« first day (1517 days earlier)      last day (3485 days later) » 

12:50 AM
Ahh. Fun. I need to locate a park based off sketchy google maps directions I don't really trust and a photo of the place
 
1:17 AM
@JourneymanGeek no GPS in da phone? sygic is really cheap road gps google store App, that works in android IF your country is properly in the maps.
although it has some online stuff, it is offline, unlike google maps which are great but more often require connect. It is possible to offline google maps, but it is a pain and limited, you can "load it" up while on wi-fi with tiles
 
1:38 AM
@Psycogeek place has no address and even the army got confused
 
singapore Area 51 :-)
 
2:03 AM
removed
 
lol. Its a wierd obscure little park in the new downtown district
 
2:19 AM
If I have 2x300GB drives in a RAID-1 array, what does enabling SPAN in the raid controller do?
Do I need to link it to another RAID-1 array?
 
2:33 AM
Nevermind, figured it out! ^_^
 
3:06 AM
._.
300gb?
SAS?
 
@JourneymanGeek Yes, I believe so.
Could just be SCSI though...
I think it's SAS actually.
 
3:31 AM
I thought SCSI only went up to 137gb?
(which is also an odd size)
 
 
1 hour later…
Ash
4:53 AM
anyone here has an account with ello ?
if ( youHaveAccountWithEllo == true )
invite me;
 
dosen't requesting an invite work?
 
Ash
well , I hope that takes some time.I didn't tried it yet. And i wanted to see if anyone already joined.
 
5:10 AM
ahh
 
 
1 hour later…
6:17 AM
@Bob @allquixotic dammit I was so busy I couldn't even drop a "watch out for something big we're coming with"
and oh, all new autocomplete?
 
helo
i have some zipped file that not utf8
when i zip here the name is strange
i even cant rename it
 
@Sathya: new autocomplete was a Big Thing ;p
 
@JourneymanGeek dang
 
6:39 AM
hhhnnnnrkkk
Got to bring an old server from work home to play around with...
Plug it all in and then find out that it doesn't have a VGA or DVI output D:
 
I mentioned this a few days ago (see the transcript for context):
yesterday, by DragonLord
Who here has heard of a benchmark called HINT?
I've ultimately decided to fork it into a new project called HI64
 
7:02 AM
cool so which one is the download of a compiled binary for winders 64 bit ?
 
No compiled binaries yet
The code builds using any recent version of GCC
 
ok then let me include tested in that too :-)
 
I haven't even finished the documentation
 
Well i dont read manuels anyways :-) it does 4-5 tests shows up in a graph? (or do you have to make that seperate) Then i can use it to compare , in this case different overclock and memory settings.
 
You'd have to build separate versions for each data type
You might want to read the original README in the old-files directory
 
7:12 AM
I scanned over some of that yesterday. I use this thing sometimes for quick and dirty compare of windows only systems.
 
Also, the program writes files to a directory called datafoo where foo is passed as parameter to HI64
 
Because it was made so long ago, unlike other benchies it finishes in seconds, and provides relative results.
 
Be careful with it--it likes to eat up all your RAM and cause heavy paging before finishing
That behavior is in fact by design in the original code
I may add a feature to limit memory usage in the future
 
@DragonLord read that before, i assume the purpose would be to test "All" the memory which would include disk paging virtual junk
 
@Psycogeek Well, HINT was originally designed to test systems big and small for their performance at all storage levels, from cache to disk
I probably won't be able to port the parallel version of HINT to Windows HI64 without someone else contributing
 
7:52 AM
@Bob hi bob see my post update and tell me what to do ! thanks
 
Some QUIPS charts for the Dragon
This one is QUIPS plotted against time
 
...and this one is QUIPS plotted against memory use
These are generated from a 64-bit version of HINT with modifications that are now part of HI64
ADVANCE in hint.h is set to 1.05 for finer data sampling, but much longer run time
System has 24 GB of memory and a Core i7-4800MQ processor
...which has 32 KB L1 instruction and data caches for each core
...256 KB L2 cache per core
...and 6 MB of L3 cache shared across all cores
The memory plot shows a very rough transition from L3 cache to main memory
hint.byu.edu/tutorials/graphs/index.html provides more information about how to read these plots
@Psycogeek? You left as I was posting the QUIPS graphs above.
The plots themselves were created with OpenOffice Calc
 
8:21 AM
Morning people
Is there any where to tell what program is responsible for an entry in the event log? I have an error, the source simply says "hostscan_unknown"
 
Yay, got Refiner \o/ No gold though :(
 
Bob
9:06 AM
@Mahdi Hm.
This is a fresh Windows install, right?
...I don't really have any other ideas.
Do you know what programs you installed?
Possibly malware... or the antivirus itself.
 
9:50 AM
heh... my wife has just made the distinction that Doctor Who? is just a nerdy James Bond.
 
10:38 AM
@Bob this is the fresh windows and installed nod32 endpoint antivirus but it's something really wired whats you guess ?
 
Bob
@Mahdi I... really don't know. You obviously have other things (Firefox, Chrome, IDEA, etc.) installed. What else?
I'm talking any program you installed, any program you downloaded/ran, since installing Windows.
Because this is such a weird issue with no obvious cause.
 
intelliJ studio 2013 office some other normal developer tools
 
Bob
My current guess would be the antivirus. How often does this occur?
 
antivirus is new
no
 
11:09 AM
Any reason why typing Alt+26 wouldn't return a rightwards arrow in OneNote?
It works in this field.
 
11:20 AM
@slhck Alt invokes quick access, I guess
@slhck Try that maybe ;D
Oh, @ autocomplete now has avatars in it
 
@OliverSalzburg → That works!
@OliverSalzburg Since a few days it seems
 
Huh
img.user-image:hover {
    transition:1s;
    max-width:100%;
    max-height:100%;
}
Neat!
Should have added that ages ago
 
11:35 AM
... like many things on SE
 
12:06 PM
Well, that snippet is from my chat theme ;D
 
lol. and kicks, and reasonable roomv owner powers and /me and.... :p
 
@allquixotic: In case you are interested, my problem from the other day has been resolved. I added an answer to my own question.
 
Hi there ! Is it possible to add folder compression at OS / filesystem stage ? i.e. I would not bother with this myself...
 
12:21 PM
Sure
 
Bob
@Basj Which OS?
 
@Bob In order to avoid a X Y problem, can I explain you why I'm asking this ?
 
Bob
@Basj Go ahead.
 
@Bob I'm building this website : bigpicture.bi/demo . Here you can see a better example : bigpicture.bi/1898oa
Each time someone saves, then the version is updates and saved into disk as a new file. Ex : bigpicture.bi/1898oa~1, bigpicture.bi/1898oa~2, etc.
Of course, I could use a database instead but my shared hosting doesn't allow big DB , so I go with files.
 
Bob
...why do you need folder compression for that? O_O
@Basj ...yea, the correct solution is a DB -_-
compression with shared hosting is an entirely different matter
 
12:28 PM
@Bob Now imagine I have 1898oa~1 1898oa~2 1898oa~3, 1898oa~4, etc. They only differ by a few bytes. Of course I could only store the diff
But I would have loved to set : Folder /MYDATA/ Compression On
And then hope that the OS / filesystem does all the complex compression stuff for me :)
 
@Basj: so, basically you find a manure pile, and expect a pony? ;p
 
Bob
if you were on a VPS with Linux, I'd say there's FUSE filesystems that compress
if you were on Windows, I'd say NTFS folder compression
 
@Bob: or he could use a database ;p
 
@JourneymanGeek not english, I don't know what is a manure pile ;)
 
@Basj: Manure is what comes out of the south pointing end of a north facing horse
or cow or other ruminant.
Not fresh, of course.
 
Bob
12:32 PM
in your case, you'll have to either invent your own on-disk DB, use something like sqlite, or use a proper DB
considered changing hosts?
 
I don't know at all what to do, that's why I come here :)
The only thing I know is : I have shared hosting (Apache / Linux), DB access is slow and very limited in size.... but hard drive storage is large and fast
 
Bob
@Basj solution: use a different host -_-
 
@Bob Why is a DB the solution ?
I only need to store files... There is no user management right now
just updated versions of files
So why is DB really the solution ? For what reasons ?
 
Bob
@Basj whatever you do is a DB
even chucking files into a folder, you're essentially using the filesystem as a DB
 
@Basj: The thing is ... you want compression across files.
 
Bob
12:41 PM
if you really want to compress, your only choice is to roll your own or find an existing library
 
But ... OS / filesystems will usually compress files themselves, but not try to compress multiple files with each other.
 
@Bob I believe you. But why is it better to have a complex thing ? (=a DB, having to use SQL, MySQL) instead of a simple ready-to-use thing (=filesystem)
 
Bob
as I said, you can find FUSE filesystems that can do this on Linux, but that requires more access than shared hosting typically provides
 
So, if you want to actually spare out a considerable amount of size; you'll want to bring all your data into a single file.
 
@Bob true
 
Bob
12:42 PM
i.e. you're going to have to go up to a VPS if you really want to go this route
 
And when you want to do that single file efficiently ... you end up with a database. ;)
 
Bob
@Basj I only suggested sqlite because that comes with compression extensions
most RDBMSes (MySQL, MSSQL, etc.) can also compress
it just saves you the work of doing it manually, and they're optimised for much faster random access than your typical container
 
Database are or can be simple ready-to-use, the way to access it just becomes different.
 
Bob
you can manually pack files into a zip container if you want, but they're not particularly efficient when you get to larger sizes
 
So, to sum up your ideas, what would do, if I don't change hosting (ie I cannot change fs like FUSE, and I cannot have a biiiiig DB because DB is limited in size / access) ?
 
12:45 PM
Yeah, at larger sizes rewrites become a problem; DBs have stuff like incremental writes, journals and similar things that prevent rewrites until they are necessary.
 
Bob
@TomWijsman I believe the zip format also allows individual file updates/adds, but I'm not sure how efficient it is
 
@Basj: Which size are we talking about? Storing some text shouldn't be a problem.
@Bob Possible, depends on the format and implementation.
 
Bob
I'd guess not as efficient as most DB implementations, since it has a different focus
 
do you think jsfiddle.net uses a DB or files to store ?
Given that I cannot change fs like FUSE, and I cannot have a biiiiig DB because DB is limited in size / access, would you use 1) personal method : manual zipping etc. and way to group all revisions in a single ZIP file on each SAVE or 2) SQLite ? or 3) another solution ?
 
Which size are we talking about?
 
Bob
12:49 PM
@Basj What's the expected size of an individual file, and total size?
 
@TomWijsman This project is unknown now (except you and a few others) and it has few features. I will develop new features, and I hope some people will like it : brainstormi.ng/demobrainstorming, for example for doing "brainstormings")
 
Bob
you're going to have to move off shared hosting before it gets extremely big anyway
 
@Bob probably this will never happen ;)
I hope it will happen
@Bob Size of the demo page is ~ 4 Kb..... My personal "brainstorming" file is 50 Kb
 
Putting text like that and save it is a few hundred bytes; if we're talking about a database limit of a few hundred megabytes, you can save about a million of them.
How can it be 4 Kb? :-S
 
If 1000 users * 100 versions * 50 Kb = 5 GB
@TomWijsman because I save all the x, y position, fontsize, etc.
I haven't tried to optimize at all. Of course I could remove "markup" to make it smaller
But as I will do compression, there is no point for premature optimization
 
12:53 PM
@Basj Yeah, with an optimized minimal XML it could go to a few hundred bytes I think.
 
@TomWijsman yes of course, true
 
Just saying, your example would fit a million times in the average database limit, I think.
 
@TomWijsman with compression or no compression ?
 
Or in case of 4 Kb, a million divided by 40 which is still a lot.
No compression... :)
 
what's your computation exactly ?
 
12:56 PM
500 megabytes database / 500 bytes per text file = 1.000.000 text files per database
 
without versioning
for which range is SQLite adapted ?
 
For starting out, you're not going to hit limits any time soon; but once you get into all the -ility words (scal-, real-, ...), you'll want something more serious.
 
Bob
@TomWijsman XML is not optimal for space, at all
you want a binary format
 
@Bob sure but I did not want to go into premature optimization now
About SQLite, I always had this question : why isn't it standard for Wordpress for example ?
 
@Bob It helps a lot with compression; binary is better, but harder to support and adjust especially when using versioning.
 
Bob
1:02 PM
@TomWijsman The actual text content will compress just fine; binary formats (e.g. binary XML) just make the metadata smaller
 
All the people that I know who use a Wordpress (or other CMS) *don't* have millions of visitors per day....... So why bother with a MySQL DB (that's annoying to create a DB on a shared hosting, etc.) if one can just use ........ a file DB ( SQLite)

Moreover for backup, having SQLite is 100 times easier : no need to PhpMyAdmin and "export" , just backup the server files ...
 
@Basj Because ready databases are very often available, therefore making the need of setting something up on your own unnecessary.
 
Bob
@Basj Most people running serious systems have proper automated backups.
pMA is not used with anything serious
And those backups are generally more granular - better for incrementals, and allow partial restores
 
As most people who download Wordpress and try to install it won't have millions of visitors at the beginning... why isn't SQLite enabled by default ?
 
Bob
also, wordpress is a mess
 
1:04 PM
@Bob Yeah, binary XML could allow to easily transform it back into normal XML and therefore allow to adjust it in case the DTD changes.
 
Bob
they don't even support postgres since their codebase is so dependent on mysql-specific features
 
yeah that's a bad thing
 
Bob
@Basj Do you do versioning manually?
Have you considered, say, git?
 
@Bob no no no, I'm not speaking about versioning of my source code, I speak about versions for the final user : bigpicture.bi/demo~13, bigpicture.bi/demo~14, bigpicture.bi/demo~15
 
@Basj That question can be turned around: Setting up a MySQL DB creates a file on disk as well, it's not any more bothersome; if you can use it, why bother with a single file DB?
The answer lies in reliability (automated backups), scalability (reddit frontpage effect) and words like that.
 
1:09 PM
@TomWijsman If you want to move a SQLite DB (for example to copy it locally), you just have to move "a file" !.... If you want to import a MySQL DB locally, you have to export it / import it ... you have to create a local MySQL server, etc.
 
A file is just that, a file; whereas a DB is a bigger concept, which involves a lot more than just being a file.
@Basj Or you use a tool that copies from server to server, YMMV.
 
Bob
@Basj I mean using git to handle versioning of user files, since it already compresses between revisions
 
@TomWijsman I only know a little bit reddit... What do you mean by reddit frontpage effect ?
 
Furthermore, you shouldn't need to copy your user's data to your local MySQL server for any reason.
 
Do you just mean a frontpage with lots of data, or something else @TomWijsman ?
 
Bob
1:11 PM
@Basj see: Slashdot effect
 
@Basj That your site appearing on something like the reddit frontpage can cause a ton of people to visit your site at once.
It used to be named Digg effect, but the name changes every once in a while. :D
 
Bob
Pretty sure Slashdot was the original name
 
@Bob ohhh right thanks
 
And it's not so much about the amount of visitors per day, but rather about of simulatenous visitors at certain times of the days.
 
@TomWijsman yessss! when my tool will be developed enough with good features, I would like to post it on reddit ;) So what's your advice on that ? Forget about SQLite ?
 
1:13 PM
@Bob Ah, you're right; I just read a bit too much of the Digg effect blog posts to ignore that fact.
@Basj It depends on how needy you are of the database.
 
on bigpicture.bi/1898oa, I only need DB when reading on load / writing at the end of the session
 
Bob
Oh yea, another advantage of a proper RDBMS is good concurrency
 
something else : is there an easy way to give a Linux folder a max file size quota ?
example : 10 GB for these data
 
Bob
And caching
 
(if while i'm sleeping a spambot creates revisions every 1 millisecond, I will be flooded)
I know this is difficult to do, but as a first thing, I would like to set 10 GB max size for my DATA folder
 
Bob
1:17 PM
You're going to want rate limiting too
 
@Basj Not sure if that's possible on shared hosting, it also depends on the filesystem.
 
@Bob yes, rate limiting too... How ?
@TomWijsman is thre a standard command on Linux for that, with bash ?
(I have SSH access)
 
Depends on the filesystem. :)
 
how to know which fs again ?
 
But if you really want something simple, put a reasonable limit on the amount of text and elements in a file and a reasonable limit on the amount of files; that should prevent your test environment from filling unreasonably large.
But realistically, as suggested, rate limiting is the way to go; just googling those two words find you a lot of implementation details.
 
1:19 PM
I have xfs or ext3
 
Running fdisk or the dev node, then printing it out should tell you.
@Basj ext3 can't, people work around that with LVM volumes; XFS can, see docs.oracle.com/cd/E37670_01/E37355/html/ol_quoset_xfs.html
 
@Bob thanks
thanks everyone for the advices... I'm gonna read this thing, and I'll be back today maybe ! See you!
 
Ow. Was running around a park today (new experimental, friendlier not entirely completely mandated state/miltary fitness classes)
Wasn't too bad but a bit higher intensity than I'm used to
Unrelatedly
-1
Q: I've uninstalled/deleted bash and it's killed the server

user127379I changed the default shell to zsh. And then proceeded to uninstall bash (this was via yast on SLES). To my horror I start to see everything deleted/uninstalled, apache, mysql . I can't putty into the box any more. Anyway to revive this one? It's a VM on a server with a iDrac connection, but...

 
1:46 PM
Q
 
2:22 PM
1
Q: Heterogeneous (cpu+gpu) System Architecture using C++

RijuI want to develop an image processing system on a computer that has both multi-core CPU and GPU (I don't have such a computer yet, so this is part of the question too - what cpu and gpu models are best suited for such a system). Every minute, the system will capture an image from a video camera. ...

@Bob @JourneymanGeek that Q needs upboats to get him to 20 rep so he can chat :D
You should come to chat -- not sure if you have enough rep to chat, but once you have 20 rep, this is a great chat discussion — allquixotic 47 secs ago
I'd almost say this is a great time to use a sockpuppet (hey, it's SE's fault for imposing a 20 rep requirement for chat!) buuuut... I'm not gonna
 
Bob
@allquixotic eh... technically, I shouldn't be upvoting that...
meh
click
 
but! such an interesting question to talk about here!
meh, he probably is asleep, question asked 10 hours ago... if you figure the middle of his productive workday was 10 hours ago, he's definitely asleep
Riju... eastern Europe? Japan?
we could really game the system and upvote while VTCing :D
good question, but it's off-topic! sorry! /hug /smack
 
Bob
> location USA
assuming he didn't lie in his profile :P
 
oh, ok. so he was asking the question from between midnight EST to 8 PM Alaskan time
and there he is!
 
Nope I wasn't lying abt location :)
 
2:29 PM
hang in there, little buddy! if you don't have 20 rep, we're gonna try to get it for you! OH! you can chat!!!! yay! \o/
 
\o/
 
ok, so, in case you aren't used to SE, a lot of stuff that's off-topic on the Q&A sites is on-topic in chat, and we really like talking about stuff such as your question in here
 
That's great to hear. I am very new to SE>
 
you can edit your messages or reply to others' messages (lets others see specifically which prior message you are referring to) by using the controls when you hover over messages
 
Thanks for getting me here allquixot
 
2:31 PM
e.g. you can edit your typo, or reply to my message by clicking on the "turn-like arrow" on one of my messages
 
What are some good vpn clients out there?
 
@DemCodeLines OpenVPN is really the only one I can think of off the top of my head that isn't buggy, proprietary, and using such a wacky protocol that it's likely to be blocked by most networks
 
@allquixotic Does Windows use "wacky protocol that it's likely to be blocked by most networks?"
 
the nice thing about OpenVPN is that, aside from extremely, extremely sophisticated deep packet inspection, or mandatory MITM, it is nigh impossible to distinguish an OpenVPN connection from, say, an HTTPS website that just happens to have a lot of stuff going on
@DemCodeLines the built-in VPN in Windows? yeah, it does; if I recall correctly, it uses GRE, which is a separate protocol on top of the Internet Protocol (IPv4/IPv6). So instead of, say, TCP/IP or UDP/IP, it uses GRE/IP. Unless your internet connection is completely unfiltered, it'll be blocked.
 
Bob
in The Comms Room, 11 hours ago, by Journeyman Geek
Unless they're doing it cause 9 means something like death in chinese
@JourneymanGeek didn't @allquixotic say it was Tamil?
9 in Mandarin Chinese sounds more like wine
 
2:34 PM
@Bob lol I said that here but I didn't actually know whether 9 means death in Tamil, I was just making fun
 
So Windows 10 finally has Copy-Paste in CMD...
I think that alone gets many sysadmins to upgrade.
 
@Riju regarding your question, it seems like you have hardware and software confusions going on
 
absolutely.
 
@Bob beats me, my chinese is about as good as my shih tzu.
 
a reply mentioned openCL
 
2:35 PM
on the software side, for the purpose of running stuff on the GPU, OpenCL is the "industry standard" that crosses all vendor lines
 
is this proprietory?
aha
 
Bob
@DemCodeLines Windows always had it.
It's always been there.
 
meaning, if you have a modern GPU (manufactured from 2010 onwards), any self-respecting graphics driver for that GPU will have an OpenCL implementation
 
Bob
Personally, I think Ctrl+V is not a good keyboard combo to use for it, since Ctrl+char usually has a special meaning.
 
some had it even earlier than that, but 2010 is a good starting point
 
2:36 PM
 
@Riju another option would be CUDA, but if you write your code targeting CUDA, only Nvidia GPUs will be able to run it
 
that is awesome. since i am going to get a new one anyway, I can make sure that the gpu has openCL implementation
 
Today at lab, write a Visual Basic program that connects to Oracle SQL using some fucked up IDE from 2001 !
 
CUDA can sometimes (depending on the algorithm, the GPU, etc.) be faster than OpenCL, but OpenCL can still be used to achieve really really high speeds on embarrassingly parallel algorithms, regardless of what type of GPU you have, as long as it's recent
 
@Bob I meant keyboard CTRL+V.
 
2:38 PM
as for what GPU to buy, it really depends on how complicated you expect your algorithms to get
 
right @allquixot I already considered that and didn't want to get myself in a corner by using CUDA. openCL definitely sounds promising.
 
How the heck do I get OpenVPN to connect to a specific server? Kinda like the way I get Windows to connect to that server. Is that possible in OpenVPN?
 
...
duh? Else it wouldn't be much use, no?
 
Right, which is why I can't figure out how to do it.
 
@Riju personally, I recommend AMD graphics cards for GPGPU (general-purpose programming on the GPU, as opposed to 3d rendering, which is the other task GPUs are used for), because they have a higher percentage of their resources dedicated to performing floating point operations.
 
2:40 PM
Where to put server url, username and pass.
 
Bob
@DemCodeLines OpenVPN operates on certificates and config files, by default.
 
when you're looking at graphics cards for GPGPU, you have to consider: is your algorithm going to use fixed-point arithmetic (think of int in C/C++), or is it going to use float or double (single and double-precision floating point)?
 
@DemCodeLines: the last time I did it, it was in a config file
 
@Bob Anyway to change that?
 
Bob
@allquixotic I thought AMD cards were better for integer operations.
 
2:41 PM
if you're going to use both, you still need to consider floating point performance; if you're never going to use floating point, then you don't have to consider FP performance
 
There is still a config file, however, it doesn't ask for username and stuff.
 
Bob
@DemCodeLines Why would you need to?
As I said, it operates on certificates.
 
I am just used to using Windows VPN client and would love it if OpenVPN gives me a similar interface.
 
Bob
Far more secure than a username and password.
 
Sure, but how do I connect using certificates then?
 
2:43 PM
@Bob I think I'm getting it wrong -- what I'm thinking of is that AMD cards, such as the GCN 1.0 and 1.1 cards, have a higher fraction of their GFLOPS output for double-precision FP, compared to Nvidia cards
 
@DemCodeLines Generate a pair
 
so if you have (to use a nice round number for example) 100 GFLOPS for single-precision FP, usually there's a fraction, like 1/4, 1/8, 1/32, etc. for your equivalent GFLOPS in double-precision FP.
 
@HackToHell No clue how to. Will try to look it up.
 
all I remember is that the ones on GCN 1.0 and 1.1 cards are way better than Nvidia's GeForce GPUs (but Teslas are better)
 
2:44 PM
O
 
Bob
@DemCodeLines There's plenty of OpenVPN guides.
The general gist is, you create a keystore on the server. Then you use it to generate a client certificate and embed it in an OVPN conf file.
 
I mistyped that, and then, totally squirreled what I wanted to say
 
Bob
Then you load that conf into the OVPN client.
 
What if the server isn't running OpenVPN and is instead a typical workplace VPN server?
 
I remember seeing a script that automated that, can't find it
@DemCodeLines What's a typical workplace VPN ?
 
2:46 PM
@HackToHell idk
But I know the place I am trying to connect to doesn't use OpenVPN.
 
@Riju @Bob The GTX 980 has an advertised 4612 GFLOPS of single-precision FP performance. But its double-precision FP performance is only 144 GFLOPS. Not good. The Radeon R9 290X has an advertised 5632 GFLOPS of single, and 704 GFLOPS of double. So the Nvidia card has 3.12% of its SP perf when doing DP math; the AMD card has 12.5% of its SP perf when doing DP math.
 
I mean, Windows connects to it fine on a secure wifi network. However, on the public network, it just fails and returns all sorts of errors beginning with the number 8.
 
That's what I was thinking of.
 
@allquixotic I think I can certainly make my algos work with single-precision FP
 
Bob
@allquixotic I was thinking of their better integer performance for cryptocurrency mining :P
 
2:48 PM
@Riju okay; if you don't need double-precision FP, then all you need to worry about is (assuming, which may not be a correct assumption, that you don't care about power consumption) -- given your budget, how many single-precision GFLOPS can you afford to purchase?
 
Bob
@Riju Depending on how much you can spend, a workstation card might actually be more appropriate.
 
so we really need to know your budget before we can go any further, because the range of cards goes from old and low-end ones, to just-released extremely high-end ones
 
@allquixotic What's the name of the unit that does math in GPUs ?
 
and if you need supercharged performance and you have a hideously huge budget, as in several thousand USD, you could even consider multi-GPU
 
There are specialized units for that no ?
 
2:49 PM
@HackToHell all GPUs do is math, technically :P
 
Uhm yeah, but I remember reading somewhere that there are specialized math units say ray tracing and for vector stuff.
 
the GPU's "cores" (called streaming processors or shader cores, depending on who you ask) are the generally-programmable units that, basically, act almost as if they were a CPU core, except they're each pretty weak, and there are between several hundred to many thousands of them on a single GPU chip
the general purpose cores started out running OpenGL 2.1 / DirectX 9 "fragment shaders", which basically allow you to do arbitrary vector math on the GPU, but now you can basically run arbitrary C code on them
(OpenCL kernels look a little bit like C)
 
@HackToHell: most raytracers are a shitload of regular cores for some reason
 
the special-purpose math units that are hard-wired to do specific things are usually reserved for stuff like 3d rendering
the ROPs and texture units are very specialized and only really work for graphics rendering, not GPGPU
then there's the video decode / encode units which are separate from the general-purpose cores and similarly are either electrically or firmware-designed to only perform specific operations
 
Aha !
480 Stream Processing Units
24 Texture Units
32 Z/Stencil ROP Units
8 Color ROP Units
From the AMD site
 
2:53 PM
@Bob @allquixotic Wow...no 1 gpu will be good for now :) My budget is around 1k USD for the gpu. What is a workstation gpu?
 
@Riju there are different "classes" of GPUs
 
@JourneymanGeek Oh so it's just normal old cored
 
desktop-grade GPUs are typically optimized for 3d rendering (playing today's modern PC games), but they are still capable for GPGPU
 
I supposed stream processing units are the cores @allq mentioned
 
then there are dedicated "GPGPU cards" which have no 3d rendering capabilities at all, but they have an extremely high performance on both single-precision and double-precision floating point (OpenCL or CUDA)
workstation GPUs are similar to desktop-grade GPUs, except they're more expensive, and they have a better GFLOPS rating for double-precision float
they could also have special features like genlock and synclock, which are used for real-time video production
 
2:55 PM
!! wiki Render output Unit
 
The render output unit, often abbreviated as "ROP", and sometimes called (perhaps more properly) raster operations pipeline, is one of the final steps in the rendering process of modern 3D accelerator boards. The pixel pipelines take pixel and texel information and process it, via specific matrix and vector operations, into a final pixel or depth value. The ROPs perform the transactions between the relevant buffers in the local memory – this includes writing or reading values, as well as blending them together. Historically the number of ROPs, TMUs, and pixel shaders have been equal. However, as...
 
^^ bloody confusing
 
so, ignoring lower-end stuff like APUs, IGPs and laptop GPUs, we have: desktop-grade, workstation-grade, and server-grade GPUs
 
@HackToHell ^^
 
desktop: the most inexpensive; well-rounded; good but not quite as awesome as the others at GPGPU. best value for your money
 
2:56 PM
@JourneymanGeek Awesome !
 
@allquixotic I will go for a dedicated GPGPU then. I don't need 3D rendering but better performance with single-precision and double-precision floating point would be great
 
workstation: very expensive; also does video output like a desktop GPU; enterprise-grade "video production professional" features; good GPGPU performance
 
Wikipedia should link to that
 
server: hideously expensive; no video output; highest GPGPU performance
 
@HackToHell: there's newer variations but that's the one I remember
 
2:57 PM
@Riju OK... I'm not sure if you'll be able to afford a Tesla card or AMD's equivalent within $1k USD, but you could afford a very high-end desktop GPU with that price point
generally, with the price point you're at, I would not recommend buying a previous-generation product
GPUs increase in performance between about 10 to 50% per year (depending on whether they're moving to a new fabrication line or just iterating the GPU design)
 
nope. nothing hideously expensive. Wow...your answers cleared a lot of cloud. Is there a way I can link your answers back to the original question? That would be good for my future reference as well as someone else who find him/herself in my position.
 
@JourneymanGeek I suppose it's enough to understand the basics
Or you could get a FPGA if you are mining bitcoins ;p
 
Bob
@Riju You can create a bookmark in chat.
On the right hand side, click the little room link just under the chat description.
 

« first day (1517 days earlier)      last day (3485 days later) »