential tricks and software tools that enabled these Thunderbolt 1/2 machines to work with Thunderbolt 3 enclosures.
4) Questions and Answers : Macintosh : MacOS High Sierra - (17G12034) - Web Driver 387.10.10.10.40.135 (Message 2628)
Posted 22 days ago by David Riese
Post:
I agree - Apple has to consider giving hobbyists/power users access to GPU controls, including fan controls.

IMHO, Apple needs to rethink its entire paradigm for supporting PowerUsers and Hobbyists. I could assemble a Mac system that would have the computing power to crunch ~20M Collatz credits a day for about $3000 - or less than half of the cost of a base-model MacPro 7,1. So, while I find the new MacPro to be absolutely stunning visually, the cost/benefit analysis leaves me cold.


And I think that's exactly why TL04 has done what he did to get the Apple software on his pc. Great security, great software but waaaay too 'mine, mine mine' thinking. Personally I would like to see them open a side business of selling parts, not machines PARTS, to let people build their own Apple machines without having to cobble together stuff. And yes that would mean selling the software as well maybe in a way to let you buy say 5 licenses and you could own a max of 5 machines. ANY Apple machines, meaning the software goes with you to your new machine if you build one but you have to get rid of the old machine to do it, assuming you already have 5 machines. None of this MS crap where you have to buy new OS software everytime you ugrade your machine.

We can dream can't we?!!!


I would love it if Apple would sell parts that would allow users an opportunity to mix and match components and build customized systems. I am guessing that Apple would argue that the Windows and Linux platforms already offer that DIY experience, making it unnecessary for Apple to compete in that market segment. But, as you said, we can dream, particularly after listening to Tim Cook's moving tribute to Steve Jobs: https://www.youtube.com/watch?time_continue=123&v=iaeKuzB9DS8&feature=emb_logo
5) Message boards : Number crunching : Post your highest find & number of steps (Message 2626)
Posted 23 days ago by David Riese
Post:
2805 steps for 6659467027936692178847 on 2019-11-11 01:33:11

By the way, you can find all-time, recent, and daily high step results here: https://boinc.thesonntags.com/collatz/high_steppers.php
6) Message boards : Cafe : The say hello thread :) (Message 2619)
Posted 23 days ago by David Riese
Post:
Hello from Auburn, Alabama, the "loveliest village on the plains" and home to Auburn University!
7) Questions and Answers : Macintosh : MacOS High Sierra - (17G12034) - Web Driver 387.10.10.10.40.135 (Message 2612)
Posted 25 days ago by David Riese
Post:
I agree - Apple has to consider giving hobbyists/power users access to GPU controls, including fan controls.

IMHO, Apple needs to rethink its entire paradigm for supporting PowerUsers and Hobbyists. I could assemble a Mac system that would have the computing power to crunch ~20M Collatz credits a day for about $3000 - or less than half of the cost of a base-model MacPro 7,1. So, while I find the new MacPro to be absolutely stunning visually, the cost/benefit analysis leaves me cold.


[quote]5-4-2020 at 8:42 PM - PDT

Thanks guys. I'll miss both of you, too.

There's a small chance that I'll have the $$$ for the
AMD/ATI Card in 4 Months. That will at least run
StarCraft:Remastered without overheating. Right
now with the damage done to the 1070, the Card
is heating up to 78C on the Game. NOT good.

Apple REALLY needs to rethink Video Driver Controls,
and allow GPU Fan Control along with System Fan
Control. BUT, as their past deeds show, THEY WON'T!


TL[/quote
8) Questions and Answers : Macintosh : MacOS High Sierra - (17G12034) - Web Driver 387.10.10.10.40.135 (Message 2605)
Posted 27 days ago by David Riese
Post:
I am so sorry that you have run into this problem using your 1070 to crunch on behalf of Collatz. I don't know what to say other than we will miss you!

Best wishes, and perhaps we will see you down the (virtual) road.
9) Message boards : Science : Here is how a good project, GIMPS, communicates its results (Message 2590)
Posted 30 Apr 2020 by David Riese
Post:
I agree that this link is useful for the public to see the upper limit of user progress being made. But there are so many more results that need to be listed. Just because people are currently running ~7.7 × 10^21 doesn't mean that these have been fully processed on the project's end. Necessary results are...
- an explanation of where he started (2^71)
- which numbers are tested (every integer after 2^71 ???). If he does every integer, there is an algorithm that is most likely faster: https://boinc.thesonntags.com/collatz/forum_thread.php?id=153#2325
- the highest integer tested for which there are no gaps between it and 2^71
- some statement about whether or not the results of these tested integers have been processed to be checked for 128-bit-integer overflow or for exceeding of the algorithm's max steps (these would be the integers that could disprove the Collatz conjecture!)


One more thing. With respect to your second comment, I don't know what you mean when you ask "which numbers are tested". My understanding is that Collatz maintains a master lookup table of ALL numbers that have been shown to satisfy the Collatz conjecture. If this is the case, and if that master lookup table was updated on a regular basis (which I assume it must be), Collatz would not have to issue EVERY number as something to be tested. My point is best illustrated with a very, very, very, simple example. Say a task asked a cruncher to evaluate the number 133. The sequence of numbers would be as follows: 133, 400, 200, 100, 50, 25, 76, 38, 19, 58, 29, 88, 44, 22, 11, 34, 17, 52, 26, 13, 40, 20, 10, 5, 16, 8, 4, 2, and 1. So, not only does 133 fulfill the conditions of the Collatz conjecture, so do all the numbers that follow in this sequence, including 400 and 200 - both of which are greater than the assigned number. The "master" lookup table could be updated with these two numbers, thereby meaning that these particular numbers do not have to be assigned to be crunched in their own task. This would yield computational savings.

By the way, my understanding is that each work unit also contains a subset of the master lookup table of numbers (that have been shown to satisfy the Collatz conjecture) or a "rule" that represents this subset of the master lookup table. If this is the case, once a particular Collatz calculation reached one of these numbers (or fulfilled the rule), then that calculation would cease. So, in the example above, if the work unit lookup table contained every number less than or equal to 10 (>= 1 log less than the number being analyzed), then this particular calculation would cease after 7 multiplication/addition steps and 15 division steps - a computational savings of 1 multiplication/addition step and 5 division steps.

I don't code in Python (I cut my teeth on Fortran-77 running on a PDP-11/70 and a MicroVax, and nowadays I use R), so I can't comment on your optimization. But, my gut tells me that there is a lot of optimization available through the judicious use of lookup tables and or rules that represent these lookup tables. Perhaps this is what your Python code represents.


Now that I have thought about it, there is a possible reason why "power users" might be interested in Collatz. The Collatz code allows for a certain degree of user-specified optimization and there is a very active conversation about how to "tweak" these parameters to improve the performance of different GPUs ( https://boinc.thesonntags.com/collatz/forum_thread.php?id=8 - over 51000 views). In my opinion, this allows for Collatz participants to take a level of ownership in Collatz that is not available in many other BOINC projects. And, in my opinion, this higher level of ownership allows for greater participant satisfaction and improved retention of users. Think of it as a way for users to do their own experiments. What's great about this operational model is that personal experience tells me that numerous Collatz users are very generous in sharing their insights with other members of the Collatz community, even though this sharing may result in the ranking of the user doing the sharing being negatively impacted. Indeed, I can't tell you how many times other Collatz users have given me invaluable advice on how to keep my hardware running and make it more productive. The average age (from date of introduction to today) of my 17 crunchers at home and at work is 9.78 years, so I need a lot of help to keep this hardware up and running!

So, the bottom line is that there is a significant amount of code optimization going on all the time at Collatz. Every time a new GPU enters the marketplace, there is a fresh discussion of how best to optimize the Collatz code to accommodate the improved features of that new hardware. For folks who are interested in optimizing hardware and software performance, it can be a lot of fun. Maybe that's why so many "power users" come to Collatz - they want to share in the fun.
10) Message boards : Science : Here is how a good project, GIMPS, communicates its results (Message 2585)
Posted 30 Apr 2020 by David Riese
Post:
I agree that this link is useful for the public to see the upper limit of user progress being made. But there are so many more results that need to be listed. Just because people are currently running ~7.7 × 10^21 doesn't mean that these have been fully processed on the project's end. Necessary results are...
- an explanation of where he started (2^71)
- which numbers are tested (every integer after 2^71 ???). If he does every integer, there is an algorithm that is most likely faster: https://boinc.thesonntags.com/collatz/forum_thread.php?id=153#2325
- the highest integer tested for which there are no gaps between it and 2^71
- some statement about whether or not the results of these tested integers have been processed to be checked for 128-bit-integer overflow or for exceeding of the algorithm's max steps (these would be the integers that could disprove the Collatz conjecture!)


One more thing. With respect to your second comment, I don't know what you mean when you ask "which numbers are tested". My understanding is that Collatz maintains a master lookup table of ALL numbers that have been shown to satisfy the Collatz conjecture. If this is the case, and if that master lookup table was updated on a regular basis (which I assume it must be), Collatz would not have to issue EVERY number as something to be tested. My point is best illustrated with a very, very, very, simple example. Say a task asked a cruncher to evaluate the number 133. The sequence of numbers would be as follows: 133, 400, 200, 100, 50, 25, 76, 38, 19, 58, 29, 88, 44, 22, 11, 34, 17, 52, 26, 13, 40, 20, 10, 5, 16, 8, 4, 2, and 1. So, not only does 133 fulfill the conditions of the Collatz conjecture, so do all the numbers that follow in this sequence, including 400 and 200 - both of which are greater than the assigned number. The "master" lookup table could be updated with these two numbers, thereby meaning that these particular numbers do not have to be assigned to be crunched in their own task. This would yield computational savings.

By the way, my understanding is that each work unit also contains a subset of the master lookup table of numbers (that have been shown to satisfy the Collatz conjecture) or a "rule" that represents this subset of the master lookup table. If this is the case, once a particular Collatz calculation reached one of these numbers (or fulfilled the rule), then that calculation would cease. So, in the example above, if the work unit lookup table contained every number less than or equal to 10 (>= 1 log less than the number being analyzed), then this particular calculation would cease after 7 multiplication/addition steps and 15 division steps - a computational savings of 1 multiplication/addition step and 5 division steps.

I don't code in Python (I cut my teeth on Fortran-77 running on a PDP-11/70 and a MicroVax, and nowadays I use R), so I can't comment on your optimization. But, my gut tells me that there is a lot of optimization available through the judicious use of lookup tables and or rules that represent these lookup tables. Perhaps this is what your Python code represents.
11) Message boards : Science : Here is how a good project, GIMPS, communicates its results (Message 2580)
Posted 29 Apr 2020 by David Riese
Post:
Collatz has "published" results of our crunching. The results page for every user indicates the "best result" for that the particular user. The best result is defined as the number that requires the most steps to be resolved to 1. My best result required 2805 steps to resolve 6659467027936692178847. Collatz also publishes the "best result" on a daily basis, on a recent basis (last 30 days), and on an all-time basis. This information can be found here:

https://boinc.thesonntags.com/collatz/high_steppers.php

An analysis of today's best results suggests that current work units are resolving numbers of ~7.7 x10^21.


I agree that this link is useful for the public to see the upper limit of user progress being made. But there are so many more results that need to be listed. Just because people are currently running ~7.7 × 10^21 doesn't mean that these have been fully processed on the project's end. Necessary results are...
1. an explanation of where he started (2^71)
2. which numbers are tested (every integer after 2^71 ???). If he does every integer, there is an algorithm that is most likely faster: https://boinc.thesonntags.com/collatz/forum_thread.php?id=153#2325
3. the highest integer tested for which there are no gaps between it and 2^71
4. some statement about whether or not the results of these tested integers have been processed to be checked for 128-bit-integer overflow or for exceeding of the algorithm's max steps (these would be the integers that could disprove the Collatz conjecture!)


I think these are all great things to know. I suggest that you volunteer your assistance to Jon to help make them happen.
12) Message boards : Science : Here is how a good project, GIMPS, communicates its results (Message 2579)
Posted 29 Apr 2020 by David Riese
Post:
In my opinion, in order for the credit system at Collatz@Home to be unethical it would have to be attempting to monopolize a scarce resource (volunteers). However, BOINCStats reports that there have been only 65,138 Collatz@Home volunteers, in comparison to 1,346,896 Rosetta@Home volunteers. So, it appears that the credit system at Collatz@Home is not ethical.


Hi David, thanks for the user-count info and great point! Something to maybe consider is that Collatz *might* attract a certain type of credit-hungry power user who has far more resources than the average user. If I were a credit-hungry power user, I'd choose Collatz. I don't know how much *work* is actually being done at Collatz, so comparing user counts may not be a good approximation.


Something akin to what you propose has already been done. At least one project under the BOINC umbrella enabled the use of ASICs and granted a large number of credits for the work performed by those ASICs. My understanding, and I could be wrong, is that the disparity between the relatively large number of credits earned by the ASICs and the relatively small number of credits earned by GPUs and CPUs crunching the same algorithms reflected the disparity in "productivity". In any event, BOINCStats illustrates the effect quite vividly. In terms of total credits earned, I rank #184 in the world overall. However, in terms of credits earned by non-ASIC crunching, I rank #33 in the world overall. There are numerous illustrations for the other side of the coin - one cruncher ranks #3 in the world by total credits earned (~1.7 trillion credits). However, in terms of credits earned by non-ASIC crunching, s/he ranks only 29th in the world (~27 billion credits). So, yes, there are things that projects can do to "attract" users. But, Collatz has obviously not been the best at doing so. Is it better than some projects? Absolutely. Many projects? Possibly.

As for your assertion that Collatz attracts "a certain type of credit-hungry power user", I won't speculate on what drives that group. But I will comment on what drives me. As someone who studied mathematics as an undergraduate, who used to work as a computer programmer (writing very early generation proteomics and autonomous navigation software), and is now a practicing quantitative biologist, I am quite interested in this attempt to "disprove" the Collatz conjecture. Specifically, what drives me is having one of my computers identify the number that satisfies the conditions of the Collatz conjecture with the largest number of steps. When I checked yesterday, one of my computers had found the number that required the 8th largest number of steps. Maybe tomorrow one of them will find the elusive #1. Of course, even if it does, that record could be broken milliseconds later. And so the chase continues ... <g>
13) Message boards : Number crunching : Newbie question - how to get multiple GPUs to crunch (Message 2557)
Posted 25 Apr 2020 by David Riese
Post:
Hi Vester:

Thanks so much for the help, It worked like a charm. +1!

Dave
14) Message boards : Number crunching : Newbie question - how to get multiple GPUs to crunch (Message 2555)
Posted 25 Apr 2020 by David Riese
Post:
By the way, both GPUs are functional (in terms of crunching Collatz tasks) - I have tested this by restarting BOINC with only one eGPU turned on, with only the other eGPU turned on, and with both GPUs turned on.
15) Message boards : Number crunching : Newbie question - how to get multiple GPUs to crunch (Message 2554)
Posted 25 Apr 2020 by David Riese
Post:
I feel like such a newbie for asking this question - is there some setting that controls how many GPUs can crunch at a time? I am asking because I have a Mac mini 8,1 (ID: 864357; OS 10.15.4; BOINC 7.16.6.) connected to two VEGA 56 GPUs (via Thunderbolt 3 and Node Titan enclosures) and this box will crunch only 1 GPU task at a time. I don't think it is a problem with BOINC or Collatz recognizing both GPUs, given the log report below. I don't have this problem with an iMac 18,1 running two RX 580 eGPUs (ID: 852518; OS 10.14.6; BOINC 7.14.2 ) or with a MacPro 5,1 running two internal GTX 1070s (ID: 825825; OS 10.12.6; BOINC 7.14.3). BTW, my Collatz config files got deleted in the process of updating to BOINC 7.16.6 and it took me a little while to figure that out. Thanks in advance for any assistance you can supply.

Sat Apr 25 12:06:40 2020 | | Starting BOINC client version 7.16.6 for x86_64-apple-darwin
Sat Apr 25 12:06:40 2020 | | Libraries: libcurl/7.58.0 OpenSSL/1.1.0g zlib/1.2.11 c-ares/1.13.0
Sat Apr 25 12:06:40 2020 | | Data directory: /Library/Application Support/BOINC Data
Sat Apr 25 12:06:40 2020 | | OpenCL: AMD/ATI GPU 0: AMD Radeon RX Vega 56 Compute Engine (driver version 1.2 (Mar 5 2020 22:38:16), device version OpenCL 1.2, 8176MB, 8176MB available, 7267 GFLOPS peak)
Sat Apr 25 12:06:40 2020 | | OpenCL: AMD/ATI GPU 1: AMD Radeon RX Vega 56 Compute Engine (driver version 1.2 (Mar 5 2020 22:38:16), device version OpenCL 1.2, 8176MB, 8176MB available, 7123 GFLOPS peak)
Sat Apr 25 12:06:40 2020 | | OpenCL: Intel GPU 0: Intel(R) UHD Graphics 630 (driver version 1.2(Mar 15 2020 21:29:48), device version OpenCL 1.2, 1536MB, 1536MB available, 211 GFLOPS peak)
Sat Apr 25 12:06:40 2020 | | OpenCL CPU: Intel(R) Core(TM) i5-8500B CPU @ 3.00GHz (OpenCL driver vendor: Apple, driver version 1.1, device version OpenCL 1.2)
Sat Apr 25 12:06:40 2020 | | Host name: DJR-Home-Office-Mac-Mini.local
Sat Apr 25 12:06:40 2020 | | Processor: 6 GenuineIntel Intel(R) Core(TM) i5-8500B CPU @ 3.00GHz [x86 Family 6 Model 158 Stepping 10]
Sat Apr 25 12:06:40 2020 | | Processor features: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clfsh ds acpi mmx fxsr sse sse2 ss htt tm pbe pni pclmulqdq dtes64 mon dscpl vmx smx est tm2 ssse3 fma cx16 tpr pdcm sse4_1 sse4_2 x2apic movbe popcnt aes pcid xsave osxsave seglim64 tsctmr avx rdrand f16c
Sat Apr 25 12:06:40 2020 | | OS: Mac OS X 10.15.4 (Darwin 19.4.0)
Sat Apr 25 12:06:40 2020 | | Memory: 8.00 GB physical, 208.28 GB virtual
Sat Apr 25 12:06:40 2020 | | Disk: 233.47 GB total, 195.27 GB free
Sat Apr 25 12:06:40 2020 | | Local time is UTC -5 hours
Sat Apr 25 12:06:40 2020 | Rosetta@home | Config: excluded GPU. Type: intel_gpu. App: all. Device: all
Sat Apr 25 12:06:40 2020 | collatz | Config: excluded GPU. Type: intel_gpu. App: all. Device: all
Sat Apr 25 12:06:40 2020 | Rosetta@home | Config: excluded GPU. Type: ATI. App: all. Device: all
Sat Apr 25 12:06:40 2020 | | cc_config.xml: bad URL in GPU exclusion: http://einstein.phys.uwm.edu/
Sat Apr 25 12:06:40 2020 | collatz | General prefs: from collatz (last modified 11-Mar-2020 11:45:07)
Sat Apr 25 12:06:40 2020 | collatz | Host location: none
Sat Apr 25 12:06:40 2020 | collatz | General prefs: using your defaults
Sat Apr 25 12:06:40 2020 | | Reading preferences override file
Sat Apr 25 12:06:40 2020 | | Preferences:
Sat Apr 25 12:06:40 2020 | | max memory usage when active: 4096.00 MB
Sat Apr 25 12:06:40 2020 | | max memory usage when idle: 7372.80 MB
Sat Apr 25 12:06:41 2020 | | max disk usage: 200.26 GB
Sat Apr 25 12:06:41 2020 | | max CPUs used: 3
Sat Apr 25 12:06:41 2020 | | (to change preferences, visit a project web site or select Preferences in the Manager)
Sat Apr 25 12:06:41 2020 | | Setting up project and slot directories
Sat Apr 25 12:06:41 2020 | | Checking active tasks
Sat Apr 25 12:06:41 2020 | Rosetta@home | Task r3x_6850_fold_SAVE_ALL_OUT_920325_446_0 is 0.08 days overdue; you may not get credit for it. Consider aborting it.
Sat Apr 25 12:06:41 2020 | collatz | URL https://boinc.thesonntags.com/collatz/; Computer ID 864357; resource share 40
Sat Apr 25 12:06:41 2020 | collatz | Your settings do not allow fetching tasks for CPU. To fix this, you can change Project Preferences on the project's web site.
Sat Apr 25 12:06:41 2020 | collatz | Your settings do not allow fetching tasks for Intel GPU. To fix this, you can remove GPU exclusions in your cc_config.xml file.
16) Message boards : Number crunching : Mining rig for collatz etc crunching? (Message 2544)
Posted 23 Apr 2020 by David Riese
Post:
Nice rig! Wish I had your expertise!
17) Message boards : Science : Here is how a good project, GIMPS, communicates its results (Message 2543)
Posted 23 Apr 2020 by David Riese
Post:
In my opinion, in order for the credit system at Collatz@Home to be unethical it would have to be attempting to monopolize a scarce resource (volunteers). However, BOINCStats reports that there have been only 65,138 Collatz@Home volunteers, in comparison to 1,346,896 Rosetta@Home volunteers. So, it appears that the credit system at Collatz@Home is not ethical.

Collatz has "published" results of our crunching. The results page for every user indicates the "best result" for that the particular user. The best result is defined as the number that requires the most steps to be resolved to 1. My best result required 2805 steps to resolve 6659467027936692178847. Collatz also publishes the "best result" on a daily basis, on a recent basis (last 30 days), and on an all-time basis. This information can be found here:

https://boinc.thesonntags.com/collatz/high_steppers.php

An analysis of today's best results suggests that current work units are resolving numbers of ~7.7 x10^21.
18) Questions and Answers : Macintosh : MacOS High Sierra - (17G12034) - Web Driver 387.10.10.10.40.135 (Message 2512)
Posted 17 Apr 2020 by David Riese
Post:
I agree that it would be nice if the MacOS allowed for control of GPU fan speeds. As a stopgap measure, I use TG Pro (Tunabelly Software) to crank up the system fans. As a result, my MacPro 5,1 running a flashed RX 580 and Mojave <https://boinc.thesonntags.com/collatz/show_host_detail.php?hostid=858960> sounds like a small jet plane. Good luck with the upgrades! Once you get the Hackintosh running again, feel free to benchmark your RX580 against the one in this rig.
19) Message boards : Cafe : The Last Person To Post Here Wins 11 (Message 2482)
Posted 10 Apr 2020 by David Riese
Post:
Hey TL, I love Mikey's idea that you pull the existing HD out of your box, replace it with a blank HD, install OS 10.12 on it, and see if that enables the GTX cards to work again. A standard install of OS 10.12 should fit on a 128 GB thumb drive ($20-$30) and should boot pretty fast, too.
20) Questions and Answers : Macintosh : MacOS High Sierra - (17G12034) - Web Driver 387.10.10.10.40.135 (Message 2481)
Posted 10 Apr 2020 by David Riese
Post:
One more thing. I would keep your eye on the price for the flashed AMD VEGA 56 at MacVidCards (currently $450). That GPU, like the RX580, is compatible with Mojave in a MacPro 5,1.

Moreover, in an Akitio Node Titan eGPU enclosure connected to a MacMini, an unflashed VEGA 56 generates ~6 million Collatz credits/day. That's significantly more than the ~3.6 million Collatz credits generated by each of my two unflashed RX 580s in OWC Mercury Helios eGPU enclosures connected to an iMac. I know that some folks say that the VEGA 56 is power hungry, noisy, and hot. I haven't done any formal measurements, but I have not observed that my VEGA 56/MacMini setup is any warmer than its predecessor (MacPro 5,1 with a GTX 1080 that generates ~6 million Collatz credits/day).

So, I think the VEGA 56 will keep you in a price-performance sweet spot longer than the RX 580. Indeed, the benchmarking I have seen indicates that the AMD 5700 XT is not significantly faster than the VEGA 56, although it reportedly draws less power and generates less heat.


Next 20


©2020 Jon Sonntag; All rights reserved