so the used/refurbished market in sticks gets it's pipeline full leading to nice discounts so I can buy 'em up cheap! I've got four Xeon's looking to have a hot date with in-memory databases.
Samsung's started volume production of 128GB DDR4 RDIMMS. You read that right – a single registered dual inline memory module packing 36 individual 4GB chunks of 3D TSV DDR4 DRAM, for a total of 128GB, can now be yours. The Register is aware of servers with 96 DIMM slots, which means … WOAH! … 12.2 terabytes of RAM in a …
Microsoft bashing in an article about new RAM...
The sad thing is that this being the Register I am not surprised.
The sadder thing is that it is misplaced bashing: In fact, every Windows edition after Vista has been faster on the same hardware -- including amount of system RAM -- than the one preceding it. I am not sure the trend holds with Windows 10 Build 10586 as it's so new and I haven't installed it on enough machines to draw any meaningful conclusions, but it does hold for everything else.
Back on topic: Good $deity, I want some of those modules! Which body parts do I have to sell to afford them...?
Trend sort of holds true with Windows 10 10586. One of the things it does is perform well with less RAM because it has vastly improved the logic for paging stuff in and out (using compression) - http://blogs.windows.com/windowsexperience/2015/08/18/announcing-windows-10-insider-preview-build-10525/
I'd imagine there's been some optimisations for general memory usage as well
If you played with using Windows 2000 Advanced Server and/or Windows 2003 Enterprise, you'd find that the memory manager was quite different in a more than pleasant way. It's nice to see that Microsoft is extending this kind of thinking lower down the stacks rather than reserving it to Enterprisey stuff.
"The sadder thing is that it is misplaced bashing: In fact, every Windows edition after Vista has been faster on the same hardware -- including amount of system RAM -- than the one preceding it"
Complete and utter lie, every version of windows was always slower than the previous version on the same specification with the possible exception of Vista->windows 7 upgrade.
The only way the lie could be even remotely true would be if the shrills were saying that newer operating systems run better on newer hardware when the software is optimised for the new hardware. Yeah, well DUH! by that token the lastest version (name your linux distribution) is the fastest ever but will still run productively on older hardware, unlike windows
Wrong and willful lies, if you had actually installed every version of windows on a minimum of 100 machines per version then like me you would know what you are talking about and yes I held MS certification on each version since 95a (the version on floppies)
Even if you are not just another MS shill then you are clearly talking out your behind, no one who wasn't being paid to lie, who did actually upgraded existing hardware of the time that was well within the OS specification would be unaware of the facts.
Newer Windows versions always ran slower on the same hardware than the previous windows version with the exception of vista->windows 7 and even after upgrading RAM and CPU, dropping back would give still better performance. Again this is something I have actually tested rather than just seen on some website.
Almost totally wrong.
XP. When we (at Nokia) had this thrust up our collective jaksie's it was pretty good. Yikes, they even gave us 512MB to increase productivity. From WIn-NT it was a miracle!
Then, Service Pack 3 came along.
We at Nokia increased the carbon footprint on the planet because we didn't want to wait for our machines to start every morning, we simply left them on overnight.
"Complete and utter lie, every version of windows was always slower than the previous version on the same specification with the possible exception of Vista->windows 7 upgrade."
Errm... this is not correct. I have on the desk next to the Mac I'm typing this on an Asus A53SD laptop, currently fitted with 8 GB RAM and 1TB drive. The drive is partitioned into a Win 7 Ultimate and a Win 10 Pro (formerly Win 8.1 Pro) bootable partitions. It has a quad-core i7 running at 2.2 GHz. It quite simply runs Win 10 (and used to run Win 8.1) faster than Win 7.
Let me repeat that: Win 8.1 used to, and Win 10 still does, run faster, on exactly the same hardware, as Win 7. I have verified this experimentally by actually checking the times to boot the system and to load and use assorted software, from Microsoft and others, running on _exactly_ the same hardware. The system shipped with Win 7 Home Premium; first thing was that I updated it to Win 7 Ultimate. I put Win 8 on the other partition. Win 8 stank, so I usually used Win 7. When the Win 8.1 upgrade came out, I installed that, not expecting much improvement. However, once it became clear that I could banish the damn tiles to Sinofsky's deepest hell, the simple fact that Win 8.1 Pro was faster at most things won me over. I updated to Win 10 Pro, and cleaned up the spyware... and Win 10 is actually faster, on the same hardware, than Win 8.1. Which means that it's faster, on the same hardware, than Win 7, and not by a small margin, either.
And this particular machine is four years old, hardly newer hardware.
Your statement runs directly contrary to my actual, day-in, day-out, experience. There are a lot of things to dislike about Windows in general and Win 10 un particular. it is, however, simply untrue to state that it is slower, on the same hardware, than Win 7. (At least, as long as the hardware is adequate. If you try to run Win 10 on a knackered old Pentium 4 system with 2 GB RAM and 250 GB drive, you _will_ be sorry. And, yes, I tried that just to see what would happen. It wasn't pretty. I put Win XP back on that old machine and it behaves much better. The thing is, Win 7 and Win 8.1 didn't work very well on that machine, either, and for the same reason that Win 10 didn't like it: the hardware dates from 2005 and is well past its sell-by date.)
@ James O'shea, what were the release dates of windows 7 and the Intel i7?
yes thats right, windows 7 came out first so i7 hardly existing hardware
So, I guess you missed the part in the post you are refuting about windows being optimised for new hardware? and in your case hardware that wasn't even on sale when windows came out.
windows7 was optimised for 2 cores where as windows 10 is optimised for more than 2.
Your laptop came out long after windows7 was released and would have been cutting edge for the time, i.e. not something that the majority of windows 7 users would have been using. So if you were saying you had a core2 and it ran quicker then that would mean something, saying you have hardware that the OS has been specifically optimised for is so much less.
Sounds like you need to take your PC back to BestBuy or JiffieLube for a tune-up.
There's always a reason for these "seemingly no reason" pauses. If you are unable to use available tools to find out what's going on, try getting your 12yo nephew to help. I'd also suspect some additional software installed that is doing something in the background (phoning home?).
> The sadder thing is that it is misplaced bashing: In fact, every Windows edition after Vista has been faster on the same hardware.
I once had a Laptop with Vista on it.
I imagine ANY Windows edition before or after Vista (excluding ME that is) has been faster on the same hardware. Even a rabid Microsoftie aught to realise by now that he or she shouldn't mention Vista in this day and age.
In the dim and distant I put Win 95 on some old XP desk tops and they worked in a limited way. No internet drivers IIRC prevented further endearment. Anyone remember why I eventually upgraded?
Memory Lane is one step too far for me, these days. Did they have Office in those days, I vaguely recall being stranded with Notepad or was there WordPro?
Well from what I can see, memory limitations are not baked into CPUs but into motherboard chipsets.
And there are quite a few Skylake boards that are apparently ready for 64GB.
Besides, it is not logical that a 64-bit processor able to address 1GB of RAM would be limited in any way to not be able to address 128GB.
Memory DOES come in power-of-2 sizes, always has been, always will be. Width, as seen from the bus, will be a multiple of 8, or 32, or 64, and address size will also be a multiple of 2 because of the way address decoding and address mapping between memory banks works.
None of this disk drive sizes marketroid malarkey.
@Sorry T.H.I.A.S - I want to simultaneously upvote you and downvote SI now! I'm old enough to have been raised on the full binary terms and only recently discovered that SI decided to deem that they;re passe. Whilst generally in favour of SI, that's one case where I feel they've made the wrong decision.
Have your upvote - I'm off to find the SI HQ and see if I can give 'em what for with the pointy end of my brolly...
I'd love to be able to afford a pair of these but I don't think that's going to happen any time soon :-(
I was wondering about pricing though and as a quick googling turned nothing up I though it was about time to extrapolate... A 64GB DDR4 can be had for about £600 so I'd guess that after the initial super premium pricing is over the 128GB module will be around £1500 / stick. For your 96 stick server you are talking a whopping £144,000 for the memory alone, best double check the halon system is working.
Can address over that. 64-bit chips can do 64EiB - Exbibytes - so 1024 PiB / 1048576TiB. Theory and practical the address limit allows. Physically possible? That's another matter at the moment.
RAM can run pretty hot, servers with ASHRAE 3 compliance for instance will happily sit and run with all day with an intake of 45 degrees celcius.
Got to think, it's not just datacentres that servers sit in, but some quite enclosed / passively cooled spaces for telecoms industries or such.
Though yeah, if you thought standard DIMMs ran hot you should see the monster cooling reqs for flashDIMMs. Maybe even these 128GiB sticks will have some impressive heatsinks around them like some LRDIMMs do.
"64-bit chips can do 64EiB"
Not actually true as far as I know. For example, see https://en.wikipedia.org/wiki/X86-64#Physical_address_space_details for an assertion that it was limited to 48-bits back in 2010. The preceeding paragraphs (on the virtual address space) also assert that the 48-bit limit is baked into the AMD64 spec. 48 bits is 256TB, so we are now within two orders of magnitude of being able to put a moby in an x64 machine.
Yeah, and we were limited to 640k in the 286-486 days, then came Extended Memory pushing it up to 1 MB, then 64 MB, and then again to 4 GB. I'm sure that once we start seeing commodity servers ship with more than 16 TB, AMD and Intel will just put out a processor extension to add in some more address bits before it really becomes an issue.
It wasn't too long ago that everyone was worried about disks being over 2 TB, now we have modern OSes booting from single-partition 8 TB disks without needing drivers or any modifications to the kernel.
Yeah, and we were limited to 640k in the 286-486 days, then came Extended Memory pushing it up to 1 MB
The 286 was already capable of 16MB, but DOS, running in 8086 real mode, could only deal with 1MB, of which the area from 640k to 1M was used for the BIOS, controllers and a window into expanded memory if you had that. Everything over 1M was designated as extended memory.
Several other OSes of that time were capable of using all available memory without having to move back and forth between real and protected mode.
And of course there were also other architectures that weren't saddled with icky design decisions for the sake of backwards compatibility that few people actually used.
This post has been deleted by its author
no need to follow AMD and support 256TB, Intel knows better and can make it cheaper.
Early Intel 64 implementations (Prescott and Cedar Mill) only allowed access to 64 GB of physical memory while original AMD64 implementations allowed access to 1 TB of physical memory. Recent AMD64 implementations provide 256 TB of physical address space (and AMD plans an expansion to 4 PB), while some Intel 64 implementations could address up to 64 TB. Physical memory capacities of this size are appropriate for large-scale applications (such as large databases), and high-performance computing (centrally oriented applications and scientific computing).
Give piss-poor programmers more space to write crap code?
They might even revert to university-learned Java. God help us.
I remember using a computer (PDP-11) with 128 KILOBYTES of memory. Considered a luxury then. Memory cost thousands - about the asking price of a decent 3-bedroom family house in those days.
Considering, therefore how much a 'word' of code cost, you were FUC*KING cautious!
Just think about all the VMs you can host on a big with that much memory? Most of them won't need the CPU... or even just the huge readonly DB replicas you could host on there for your blog? Have a master box to handle updates, and a ton of these facing the customers or running reports. Sweet!
Who cares how crappy my SQL queries are? What do you mean that doing 15 joins just to return the ID of the boxtop that matches the random query across your data wharehouse is stupid? Toss RAM at it and it's now fast again!
At least until the next upgrade of the software where the vendor has written even crappier queries...
If you're planning on running Windows, you might want to visit https://msdn.microsoft.com/en-us/library/windows/desktop/aa366778%28v=vs.85%29.aspx This lists the Official Microsoft Limits to How Much RAM You Can Stuff Into a Machine for all desktop and server OSes going back to XP. Sorry, 12 TB ain't on the menu. Win 8 maxes out at 512 GB and Win 10 at 2TB. That's the actual physical limit, not the theoretical kernel-mode or whatever address limit. Sticking more than that into a machine May Not Achieve The Desired Result. Unless, of course, the desired result is to cause your computer to melt.
Biting the hand that feeds IT © 1998–2020