Just to reply again, but my desktop PC probably runs at 100% for less than 30 hours a month. The extra 100 watts for 30 hours or so is going to be pennies. So over 3-5 years, making up even $50 in processor price difference is going to be hard to do.
Well its a lot, but I think you overestimate the cooling. 120-130W is pretty common for most high end CPUs today. And GPU's do 200+ in a double high PCI slot.
You got the Megahertz myth wrong! The only myth is that Megahertz isn't important!
Tong in cheek and all that, but....
Frankly, today both AMD and Intel are at an IPC wall nearly as much as they have been at a clockrate wall. So, yes faster clock rate is pretty much the only way to get performance if your application doesn't scale with cores. Which is a shocking number of them.
The part I find interesting, is that if they can beat the haswell with this part then they probably have an IPC advantage over intel again. Remember the top end haswell turbo boosts to 4.9Ghz.
The only real difference then is how much power they take doing it. Which may not be as significant for an average system that is idle or running without the turbo boost for a significant portion of the time.
If AMD is within a few percent in the single thread area (especially on specint) and the thing is priced like a traditional AMD then they my next machine will probably be something like this.
At least when Bush was elected he was upfront about being a war monger.
The second time, the first time around he sounded like some kind of leftist hippy criticizing all the nation building and police actions the US was involved in around the world. I remember some interviewer pressing him on the issue, by listing assorted military actions and basically challenging him to say he wouldn't have done the same thing. I think out of the 5 or 6 listed Bush only was willing to really challenge one of them.
That was part of Bush all the R's forgot the second time around. Mr upstanding sticks to his principals did a 180 on half the issues he ran on the first time around.
Not to say Gore was any better, that guy came off as big brother incarnate. Same with Obama, Mr "Hope and Change" hasn't changed shit. In fact his "crowning achievement" is a healthcare bill that is a insurance company wet dream. And more of the same government support of big business over the individual that has been going on forever in this country.
Yes, but Microsoft Windows is more like IBM's old attempt at a PC monopoly, not like IBM's mainframe business.
I don't think so, the day Compaq released a legal PC clone that ran 99% of the same software that the IBM PC. IBM lost any "PC monopoly" they might have had. They tried to regain it with microchannel on the PS/2 but microchannel was just as easily replaced and I can't think of a single micochannel card that didn't have similar functionality on ISA/EISA or eventually VLB. Not that the PC was anywhere near a monopoly before the mid 80s and the bazillion clones. So really, IBM never had a PC monopoly.
No so with windows, the closest thing to a windows clone is reactos which still can't run the majority of PC applications without some serious finger crossing or hacking.
CP/M didn't have the kind of business backing or vendor lock that the mainframe customers did in the business space. Banks, airlines, etc didn't base their company processes around them. The people running wordstar, visicalc, etc all had alternative platforms the products ran on. So, as CP/M died (or was killed depending on perspective) everyone just moved their files to alternative platforms. Plus, it was pretty easy to move to an apple ][ or especially PC from CP/M because of the emulator cards that were everywhere. I actually still have a bootable CP/M board for my Apple ][+.
Anyway, some it also seems to be the fact that the customer base was more reliant on COTS type software packages rather than custom programming. The machines were to expensive for the hacker types, and to wimpy for the "serious business/system" programmer.
Windows is a much harder platform to emulate right now. That may not be the case in 10 years, maybe wine reactos, or some pure virtualization solution will replace it. But I doubt it, unless every application vendor ports their app to android and makes it functional I don't see that happening. Right now I have hardware/software that won't run in vista/7 because the vendor hasn't released anything since the windows 2k days.
third party alternatives pretty quickly killed IBM's business.
IBM is busy trying to spin off their x86 server business, storage business/etc. What is the one they you don't hear them talking about selling?
IBM's mainframe business is the model for the rest of the company. The margins are huge, where else can you charge 6 and 7 figure sums for yearly maintenance on an OS, much less all the other stuff that goes with it.
Actually, I just killed chrome 27.0.1453 too. Which I thought had a single thread per tab. I guess it depends on what your doing because there seems to be some resource locking or thread pooling going on.
The ability for chrome to consume more than one CPU and more than 4GB of ram is one of the reasons I don't run it. When some javascript or web site goes haywire with firefox, firefox dies long before it affects my machine. No so with chrome which has the ability to bring all operations on my machine to halt.
This is a strong overstatement, if M$ never released another version of windows you would still find windows being installed on new machine 30 years from now. Simply because there are places where its the "one true choice" for any number of reasons. Right now they are loosing a lot of home users, to the tablet market as people discover they don't need a PC to facebook. But users running $NICHE_BUSINESS software or whatever will continue to buy windows even if the desktop license costs go up 10x.
That is basically the IBM business model, milk a smaller and smaller set of mainframe customers for more money. Which is why its nearly impossible to find numbers on mainframe machine sales, rather its listed as "total installed MIPS" or some other BS that fails to account for the fact that adding CPU's caches a couple years ago gave them a 10x performance improvement per machine, plus selling 5.5Ghz machines helps too.
The total install base is shrinking, but the workload is increasing slightly. This doesn't account for POWER, where they haven't figured out how to sell machines at 1000x markup over x86 because the customer base is quite capable of porting an AIX application to linux.
If you wrote an NTP client in javascript, you would soon discover that its nearly useless. NTP is all about attempting to determine clock and network jitter and correct for it. If your running in a language like javascript your going to have a really hard time getting that jitter low enough to produce a usable answer. On most browsers you probably can't even guarantee second granularity because of javascript delays caused by scripts running in other tabs much less the OS's scheduling of the browser.
And you will note, that page says "Accurate within.1 seconds" in the java plugin that runs.
So, they used java and its sill massively inaccurate (its probably nearly impossible to get more accurate in a web page due to the inability to do low level scheduling tweaks).
The clock on your average phone is going to have a couple more decimal places.
Getting the proper time out to a couple decimal places is actually pretty hard on a modern computer, and hence NTP and local daemons that attempt to determine the clock drift on your PC.
Are you using a 3rd party case? I have the HP case and it works fine, although I've heard of people having issues with other cases. I also think there was a bad firmware version in there that screwed up the inductive charging, I would make sure your not running that version.
With mine, I put it on the base and make sure I hear it go boing and then forget about it. If it doesn't go boing (or sometimes goes boing more than once) I do a better job positioning it on the charger. That is what is nice about the veer, it sort of sucks itself into place with the magnets.
Of course, YMMV is sort of the trend with the touchpad, and to be expected from a 1st gen device dropped a few weeks after release. My touchpad needs fairly regular reboots because it has the sound corruption bug, plus if I leave a bunch of cards open for a couple weeks it seems like it leaks memory so that I get the "to many cards open message" even when I don't have any open.
Dock based, inductive charging is ~85% efficient, due to being something like 5mm of separation between the coils, running at very high frequency, and being actively controlled. So, this isn't your granddaddy's wireless power fantasies.
The loses in the 50% efficient wall warts shipping on most android phones are a worse problem.
What amazes me is that inductive charging hasn't taken over. I was a skeptic, when I got my touchpad a couple years ago. The ability to just drop the pad on a dock without worrying too much about positioning/etc quickly sold me on the idea. Same thing with the veer I purchased as well. Just drop it on the dock and the magnets align it.
Now every-time I plug in the wifes ipad, or android phone I cringe. Small easily broken connectors are something that should be a last resort.
Oh, and the touchpad prompts the user before allowing communication on the USB port.
Especially if your optimizing for different subsets of the instruction set. Trycrypt for sure, is using AES-NI paths on CPUs that support it, otherwise its using generic ones.
My comment was more in regard to gcc, which I don't believe i've ever seen automatically generate multiple arch dependent code paths. I think gcc assumes you will make multiple compilation paths with different -march's and glue them together if you want something better than the generic target.
Most precompiled software is optimized for a pretty old common target. I would be highly surprised if default code emitted by e.g. GCC is slower on a Core now, than it was five years ago.
Most software is not 1/2 of these benchmarks. I agree you build a generic binary for x86_64 with gcc its basically running the same code on all CPUs. But, povray, cinebench and x264 are all highly optimized. It wouldn't surprise me if all of them are using icc, and have tweaked in on way or the other by intel.
Frankly, this is nothing new. The CPU/GPU/machine vendors have traditionally been willing to help high profile or applications selected for benchmarks run well on their hardware. The problem is that these benchmarks aren't representative of a lot of actual work being done on PCs. If you choose a different video encoding library, you may find completely different results.
Basically, IMHO a significant percentage of the generic performance improvement on intel CPU's in the last 5 years have been benchmarking tricks. Nearly every in house application i've been benchmarking with generic x86_64 gcc targets disagrees with the conclusions drawn on places like tom's hardware. Part of that is gcc (its still not as good as icc, at optimizing for a specific target) part of it is that the basic integer/move/compare/etc instructions that comprise most code haven't gotten much faster. In my tests the only thing intel has done since core2 that affects that kind of code in any significant way has been fixing rep stosd so that it runs at a reasonable rate with respect to the dozen alternative ways to copy memory. This is also why there is a lot of talk that the tables are turned in the AMD/Intel land if your running linux/gcc. Its because AMD looks good and even beats intel in many cases on generic semi optimized linux/gcc code paths that don't leverage SSE4.
The app I work on needs a couple GB/sec (not bit) of encryption bandwidth. So, I love having AES-NI.... But.... We have encryption/compression acceleration hardware in the machine. Which means we won't be using AES-NI because the CPU's haven't increased their compression speeds to match their new found encryption speeds. Plus, we (like many other things, including truecrypt) have always offered the user the ability to encrypt with algorithms that were more CPU friendly than AES.
Anyway, two points. In the high bandwidth web servers I've seen, the SSL public key handshake part of the initial negotiation was the problem with SSL, not the AES/etc. As such, i've seen accelerators for SSL setup too! Frankly, its possible to get ~100MB/sec out of even older CPU's doing software AES. So a 8x CPU machine can pretty much keep a 10Gb link saturated if all your doing is AES... Doesn't leave much time for anything else, but you can either pay for more CPU's or buy accelerators if you actually need the bandwidth. Frankly, adding a couple extra CPU's is not really going to be your problem if your transferring more than 10Gbit of web traffic.
But this is all xeon territory, not desktop CPU's. Which is where truecrypt comes in. If your running it on a recent SSD and you just have to have AES instead of something faster, AES-NI is going to be your thing. But its quite possibly the one desktop application with the largest performance difference between a modern CPU and one a couple generations back. Your not going to be pulling 10x+ performance increases on many other applications. So putting it in a general cpu comparison except as a footnote is a little disingenuous. Same for some of these other things where openCL/CUDA offloads have become the norm so the new AVX/etc instructions aren't necessarily such a big win. Not that I mind having them either!
Just to reply again, some of these benchmarks are obvious bullshit. Like the AES one, the new cpu's have AES-NI instructions for accelerating AES.
So, yes if you happen to be doing AES, and your running code that can take advantage of AES-NI then the new CPU's are going to fly. But the whole benchmark is so tilted its not even funny. Why not use a benchmark that renders some SSL encoded web pages? Because the benchmark is going to be bottlenecked by the network stack and the rendering engine. Not the tiny percentage of the time the CPU spends doing AES.
He really should have broken the benchmarks into two camps, benchmarks actually running the same code on both sets of CPU's and ones that can leverage some new instructions available only on the new CPUs.
You should look at that page again, i'm not sure is benchmarking is "fair". If he wants to compare the older CPU's it might have been ammusing to use the 5 year old version of the code too. That way, your not seeing the affects of code optimized for the latest CPU's at the expense of the old ones.
Ignoring things that are using SSE modes not available on the core2 (cinebench, etc!).
The i7-4770 is actually clocked at 3.9Ghz when running a single core. So for something like 7 zip, 4807/3.9=1232 units/ghz. Vs the core2 at 2519/2.5=1007 units/Ghz. The numbers don't look that great.
We are talking a CPU with more cache, faster RAM, faster PCIe... Its only getting a puny 20% performance improvement in 7-zip single threaded over nearly 5 years!
Now of course that is the pessimistic view, the optimistic is that if you application can take advantage of 256bit AVX, or a dozen cores then the CPU's have gotten significantly faster.
But, I've seen this a few times, someone goes out and purchases a new machine to replace an old high end machine and the new machine is actually slower at some number of given tasks. This happened where I worked, we replaced a couple 3.2Ghz opteron machines with some nice new 2.0 Ghz xeons, and the latency of our application went up ~25% and the throughput went down by ~10%. I'm sure if we had purchased better xeons that wouldn't have happened. But, the guys in purchasing assumed that spending the same amount of money for a new machine as they spent 5 years ago on the old ones would yield a better machine. They were wrong!
barely understood linear algebra when I learned it. I didn't understand it until I started using it, in classes like numerical programming or quantum mechanics, etc.
And, I would move that this is the fundamental problem with a _LOT_ of education. Sure, you can rotate and transform that matrix in linear algebra class to solve for n values.... But do you really understand what your doing, what its really for? Nah, that takes real applications, and being presented with problems you cannot solve without the math. Or for that matter problems that require you to use that math in a way not taught in class.
So, i was in the same boat you were, busy learning some aspect of diffeq in semiconductor devices that in theory I already new based on the fact that I passed the diffeq class. Same thing with some AI/ML classes I took (yah I was one of the few that completed the big MOG) wow were my statistics bad/rusty. But I might move, that it wasn't until I took that class that I really understood some of the stuff that we spent 1 day on when I was in college 15 years ago.
So, maybe the problem is that we even try to teach math without the applications. Of course teaching it with the applications means that instead of spending a semester on linear algebra we have to spend three or four because the applications get in the way of "learning the math".
Slashdot Mirror
Just to reply again, but my desktop PC probably runs at 100% for less than 30 hours a month. The extra 100 watts for 30 hours or so is going to be pennies. So over 3-5 years, making up even $50 in processor price difference is going to be hard to do.
My point was, that if they can match or beat a 4.9 Ghz part with a 5Ghz part, then the IPC's are going to be similar.
I didn't say the IPC changes with clock rate...
TDP for this amd part: 220W
Well its a lot, but I think you overestimate the cooling. 120-130W is pretty common for most high end CPUs today. And GPU's do 200+ in a double high PCI slot.
You got the Megahertz myth wrong! The only myth is that Megahertz isn't important!
Tong in cheek and all that, but....
Frankly, today both AMD and Intel are at an IPC wall nearly as much as they have been at a clockrate wall. So, yes faster clock rate is pretty much the only way to get performance if your application doesn't scale with cores. Which is a shocking number of them.
The part I find interesting, is that if they can beat the haswell with this part then they probably have an IPC advantage over intel again. Remember the top end haswell turbo boosts to 4.9Ghz.
The only real difference then is how much power they take doing it. Which may not be as significant for an average system that is idle or running without the turbo boost for a significant portion of the time.
If AMD is within a few percent in the single thread area (especially on specint) and the thing is priced like a traditional AMD then they my next machine will probably be something like this.
I sure hope they release a smallish 4k display (sub 30") for a reasonable price ($1.5k).
I'm not going to buy one of their machines, but I might buy a monitor or two from them if they get the specs right.
The problem is that HP-UX is still the same today as it was in 1999.
Don't forget IBM's as400/IBM i/or whatever its called this week. Plus HP still has the nonstop too.
At least when Bush was elected he was upfront about being a war monger.
The second time, the first time around he sounded like some kind of leftist hippy criticizing all the nation building and police actions the US was involved in around the world. I remember some interviewer pressing him on the issue, by listing assorted military actions and basically challenging him to say he wouldn't have done the same thing. I think out of the 5 or 6 listed Bush only was willing to really challenge one of them.
That was part of Bush all the R's forgot the second time around. Mr upstanding sticks to his principals did a 180 on half the issues he ran on the first time around.
Not to say Gore was any better, that guy came off as big brother incarnate. Same with Obama, Mr "Hope and Change" hasn't changed shit. In fact his "crowning achievement" is a healthcare bill that is a insurance company wet dream. And more of the same government support of big business over the individual that has been going on forever in this country.
Yes, but Microsoft Windows is more like IBM's old attempt at a PC monopoly, not like IBM's mainframe business.
I don't think so, the day Compaq released a legal PC clone that ran 99% of the same software that the IBM PC. IBM lost any "PC monopoly" they might have had. They tried to regain it with microchannel on the PS/2 but microchannel was just as easily replaced and I can't think of a single micochannel card that didn't have similar functionality on ISA/EISA or eventually VLB. Not that the PC was anywhere near a monopoly before the mid 80s and the bazillion clones. So really, IBM never had a PC monopoly.
No so with windows, the closest thing to a windows clone is reactos which still can't run the majority of PC applications without some serious finger crossing or hacking.
CP/M didn't have the kind of business backing or vendor lock that the mainframe customers did in the business space. Banks, airlines, etc didn't base their company processes around them. The people running wordstar, visicalc, etc all had alternative platforms the products ran on. So, as CP/M died (or was killed depending on perspective) everyone just moved their files to alternative platforms. Plus, it was pretty easy to move to an apple ][ or especially PC from CP/M because of the emulator cards that were everywhere. I actually still have a bootable CP/M board for my Apple ][+.
Anyway, some it also seems to be the fact that the customer base was more reliant on COTS type software packages rather than custom programming. The machines were to expensive for the hacker types, and to wimpy for the "serious business/system" programmer.
Windows is a much harder platform to emulate right now. That may not be the case in 10 years, maybe wine reactos, or some pure virtualization solution will replace it. But I doubt it, unless every application vendor ports their app to android and makes it functional I don't see that happening. Right now I have hardware/software that won't run in vista/7 because the vendor hasn't released anything since the windows 2k days.
third party alternatives pretty quickly killed IBM's business.
IBM is busy trying to spin off their x86 server business, storage business/etc. What is the one they you don't hear them talking about selling?
IBM's mainframe business is the model for the rest of the company. The margins are huge, where else can you charge 6 and 7 figure sums for yearly maintenance on an OS, much less all the other stuff that goes with it.
Btw: JavaScript isn't blocked by other pages.
Actually, I just killed chrome 27.0.1453 too. Which I thought had a single thread per tab. I guess it depends on what your doing because there seems to be some resource locking or thread pooling going on.
The ability for chrome to consume more than one CPU and more than 4GB of ram is one of the reasons I don't run it. When some javascript or web site goes haywire with firefox, firefox dies long before it affects my machine. No so with chrome which has the ability to bring all operations on my machine to halt.
but browsers are multithreaded (otherwise a while loop on one page would block the whole browser).
Really? Have you tried that in all the browsers lately?
Because I just tried it in firefox 20.0.1 to make sure they haven't changed it.
while (x100000) { x++; }
actually locked up the UI in the other windows until the script warning popped up.
Windows and Office are dying
This is a strong overstatement, if M$ never released another version of windows you would still find windows being installed on new machine 30 years from now. Simply because there are places where its the "one true choice" for any number of reasons. Right now they are loosing a lot of home users, to the tablet market as people discover they don't need a PC to facebook. But users running $NICHE_BUSINESS software or whatever will continue to buy windows even if the desktop license costs go up 10x.
That is basically the IBM business model, milk a smaller and smaller set of mainframe customers for more money. Which is why its nearly impossible to find numbers on mainframe machine sales, rather its listed as "total installed MIPS" or some other BS that fails to account for the fact that adding CPU's caches a couple years ago gave them a 10x performance improvement per machine, plus selling 5.5Ghz machines helps too.
The total install base is shrinking, but the workload is increasing slightly. This doesn't account for POWER, where they haven't figured out how to sell machines at 1000x markup over x86 because the customer base is quite capable of porting an AIX application to linux.
If you wrote an NTP client in javascript, you would soon discover that its nearly useless. NTP is all about attempting to determine clock and network jitter and correct for it. If your running in a language like javascript your going to have a really hard time getting that jitter low enough to produce a usable answer. On most browsers you probably can't even guarantee second granularity because of javascript delays caused by scripts running in other tabs much less the OS's scheduling of the browser.
And you will note, that page says "Accurate within .1 seconds" in the java plugin that runs.
So, they used java and its sill massively inaccurate (its probably nearly impossible to get more accurate in a web page due to the inability to do low level scheduling tweaks).
The clock on your average phone is going to have a couple more decimal places.
Getting the proper time out to a couple decimal places is actually pretty hard on a modern computer, and hence NTP and local daemons that attempt to determine the clock drift on your PC.
Are you using a 3rd party case? I have the HP case and it works fine, although I've heard of people having issues with other cases. I also think there was a bad firmware version in there that screwed up the inductive charging, I would make sure your not running that version.
With mine, I put it on the base and make sure I hear it go boing and then forget about it. If it doesn't go boing (or sometimes goes boing more than once) I do a better job positioning it on the charger. That is what is nice about the veer, it sort of sucks itself into place with the magnets.
Of course, YMMV is sort of the trend with the touchpad, and to be expected from a 1st gen device dropped a few weeks after release. My touchpad needs fairly regular reboots because it has the sound corruption bug, plus if I leave a bunch of cards open for a couple weeks it seems like it leaks memory so that I get the "to many cards open message" even when I don't have any open.
Inductive charging is highly wasteful.
Dock based, inductive charging is ~85% efficient, due to being something like 5mm of separation between the coils, running at very high frequency, and being actively controlled. So, this isn't your granddaddy's wireless power fantasies.
The loses in the 50% efficient wall warts shipping on most android phones are a worse problem.
What amazes me is that inductive charging hasn't taken over. I was a skeptic, when I got my touchpad a couple years ago. The ability to just drop the pad on a dock without worrying too much about positioning/etc quickly sold me on the idea. Same thing with the veer I purchased as well. Just drop it on the dock and the magnets align it.
Now every-time I plug in the wifes ipad, or android phone I cringe. Small easily broken connectors are something that should be a last resort.
Oh, and the touchpad prompts the user before allowing communication on the USB port.
That is fairly common when optimizing things.
Especially if your optimizing for different subsets of the instruction set. Trycrypt for sure, is using AES-NI paths on CPUs that support it, otherwise its using generic ones.
My comment was more in regard to gcc, which I don't believe i've ever seen automatically generate multiple arch dependent code paths. I think gcc assumes you will make multiple compilation paths with different -march's and glue them together if you want something better than the generic target.
Most precompiled software is optimized for a pretty old common target. I would be highly surprised if default code emitted by e.g. GCC is slower on a Core now, than it was five years ago.
Most software is not 1/2 of these benchmarks. I agree you build a generic binary for x86_64 with gcc its basically running the same code on all CPUs. But, povray, cinebench and x264 are all highly optimized. It wouldn't surprise me if all of them are using icc, and have tweaked in on way or the other by intel.
PovRay an open source project has "intel attention" http://software.intel.com/en-us/blogs/2009/03/26/some-tips-for-povray-1
Frankly, this is nothing new. The CPU/GPU/machine vendors have traditionally been willing to help high profile or applications selected for benchmarks run well on their hardware. The problem is that these benchmarks aren't representative of a lot of actual work being done on PCs. If you choose a different video encoding library, you may find completely different results.
Basically, IMHO a significant percentage of the generic performance improvement on intel CPU's in the last 5 years have been benchmarking tricks. Nearly every in house application i've been benchmarking with generic x86_64 gcc targets disagrees with the conclusions drawn on places like tom's hardware. Part of that is gcc (its still not as good as icc, at optimizing for a specific target) part of it is that the basic integer/move/compare/etc instructions that comprise most code haven't gotten much faster. In my tests the only thing intel has done since core2 that affects that kind of code in any significant way has been fixing rep stosd so that it runs at a reasonable rate with respect to the dozen alternative ways to copy memory. This is also why there is a lot of talk that the tables are turned in the AMD/Intel land if your running linux/gcc. Its because AMD looks good and even beats intel in many cases on generic semi optimized linux/gcc code paths that don't leverage SSE4.
The app I work on needs a couple GB/sec (not bit) of encryption bandwidth. So, I love having AES-NI.... But.... We have encryption/compression acceleration hardware in the machine. Which means we won't be using AES-NI because the CPU's haven't increased their compression speeds to match their new found encryption speeds. Plus, we (like many other things, including truecrypt) have always offered the user the ability to encrypt with algorithms that were more CPU friendly than AES.
Anyway, two points. In the high bandwidth web servers I've seen, the SSL public key handshake part of the initial negotiation was the problem with SSL, not the AES/etc. As such, i've seen accelerators for SSL setup too! Frankly, its possible to get ~100MB/sec out of even older CPU's doing software AES. So a 8x CPU machine can pretty much keep a 10Gb link saturated if all your doing is AES... Doesn't leave much time for anything else, but you can either pay for more CPU's or buy accelerators if you actually need the bandwidth. Frankly, adding a couple extra CPU's is not really going to be your problem if your transferring more than 10Gbit of web traffic.
But this is all xeon territory, not desktop CPU's. Which is where truecrypt comes in. If your running it on a recent SSD and you just have to have AES instead of something faster, AES-NI is going to be your thing. But its quite possibly the one desktop application with the largest performance difference between a modern CPU and one a couple generations back. Your not going to be pulling 10x+ performance increases on many other applications. So putting it in a general cpu comparison except as a footnote is a little disingenuous. Same for some of these other things where openCL/CUDA offloads have become the norm so the new AVX/etc instructions aren't necessarily such a big win. Not that I mind having them either!
Just to reply again, some of these benchmarks are obvious bullshit. Like the AES one, the new cpu's have AES-NI instructions for accelerating AES.
So, yes if you happen to be doing AES, and your running code that can take advantage of AES-NI then the new CPU's are going to fly. But the whole benchmark is so tilted its not even funny. Why not use a benchmark that renders some SSL encoded web pages? Because the benchmark is going to be bottlenecked by the network stack and the rendering engine. Not the tiny percentage of the time the CPU spends doing AES.
He really should have broken the benchmarks into two camps, benchmarks actually running the same code on both sets of CPU's and ones that can leverage some new instructions available only on the new CPUs.
You should look at that page again, i'm not sure is benchmarking is "fair". If he wants to compare the older CPU's it might have been ammusing to use the 5 year old version of the code too. That way, your not seeing the affects of code optimized for the latest CPU's at the expense of the old ones.
Ignoring things that are using SSE modes not available on the core2 (cinebench, etc!).
The i7-4770 is actually clocked at 3.9Ghz when running a single core. So for something like 7 zip, 4807/3.9=1232 units/ghz. Vs the core2 at 2519/2.5=1007 units/Ghz. The numbers don't look that great.
We are talking a CPU with more cache, faster RAM, faster PCIe... Its only getting a puny 20% performance improvement in 7-zip single threaded over nearly 5 years!
Now of course that is the pessimistic view, the optimistic is that if you application can take advantage of 256bit AVX, or a dozen cores then the CPU's have gotten significantly faster.
But, I've seen this a few times, someone goes out and purchases a new machine to replace an old high end machine and the new machine is actually slower at some number of given tasks. This happened where I worked, we replaced a couple 3.2Ghz opteron machines with some nice new 2.0 Ghz xeons, and the latency of our application went up ~25% and the throughput went down by ~10%. I'm sure if we had purchased better xeons that wouldn't have happened. But, the guys in purchasing assumed that spending the same amount of money for a new machine as they spent 5 years ago on the old ones would yield a better machine. They were wrong!
barely understood linear algebra when I learned it. I didn't understand it until I started using it, in classes like numerical programming or quantum mechanics, etc.
And, I would move that this is the fundamental problem with a _LOT_ of education. Sure, you can rotate and transform that matrix in linear algebra class to solve for n values.... But do you really understand what your doing, what its really for? Nah, that takes real applications, and being presented with problems you cannot solve without the math. Or for that matter problems that require you to use that math in a way not taught in class.
So, i was in the same boat you were, busy learning some aspect of diffeq in semiconductor devices that in theory I already new based on the fact that I passed the diffeq class. Same thing with some AI/ML classes I took (yah I was one of the few that completed the big MOG) wow were my statistics bad/rusty. But I might move, that it wasn't until I took that class that I really understood some of the stuff that we spent 1 day on when I was in college 15 years ago.
So, maybe the problem is that we even try to teach math without the applications. Of course teaching it with the applications means that instead of spending a semester on linear algebra we have to spend three or four because the applications get in the way of "learning the math".