CIPWTTKT&GC v0x14 (v20): Turning it off and on all day every day - Hardware & Software

[QUOTE=Strikebango;37459797]Seems like today is AMD news: [url]http://www.maximumpc.com/article/news/amd_talks_3rd_generation_bulldozer_core_called_steamroller_promises_better_branch_prediction[/url][/QUOTE] Bulldozer is a car crash in slow motion. AMD seriously has a lot of work to do to get out of the gutter, otherwise, they're fucked in the CPU market.

Maybe Gman can tell us how AMD CPU s work.

[QUOTE=Strikebango;37460086]Maybe Gman can tell us how AMD CPU s work.[/QUOTE] Anyone can do that. The answer is: Slowly.

[QUOTE=Strikebango;37460086]Maybe Gman can tell us how AMD CPU s work.[/QUOTE] No, man, if you want to summon him you gotta say his username 3 times! gman003-main gman003-main gman003-main He's like Beetlejuice.

[QUOTE=Strikebango;37460086]Maybe Gman can tell us how AMD CPU s work.[/QUOTE] A car crash in slow motion

[highlight][h2]I AWAKEN WHO IS THE ONE WHO HAS SUMMONED ME[/h2][/highlight]

[QUOTE=gman003-main;37460311][highlight][h2]I AWAKEN WHO IS THE ONE WHO HAS SUMMONED ME[/h2][/highlight][/QUOTE] Strikebango wants you to tell us how AMD CPUs work

welp, there goes my plans. I found out that newegg.com can't order to canada and newegg.ca has no selection at all.

Why not use NCIX?

[QUOTE=Walrus.;37460360]welp, there goes my plans. I found out that newegg.com can't order to canada and newegg.ca has no selection at all.[/QUOTE] And exactly what were you trying to get from Newegg?

I thought you had to stand in front of a mirror to summon you?

[QUOTE=wickedplayer494;37460370]And exactly what were you trying to get from Newegg?[/QUOTE] A[sp]prebuilt[/sp]computer. [URL="http://www.newegg.com/Product/Product.aspx?Item=N82E16883103421"]This[/URL], to be specific

[QUOTE=Walrus.;37460449]A[sp]prebuilt[/sp]computer. [URL="http://www.newegg.com/Product/Product.aspx?Item=N82E16883103421"]This[/URL], to be specific[/QUOTE] I fucking love integrated graphics! But why prebuilt?

[QUOTE=Walrus.;37460449]A -snip, disgusting- computer. [URL="http://www.newegg.com/Product/Product.aspx?Item=N82E16883103421"]This[/URL], to be specific[/QUOTE] please link this and provide a NSFW warning, you are on a worksafe forum

[QUOTE=Walrus.;37460449]A[sp]prebuilt[/sp]computer. [URL="http://www.newegg.com/Product/Product.aspx?Item=N82E16883103421"]This[/URL], to be specific[/QUOTE] Ew. NCIX doesn't have it either, so you're hosed.

[QUOTE=wickedplayer494;37460341]Strikebango wants you to tell us how AMD CPUs work[/QUOTE] [highlight][h2]AND SO IT SHAL[/h2][/highlight] wait, what? You summon me for [i]that[/i]? Has he not heard of Wikipedia? Or Google? Anyways, basic gist of it was that AMD realized that there are a lot of parts in each core that don't really get used that much, and they decided to make them shared between two "cores" to save space so they could cram more cores onto one chip. The main things were like the 256-bit SIMD unit and the instruction fetcher and decoder. So they saved space, but that also hurt their performance - they're about 20% slower clock-for-clock compared to Sandy Bridge. They also did some changes to the design process that was supposed to make it easier for them to work with the team formerly known as ATI, which really trashed their transistor density (they have far, far fewer transistors per mm^2 than Intel, and since the die size is the main factor in price...). There were a few other issues too (the cache miss rate doubled, for instance). A lot of it, though, was marketing. Each of their "cores" was much weaker than an Intel core, but their two-core "module" would beat even a hyperthreaded Intel core. Had they marketed their 4-module Bulldozer as a quad-core, they would have beaten Intel's quad-cores (assuming they somehow managed to have similar prices). Unfortunately, they didn't - they marketed their 4-module Bulldozer as an 8-core chip. Steamroller seems to be fixing some of those (they're splitting the decoder back into two units, adding more L1 cache, and changing their design process to increase transistor density) but they've lost a lot of time - I expect Haswell will be at least imminent by the time Steamroller is released. And Intel still has a fab process lead - 22nm will beat the 28nm AMD is targeting.

[QUOTE=gman003-main;37460488][highlight][h2]AND SO IT SHAL[/h2][/highlight] wait, what? You summon me for [i]that[/i]? Has he not heard of Wikipedia? Or Google? Anyways, basic gist of it was that AMD realized that there are a lot of parts in each core that don't really get used that much, and they decided to make them shared between two "cores" to save space so they could cram more cores onto one chip. The main things were like the 256-bit SIMD unit and the instruction fetcher and decoder. So they saved space, but that also hurt their performance - they're about 20% slower clock-for-clock compared to Sandy Bridge. They also did some changes to the design process that was supposed to make it easier for them to work with the team formerly known as ATI, which really trashed their transistor density (they have far, far fewer transistors per mm^2 than Intel, and since the die size is the main factor in price...). There were a few other issues too (the cache miss rate doubled, for instance). A lot of it, though, was marketing. Each of their "cores" was much weaker than an Intel core, but their two-core "module" would beat even a hyperthreaded Intel core. Had they marketed their 4-module Bulldozer as a quad-core, they would have beaten Intel's quad-cores (assuming they somehow managed to have similar prices). Unfortunately, they didn't - they marketed their 4-module Bulldozer as an 8-core chip. Steamroller seems to be fixing some of those (they're splitting the decoder back into two units, adding more L1 cache, and changing their design process to increase transistor density) but they've lost a lot of time - I expect Haswell will be at least imminent by the time Steamroller is released. And Intel still has a fab process lead - 22nm will beat the 28nm AMD is targeting.[/QUOTE] Shared decoder will be a HUGE increase in performance from Bulldozer, I'd like to see AMD come good again but they're too far behind Intel. They also really need to fix their cache latencies which I hope they do with Steamroller.

[QUOTE=mblunk;37460458]I fucking love integrated graphics! But why prebuilt?[/QUOTE] I've explained this in the quick questions thread and I've decided that it's worth the extra money. I don't know a thing about building computers and I don't find it very interesting. e: I posted it in that thread wondering if it was a good deal and got a lecture about prebuilt computers so I wasn't set on buying that particular one.

[QUOTE=Walrus.;37460553]I've explained this in the quick questions thread and I've decided that it's worth the extra money. I don't know a thing about building computers and I don't find it very interesting. e: I posted it in that thread wondering if it was a good deal and got a lecture about prebuilt computers so I wasn't set on buying that particular one.[/QUOTE] [img]http://www.w3.org/2005/Talks/0308-semweb-em/legos.jpg[/img] I built my first PC when I was 15, not having any clue what I was doing. It booted right up as soon as I was done, after about 2 hours. The hardest part is finding the page of your mobo manual that has the front panel connector layout labeled (and that's only if the labels aren't already printed on the board), and you get to pick exactly the parts you need, including ones that won't die after 2 years due to use of the absolute cheapest components possible to maximize profits. [b]If you know how to use a screwdriver, you can build a computer.[/b]

[QUOTE=mblunk;37460599][img]http://www.w3.org/2005/Talks/0308-semweb-em/legos.jpg[/img][/QUOTE] Except easier. And costs slightly more. Seriously, it ain't hard at all. Or just go to your local PC shop, they'll put it together for you.

[QUOTE=tratzzz;37460689]Except easier. [b]And costs slightly more.[/b] Seriously, it ain't hard at all. Or just go to your local PC shop, they'll put it together for you.[/QUOTE] No way, Have you seen the price of Lego?

I think they could do it if they took the "integrated full-power GPU" idea and ran with it. I'm sure, at the very least, they could share circuitry between the GPU and some of the CPU SIMD units, and definitely do some shared cache (actually, it would be really good if they could dynamically translate SSE instructions into MIMD Radeon instructions and just pass it to that, but that might be too hard). Add in a quad-channel (or better) memory controller to handle the memory bandwidth needed for both of them, hire some coders to make Java and MySQL run some stuff on the GPU, and you'd have a decently formidable chip.

[QUOTE=gman003-main;37460735]I think they could do it if they took the "integrated full-power GPU" idea and ran with it. I'm sure, at the very least, they could share circuitry between the GPU and some of the CPU SIMD units, and definitely do some shared cache (actually, it would be really good if they could dynamically translate SSE instructions into MIMD Radeon instructions and just pass it to that, but that might be too hard). Add in a quad-channel (or better) memory controller to handle the memory bandwidth needed for both of them, hire some coders to make Java and MySQL run some stuff on the GPU, and you'd have a decently formidable chip.[/QUOTE] Java and Database on the GPU... wat.

Oh, and guess who found LATE content: [url=http://www.citrix.com/lang/English/lp/lp_2328330.asp]Cloud computing survey[/url]

[QUOTE=gman003-main;37460775]Oh, and guess who found content: [url=http://www.citrix.com/lang/English/lp/lp_2328330.asp]Cloud computing survey[/url] [img]http://www.citrix.com/site/resources/dynamic/additional/demos/bin/mm/inpg/StormyWeather.png[/img][/QUOTE] Oh my God this was just on the last page

[QUOTE=gman003-main;37460735]I think they could do it if they took the "integrated full-power GPU" idea and ran with it. I'm sure, at the very least, they could share circuitry between the GPU and some of the CPU SIMD units, and definitely do some shared cache (actually, it would be really good if they could dynamically translate SSE instructions into MIMD Radeon instructions and just pass it to that, but that might be too hard). Add in a quad-channel (or better) memory controller to handle the memory bandwidth needed for both of them, hire some coders to make Java and MySQL run some stuff on the GPU, and you'd have a decently formidable chip.[/QUOTE] Commercial OLTP workloads (like the mentioned Java and MySQL) tend to barely even touch the SIMD units, if at all. They're about as non-DLP-rich as you can get.

[QUOTE=wingless;37460763]Java and Database on the GPU... wat.[/QUOTE] Java, at least for business-type number-crunching apps, might work well. I know databases on the GPU aren't really a good idea (unless you're really misusing the database, which is surprisingly common), but [i]executives[/i] won't know that. They'll just see "Optimized for AMD CPUs" and say "Get us an AMD server!" [editline]29th August 2012[/editline] [QUOTE=wickedplayer494;37460784]Oh my God this was just on the last page[/QUOTE] Sorry, just awakened from eternal slumber, still catching up.

[QUOTE=gman003-main;37460797]Java, at least for business-type number-crunching apps, might work well. I know databases on the GPU aren't really a good idea (unless you're really misusing the database, which is surprisingly common), but [i]executives[/i] won't know that. They'll just see "Optimized for AMD CPUs" and say "Get us an AMD server!" [editline]29th August 2012[/editline] Sorry, just awakened from eternal slumber, still catching up.[/QUOTE] Java you could [i]maybe[/i] get away with, but there's just no point to a database on the GPU. Selling it as a feature would just be sabotaging and evil. More so than selling a 4 module CPU as 8 module.

I was stretching for an example, sorry. Been one of those days. Actually, wait a second. Would encryption/decryption be a decently parallel task? That might be worth offloading to the GPU (I know it's already offloaded to dedicated crypto coprocessors sometimes, doing it on the GPU might be worth looking into).

[QUOTE=wingless;37460823]Java you could [i]maybe[/i] get away with, but there's just no point to a database on the GPU. Selling it as a feature would just be sabotaging and evil. More so than selling a 4 module CPU as 8 module.[/QUOTE] You couldn't get away with Java. JVM's tend to be very sensitive to cache size and latency, and GPU memory latency is far nastier than anything you see on a PC or server.