I just checked the specs Samsung gives for the 960 PRO and it's supposed to top out around 3500MB/sec, so I'm questioning that result, now. It was Samsung's own speed test (Samsung Magician), so maybe it's a bit biased! I'll check it again with CrystalDiskMark, but it's also possible that the new firmware enabled faster speeds and the specsheet wasn't updated to reflect that.
To be honest, I was always running up against 16GB when that's all I had; may last workstation had 32GB and never peaked over 12. I've gotten this one up to 17GB once while I was looking but it usually sits around 14, the 64GB was really more of a "hey, I can, why not?" kind of thing. 32 made sense because 16 wasn't enough, except that the very same workload never broke 16GB once I upgraded... still the biggest WTF I'm dealing with at the moment. Not a complaint, just a point of confusion, I guess.
That's also not entirely true, as I've filled the RAM on every machine I've ever built or bought running Prime95 to stress test. But in normal use, yeah, 64GB (or even 32) has proven to be severe overkill. I'm sure one day I'll be glad I have it all, though.
That's just read. That drive hasn't seen a TRIM in a while, so the writes were significantly slower than they should have been. I didn't make note of that number because it wasn't impressive, not much faster than the 950 RAID; it's a fair bit faster after a TRIM. Still not 4GB/sec, but faster.
With 64GB of RAM, though, I've got plenty of write buffer; it's read performance I'm more concerned with. I have more than enough backup power to flush from RAM to disk in a power failure.
It's refreshing to see someone else actually considering how things work together, rather than just "oh look, new shiny and fast". I was starting to think I was the only one who still gave a damn.
I have a similar setup, 2x 500GB Samsung 950's in a RAID0, and I get about that from that setup, as well. I updated my 960's firmware last night and ran a speed test on it afterward; it's currently getting 4GB/sec. That's as fast as some DDR2 RAM that was commonly used not long ago.
I would love NVMe RAID. I still find myself disk-bound at times when compiling, and I'm too lazy to set up a RAMdisk.
And Intel specifically had issues with task/process switching. Based on your statements above, it appears they have finally addressed at least a subset of those.
Actually, I went AMD this time around. No regrets.
having access to NVMe will speed things up significantly
And the award for Understatement of the Century goes to...
The landscape is getting interesting again... finally!
only 16?;) I was looking at a 24 threaded system back in 2013
Come on, you know full well that, instruction set, IPC and latency factor into that; In any case, anything more powerful than Apple is willing to sell me is still more powerful than Apple is willing to sell me. There are supercomputers built in the early 2000's that still smash the performance of my paltry desktop workstation, but it's still faster than anything Apple puts out.
Since I know you already know this, the following is intended for anyone reading along.
We've seen considerable improvements in instructions per clock and had a few new instructions added to the instruction set since 2013, which makes each of my 8 cores faster than each of the 12 you had in 2013. On top of that, a highly threaded application will benefit from 50% fewer locks waiting for another core to finish its work or release a resource. If we assume that one of my 2017 cores can process 1.5x the workload of one of your 2013 cores (which benchmarks seem to bear out, using an Xeon E5-2667 as an example, as that is the most likely CPU to have found in a 24-thread workstation in 2013), in aggregate we would expect 16 of my cores to match 24 of your. Factoring in that that E5-2667 is a 6 core CPU designed for dual-CPU installations (there was no 12 core 24 thread Xeon until the end of 2013, nor were there any 3 core parts which may have been used in a quad-CPU configuration), you now have CPU interconnect latency to deal with, on top of 50% more core-to-core latency. All else being equal, 16 cores is better than 24 when the aggregate number of instructions either group of cores can process is the same.
Add to that, those two E5-2667's, in 2013, would have cost more than my entire build in 2017. Also, since I'm already at the point where my GPUs are the bottleneck, still-faster CPUs wouldn't benefit me all that much, I'd just be spending more; that's why I only briefly considered Threadripper. Currently, with both GPUs running full-tilt, my little side project leaves a full core (2 threads) free, which is how I'm able to still multi-task on the machine while jobs process.
There's this concept in business called diminishing returns. It involves spending enough to get the job done effectively, and as little above and beyond that as possible. My current build puts me right in that sweet spot.
Huh, funny, I didn't know all it took to make something true was two people making the claim with no evidence. Have you finally lost your mind? Are you 5? Because you're arguing like you're either 5 or have lost your mind.
Where is your evidence, sir? You make a claim with no evidence to back it up, then point to someone else's (also baseless) claim as evidence? That's not how this works, sorry. Point out the lie or fuck off.
If anybody with an IQ above room temperature even reads this far into this conversation, there's no way you're fooling them. If you think you are, your IQ matches theirs.
But, just assuming your right for a moment, do you really want to be seen posting on a site that lets a liar maintain Excellent karma?
It makes me wonder if you need the "system more powerful" or need to run a specific piece of software.
I do need to run that specific piece (suite, actually) of software in order to effectively work with some of my clients, as it is what they use and money talks.
As for performance, well... I perform certain tasks that are greatly sped up by the availability of 16 high-performance threads and a pair of high-end GPUs. No reason to get into specifics, but it means I can do two jobs in the same day, rather than one; or that I can carry on my primary role as a developer while my side project is processing still a a considerable bit faster than it would on the best Mac I could buy, without noticeable degradation in the performance of my IDE.
In short, my use case is atypical and Apple does not cater to atypical use cases. They'll likely never again sell a machine that will suit my needs. The old aluminum-cased Mac Pro would have been perfect back when it was current, but I didn't have a need back then and it's well behind the curve today. Unless they bring that back in all its glory and with up-to-date hardware, Apple will cease to be an option for my desktop. They'd still be an option for my laptop if not for the fact that my needs above and beyond an IDE include the ability to queue and run jobs on my desktop workstation and I can do all of that just fine from a Chromebook. The Pixel is a very nice machine and the 3:2 screen ratio lends itself very well to running an IDE; with Crouton, I run my full IDE in an Ubuntu chroot. It's glorious, really.
And if it gets stolen, it was cheap (gen 1 Pixels can be had refurb for under $400) and everything's encrypted locally and committed to a remote Git repo, so no worries of data loss. I can drop another $350 to replace it if need-be.
Which leaves my retina MacBook Pro docked on my wall (yes, I mounted a dock to my wall and the MacBook Pro hangs there), ready and waiting for remote access to test my web development projects in Safari as needed. It really hasn't been used for much else in almost 2 years.
Computers are tools and each of us is best served using the best tools for the work we do. I use a mix of Windows, OS X (still on El Capitan), iOS, Android, Ubuntu, and Chrome OS; it's what works for me as I perform a variety of different tasks best served by a variety of tools. In short, I don't pound screws with a hammer, but I do keep a hammer around for when I encounter a nail. I prefer the power drill for screws, but I respect the manual screwdriver as sometimes being the right tool, as well.
Anyone who doesn't get that isn't doing themselves any favors.
No, sadly, some industry standard software used by a handful of my clients dictates that I run Windows or macOS, while some other work I do dictates that I have a system more powerful than anything Apple is willing to sell me. That pretty much limits me to Windows 10 since older versions of Windows no longer receive updates on newer CPUs.
It's 2017, and reward that goes to someone other than me is a punishment. Or something like that, I don't know, that's just what I hear whenever a millennial is speaking.
Oh, no, they've made it worse. There are now several different print dialogs in Windows (ignoring apps which provide their own) and how easy or difficult it is to specify your print settings is now a matter of luck: does your app use the old dialog that actually allows this, or the new one that prints everything in shitmode? Photos, the Windows 10 default image viewer, uses shitmode.
</reply>
Microsoft, this is why nobody uses the software you ship with Windows. You make stupid defaults in the simplest of applications, make them difficult or impossible to change (no quality settings and poor print quality from Photos? SERIOUSLY?! And you want people to use it?) and you're amazed nobody wants to trust Edge? And that's completely ignoring your history with IE.
Seriously, one factor I consider when deciding which software to use is "have I ever seen Windows 10 install it without being asked?" Pity, I was finally going to give Minecraft a try this weekend, but I found it automatically installed with the last set of updates and uninstalled it instead.
Microsoft, seriously, if you feel that you must shove your software down my throat, that tells me you don't trust your software to sell itself on merit. If you don't trust your software, why should I.
One example is a single-page newsletter. Printed. On paper. That's a flat design and it works quite well; the exact same layout displayed full-screen on a monitor would be an ideal use case for flat design because, well, it's emulating just that. Flat design plus the need for navigation is, really, where the problems start; and that's what we see everywhere.
Really, I think it's laziness. Probably, someone at Microsoft was tasked with re-implementing the Print Preview dialog and thought "well, since the document display is flat, it would be easier to make the whole dialog flat", which spilled into "I've created this flat interface, now it would be easier to use it everywhere so I don't have to learn two different form libraries."
Okay, that's probably not a true story (it might be, I don't know), but that's the mentality behind it.
Yes, it's the right interface to use for certain things. Lazy developers made it the interface they use for everything and screwed up interfaces everywhere as a result.
Mind you, I wasn't arguing for flat design, I deplore it. You asked what a benefit was, I provided one; and if done correctly, that's the benefit. Supposedly.
What we were sold is "skeuomorphism takes up too much space, we can make design elements smaller by making them flat" and, when an interface is implemented in that manner, assuming you can truly put all of the data in one screen and negate the necessity for navigation, well, that's just wonderful.
The ability to fit more data into a smaller space. Literally the antithesis of usability unless your use case is "see all of the data at once" which, when we're talking about navigation, is certainly not the case; if you can see all of the data, there's nowhere to navigate to.
Care to point out one of these supposed lies and, perhaps, expose the truth?
Can't?
Funny. If I were the liar you and a handful of others here who can't form a cogent argument keep claiming I am, it should be trivial to prove it.
Of course you can't prove me a liar; I'd have to be one first.
And, since this account is closely tied to my actual identity, I would recommend treading lightly, you're not nearly as anonymous as you think you are and defamation suits tend not to go well when you're the one attempting to damage the reputation of another.
Slashdot Mirror
Right, it's not, it's also your state of mind. Which lies have I told here? Point them out.
I'll ask again, which lies?
Hahaha, well at least UnknowingFool is entertaining. Usually, at least; he's gotten a bit old this time around.
I just checked the specs Samsung gives for the 960 PRO and it's supposed to top out around 3500MB/sec, so I'm questioning that result, now. It was Samsung's own speed test (Samsung Magician), so maybe it's a bit biased! I'll check it again with CrystalDiskMark, but it's also possible that the new firmware enabled faster speeds and the specsheet wasn't updated to reflect that.
To be honest, I was always running up against 16GB when that's all I had; may last workstation had 32GB and never peaked over 12. I've gotten this one up to 17GB once while I was looking but it usually sits around 14, the 64GB was really more of a "hey, I can, why not?" kind of thing. 32 made sense because 16 wasn't enough, except that the very same workload never broke 16GB once I upgraded... still the biggest WTF I'm dealing with at the moment. Not a complaint, just a point of confusion, I guess.
That's also not entirely true, as I've filled the RAM on every machine I've ever built or bought running Prime95 to stress test. But in normal use, yeah, 64GB (or even 32) has proven to be severe overkill. I'm sure one day I'll be glad I have it all, though.
That's just read. That drive hasn't seen a TRIM in a while, so the writes were significantly slower than they should have been. I didn't make note of that number because it wasn't impressive, not much faster than the 950 RAID; it's a fair bit faster after a TRIM. Still not 4GB/sec, but faster.
With 64GB of RAM, though, I've got plenty of write buffer; it's read performance I'm more concerned with. I have more than enough backup power to flush from RAM to disk in a power failure.
It's refreshing to see someone else actually considering how things work together, rather than just "oh look, new shiny and fast". I was starting to think I was the only one who still gave a damn.
I have a similar setup, 2x 500GB Samsung 950's in a RAID0, and I get about that from that setup, as well. I updated my 960's firmware last night and ran a speed test on it afterward; it's currently getting 4GB/sec. That's as fast as some DDR2 RAM that was commonly used not long ago.
I would love NVMe RAID. I still find myself disk-bound at times when compiling, and I'm too lazy to set up a RAMdisk.
I'd just be less confused when I got email notification of your reply, I suppose.
And Intel specifically had issues with task/process switching. Based on your statements above, it appears they have finally addressed at least a subset of those.
Actually, I went AMD this time around. No regrets.
having access to NVMe will speed things up significantly
And the award for Understatement of the Century goes to...
The landscape is getting interesting again... finally!
Is this really an outage, or do a bunch of people just owe them 12 cents?
I really need to start looking at who I'm replying to before submitting comments. That's twice this week I've replied to you without realizing it! ;)
only 16? ;) I was looking at a 24 threaded system back in 2013
Come on, you know full well that, instruction set, IPC and latency factor into that; In any case, anything more powerful than Apple is willing to sell me is still more powerful than Apple is willing to sell me. There are supercomputers built in the early 2000's that still smash the performance of my paltry desktop workstation, but it's still faster than anything Apple puts out.
Since I know you already know this, the following is intended for anyone reading along.
We've seen considerable improvements in instructions per clock and had a few new instructions added to the instruction set since 2013, which makes each of my 8 cores faster than each of the 12 you had in 2013. On top of that, a highly threaded application will benefit from 50% fewer locks waiting for another core to finish its work or release a resource. If we assume that one of my 2017 cores can process 1.5x the workload of one of your 2013 cores (which benchmarks seem to bear out, using an Xeon E5-2667 as an example, as that is the most likely CPU to have found in a 24-thread workstation in 2013), in aggregate we would expect 16 of my cores to match 24 of your. Factoring in that that E5-2667 is a 6 core CPU designed for dual-CPU installations (there was no 12 core 24 thread Xeon until the end of 2013, nor were there any 3 core parts which may have been used in a quad-CPU configuration), you now have CPU interconnect latency to deal with, on top of 50% more core-to-core latency. All else being equal, 16 cores is better than 24 when the aggregate number of instructions either group of cores can process is the same.
Add to that, those two E5-2667's, in 2013, would have cost more than my entire build in 2017. Also, since I'm already at the point where my GPUs are the bottleneck, still-faster CPUs wouldn't benefit me all that much, I'd just be spending more; that's why I only briefly considered Threadripper. Currently, with both GPUs running full-tilt, my little side project leaves a full core (2 threads) free, which is how I'm able to still multi-task on the machine while jobs process.
There's this concept in business called diminishing returns. It involves spending enough to get the job done effectively, and as little above and beyond that as possible. My current build puts me right in that sweet spot.
Indeed. That is something you say. A lot. At least you're entertaining; but, really, this is getting old now. Point out my lie or STFU.
Also, notice which word I didn't use in this post.
Indeed, that's exactly something you'd say. Notice which word I didn't use in this post?
Huh, funny, I didn't know all it took to make something true was two people making the claim with no evidence. Have you finally lost your mind? Are you 5? Because you're arguing like you're either 5 or have lost your mind.
Where is your evidence, sir? You make a claim with no evidence to back it up, then point to someone else's (also baseless) claim as evidence? That's not how this works, sorry. Point out the lie or fuck off.
If anybody with an IQ above room temperature even reads this far into this conversation, there's no way you're fooling them. If you think you are, your IQ matches theirs.
But, just assuming your right for a moment, do you really want to be seen posting on a site that lets a liar maintain Excellent karma?
Oh, wow, two whole people called me a liar! Neither of them can cite a lie I've told, maybe they're full of shit?
It makes me wonder if you need the "system more powerful" or need to run a specific piece of software.
I do need to run that specific piece (suite, actually) of software in order to effectively work with some of my clients, as it is what they use and money talks.
As for performance, well... I perform certain tasks that are greatly sped up by the availability of 16 high-performance threads and a pair of high-end GPUs. No reason to get into specifics, but it means I can do two jobs in the same day, rather than one; or that I can carry on my primary role as a developer while my side project is processing still a a considerable bit faster than it would on the best Mac I could buy, without noticeable degradation in the performance of my IDE.
In short, my use case is atypical and Apple does not cater to atypical use cases. They'll likely never again sell a machine that will suit my needs. The old aluminum-cased Mac Pro would have been perfect back when it was current, but I didn't have a need back then and it's well behind the curve today. Unless they bring that back in all its glory and with up-to-date hardware, Apple will cease to be an option for my desktop. They'd still be an option for my laptop if not for the fact that my needs above and beyond an IDE include the ability to queue and run jobs on my desktop workstation and I can do all of that just fine from a Chromebook. The Pixel is a very nice machine and the 3:2 screen ratio lends itself very well to running an IDE; with Crouton, I run my full IDE in an Ubuntu chroot. It's glorious, really.
And if it gets stolen, it was cheap (gen 1 Pixels can be had refurb for under $400) and everything's encrypted locally and committed to a remote Git repo, so no worries of data loss. I can drop another $350 to replace it if need-be.
Which leaves my retina MacBook Pro docked on my wall (yes, I mounted a dock to my wall and the MacBook Pro hangs there), ready and waiting for remote access to test my web development projects in Safari as needed. It really hasn't been used for much else in almost 2 years.
Computers are tools and each of us is best served using the best tools for the work we do. I use a mix of Windows, OS X (still on El Capitan), iOS, Android, Ubuntu, and Chrome OS; it's what works for me as I perform a variety of different tasks best served by a variety of tools. In short, I don't pound screws with a hammer, but I do keep a hammer around for when I encounter a nail. I prefer the power drill for screws, but I respect the manual screwdriver as sometimes being the right tool, as well.
Anyone who doesn't get that isn't doing themselves any favors.
No, sadly, some industry standard software used by a handful of my clients dictates that I run Windows or macOS, while some other work I do dictates that I have a system more powerful than anything Apple is willing to sell me. That pretty much limits me to Windows 10 since older versions of Windows no longer receive updates on newer CPUs.
It's 2017, and reward that goes to someone other than me is a punishment. Or something like that, I don't know, that's just what I hear whenever a millennial is speaking.
And ALL devices that run Qualcomm's ARM SoCs run Android.
Which routers run Android? I was gonna ask which TVs, but there are actually some that do...
Oh, no, they've made it worse. There are now several different print dialogs in Windows (ignoring apps which provide their own) and how easy or difficult it is to specify your print settings is now a matter of luck: does your app use the old dialog that actually allows this, or the new one that prints everything in shitmode? Photos, the Windows 10 default image viewer, uses shitmode.
</reply>
Microsoft, this is why nobody uses the software you ship with Windows. You make stupid defaults in the simplest of applications, make them difficult or impossible to change (no quality settings and poor print quality from Photos? SERIOUSLY?! And you want people to use it?) and you're amazed nobody wants to trust Edge? And that's completely ignoring your history with IE.
Seriously, one factor I consider when deciding which software to use is "have I ever seen Windows 10 install it without being asked?" Pity, I was finally going to give Minecraft a try this weekend, but I found it automatically installed with the last set of updates and uninstalled it instead.
Microsoft, seriously, if you feel that you must shove your software down my throat, that tells me you don't trust your software to sell itself on merit. If you don't trust your software, why should I.
One example is a single-page newsletter. Printed. On paper. That's a flat design and it works quite well; the exact same layout displayed full-screen on a monitor would be an ideal use case for flat design because, well, it's emulating just that. Flat design plus the need for navigation is, really, where the problems start; and that's what we see everywhere.
Really, I think it's laziness. Probably, someone at Microsoft was tasked with re-implementing the Print Preview dialog and thought "well, since the document display is flat, it would be easier to make the whole dialog flat", which spilled into "I've created this flat interface, now it would be easier to use it everywhere so I don't have to learn two different form libraries."
Okay, that's probably not a true story (it might be, I don't know), but that's the mentality behind it.
Yes, it's the right interface to use for certain things. Lazy developers made it the interface they use for everything and screwed up interfaces everywhere as a result.
Mind you, I wasn't arguing for flat design, I deplore it. You asked what a benefit was, I provided one; and if done correctly, that's the benefit. Supposedly.
What we were sold is "skeuomorphism takes up too much space, we can make design elements smaller by making them flat" and, when an interface is implemented in that manner, assuming you can truly put all of the data in one screen and negate the necessity for navigation, well, that's just wonderful.
What we got was something entirely different.
The ability to fit more data into a smaller space. Literally the antithesis of usability unless your use case is "see all of the data at once" which, when we're talking about navigation, is certainly not the case; if you can see all of the data, there's nowhere to navigate to.
Care to point out one of these supposed lies and, perhaps, expose the truth?
Can't?
Funny. If I were the liar you and a handful of others here who can't form a cogent argument keep claiming I am, it should be trivial to prove it.
Of course you can't prove me a liar; I'd have to be one first.
And, since this account is closely tied to my actual identity, I would recommend treading lightly, you're not nearly as anonymous as you think you are and defamation suits tend not to go well when you're the one attempting to damage the reputation of another.
Care to point out one of these supposed lies and, perhaps, expose the truth?
Can't?
Funny. If I were the liar you and a handful of others here who can't form a cogent argument keep claiming I am, it should be trivial to prove it.
Of course you can't prove me a liar; I'd have to be one first.