Ranting about Microsoft and Tech Support in India

Daryl wrote:With PC hardware another secret is to find the "sweet spot". Buy newish kit but not cutting edge. A CPU that was released a year ago may for example be 3Ghz and cost $200, while the latest is 3.5Ghz and cost $400. Twice the price for a marginal improvement even if your other components can make use of the speed.

Seriously beware looking at clock speeds though.

Prime example, my dad´s previous system used a 3Ghz Pentium 4 based Celeron, my mom´s computer runs a 1.6Ghz Core2 based Celeron.
Mom´s system is clearly faster.

Also, beware the difference between cores and virtual cores.
It´s not long ago i had to do a really annoying long explanation about why an Intel i3 was not BETTER than a i5 at the same clockspeed.
Somehow, the person in question had been convinced that having 2 physical cores and 2 virtual cores was better than having 4 physical cores.

And then we have AMD, who nowadays build their cpu´s by modules instead of cores(i wont be surprised if it´s the "way of the future", but it´s not optimal right now), some Intel shills(and some people who simply don´t know better) try to claim that a module really equals a single core rather than 2.
Which is pure rubbish as there is only a single part which is completely shared between the two cores in a module, and that´s the FPU(floating point unit).
It´s still a bit more complex though, as the FPU is only shared if the computation being done is 128 bit, otherwise the FPU works as two 64 bit FPUs, one for each core in the module.
(and the reason for this setup is in part simply because AMDs FPU, ever since the first Athlon cpu´s, have been "too good", basically wasting transistors, as those FPUs were only very rarely even halfway used, essentially making the cpu´s a bit unbalanced )

Fireflair wrote:
Something else to remember, you're limited by your slowest component. And that is rarely your CPU. So having the 4GhZ with 12 cores and over clocked by 30% will be faster, yes. But a 3 GhZ with 10 cores will still get the job done just fine. Save the money and get a faster front bus, more RAM, a better video card and a SSD.

Thing to note, no computers sold today have a "front bus" any longer, doesn´t exist.
The Front Side Bus or commonly FSB was first cut away when AMD introduced it´s K8 cpu´s, with Intel following suit some years later when the Nehalem based cpu´s were launched ( Core i3/i5/i7 ).

(Which is why it´s so sad the performance gains on the Intel side wasn´t larger, AMD got a 30% boost when they did it, Intel got less than 10%, as the Nehalem architechture was really more suitable for servers than desktops)

Anyway, what replaced the FSB had it´s controller moved to the cpu, and the speed of this link as well as it´s low latencies, well, any claims today about how great a system´s FSB is, is basically just lies to fool you with.

The speeds are simply so high by now on all systems that unless you have something very special you need the system for, then exact numbers just doesn´t matter.

Even if you went and took the original Hyper Transport link AMD used on their first K8 cpu´s on a cpu of today, you wouldn´t be able to notice the difference(again, unless you do something that specifically stresses this link, which is rare outside of specialist workstations and servers).

I heard one guy use a restuaraunt as as an example of why pcs dont need a fsb anymore. he called the L1 cache a table you sat at, the L2 cache a serving trolley, and the L3 cache the maib buffet table in the middle of the room. the fsb was rollerskates with jets. he said it used to be you had food on your table, or he would skate to the kitchen and make you an order. that would take a minute, as opposed to half a second for having it on the table to start with. with the introduction of the L2, if it wasnt on the table, and it was on the trolley, slightly longer than from the L1, but quicker then going to the kitchen. L3, slower than L2, but quicker than kitchen. So it used to be 100% went through FSB at slow speeds. now, >80% goes through L1, >80% of remainder goes through L2, >80% of remainder goes through L3. result? going to the kitchen not required except in <1% of cases. so nowdays, improving the speed of the rockets does bubkus.....

Tenshinai wrote:

Fireflair wrote:
Something else to remember, you're limited by your slowest component. And that is rarely your CPU. So having the 4GhZ with 12 cores and over clocked by 30% will be faster, yes. But a 3 GhZ with 10 cores will still get the job done just fine. Save the money and get a faster front bus, more RAM, a better video card and a SSD.

Thing to note, no computers sold today have a "front bus" any longer, doesn´t exist.
The Front Side Bus or commonly FSB was first cut away when AMD introduced it´s K8 cpu´s, with Intel following suit some years later when the Nehalem based cpu´s were launched ( Core i3/i5/i7 ).

(Which is why it´s so sad the performance gains on the Intel side wasn´t larger, AMD got a 30% boost when they did it, Intel got less than 10%, as the Nehalem architechture was really more suitable for servers than desktops)

Anyway, what replaced the FSB had it´s controller moved to the cpu, and the speed of this link as well as it´s low latencies, well, any claims today about how great a system´s FSB is, is basically just lies to fool you with.

The speeds are simply so high by now on all systems that unless you have something very special you need the system for, then exact numbers just doesn´t matter.

Even if you went and took the original Hyper Transport link AMD used on their first K8 cpu´s on a cpu of today, you wouldn´t be able to notice the difference(again, unless you do something that specifically stresses this link, which is rare outside of specialist workstations and servers).

Spacekiwi wrote:I heard one guy use a restuaraunt as as an example of why pcs dont need a fsb anymore. he called the L1 cache a table you sat at, the L2 cache a serving trolley, and the L3 cache the maib buffet table in the middle of the room. the fsb was rollerskates with jets. he said it used to be you had food on your table, or he would skate to the kitchen and make you an order. that would take a minute, as opposed to half a second for having it on the table to start with. with the introduction of the L2, if it wasnt on the table, and it was on the trolley, slightly longer than from the L1, but quicker then going to the kitchen. L3, slower than L2, but quicker than kitchen. So it used to be 100% went through FSB at slow speeds. now, >80% goes through L1, >80% of remainder goes through L2, >80% of remainder goes through L3. result? going to the kitchen not required except in <1% of cases. so nowdays, improving the speed of the rockets does bubkus.....

Lol... Completely incorrect and totally confusing, but definitely funny description!


First of all, L1/L2/L3 are cache memory structures. If you go back to the -90s, then generally cpu´s had L1 and the motherboards had L2 and much less often L3.
Nowadays, most cpu´s have onchip cache from L1 to L3.

What cache memory DO, is to select data and instructions that are most used and store them, trying to keep delays for fetching stuff as low as possible.
The smaller the cache, the faster you can make it, so L1 tends to be tiny but have superfast transfer and reaction times while L3 tends to be large and relatively slow.


Now, what is/was FSB?
Oldstyle, the motherboard chipset consists of a northbridge and a southbridge.

Cpu connects to the northbridge, which is also what RAM and graphics(since AGP) connects to.

Southbridge connects network, PCI and ISA buses for addin cards, and it handles slow storage interfaces, IDE/PATA/SATA as well as USB, serial, paralell ports, keyboard and similar connections.

The FSB, is the link that connects the cpu to the northbridge.

So, EVERYTHING the cpu does, every piece of data that comes or goes to or from the cpu, regardless to or from where, goes through the FSB.

Now, a modern cpu changes things a lot.
Most importantly, it moves the RAM controller and connection onto the cpu itself. And this is THE main part of what the FSB needed lots of speed and bandwidth for.

And then, AMD switched to their Hypertransport in 2003 and Intel to their Quickpath Interconnect in 2008.

So, fastest Intel FSB ever used, managed 12.8GB/s(mostly theoretically and used in overclocked systems, like the one i´m typing this on). That was unidirectional and exclusive transfer, ie. the bus could only transfer a single packet of data at any one time and only in one direction. A severe drawback.

Fastest AMD used, was 3.2GB/s but was nonexclusive and bidirectional, so even though it sounds much slower, in reality AMDs fastest EV6 bus in 2003 was probably roughly equal to the fastest Intel had in 2008(better for some use, worse for some).

When AMD switched, their Hypertransport started out at 3.2GB/s. For a link that no longer needed to handle any transfers from the RAM. Today it runs at 12.8 up to 51.2GB/s.

Intel´s current QPI is roughly comparable(also in not being unidirectional and exclusive any longer(finally!)).


So in the end, the single most demanding transfer requirement of the FSB was removed, while what replaced the FSB has still been greatly speeded up.

Sure, gfx cards have speeded up and SATA and all, but it´s nothing even close to reaching the kind of bandwidth needed by the RAM memeory.

Hence, raising the speeds of the HT or QPI today becomes more or less irrelevant, as most things connecting through that link doesn´t need that much speed, nor does it have limitation issues that can be alleviated with greater speed ( like the old Intel FSB ).




So, need or not for FSB has nothing at all to do with any of the L1/L2/L3 caches. It has simply been replaced by a more dedicated link, while the RAM interface has been given it´s own separate link.

Talking about the cache structure is merely confusing and irrelevant.

Woops. its been a while since i heard it, so obviously i missed a fair bit of the description and just remembered the funny parts. thanks for filling in the rest Tenshinai. much appreciated.

Tenshinai wrote:

Spacekiwi wrote:I heard one guy use a restuaraunt as as an example of why pcs dont need a fsb anymore. he called the L1 cache a table you sat at, the L2 cache a serving trolley, and the L3 cache the maib buffet table in the middle of the room. the fsb was rollerskates with jets. he said it used to be you had food on your table, or he would skate to the kitchen and make you an order. that would take a minute, as opposed to half a second for having it on the table to start with. with the introduction of the L2, if it wasnt on the table, and it was on the trolley, slightly longer than from the L1, but quicker then going to the kitchen. L3, slower than L2, but quicker than kitchen. So it used to be 100% went through FSB at slow speeds. now, >80% goes through L1, >80% of remainder goes through L2, >80% of remainder goes through L3. result? going to the kitchen not required except in <1% of cases. so nowdays, improving the speed of the rockets does bubkus.....

Lol... Completely incorrect and totally confusing, but definitely funny description!


First of all, L1/L2/L3 are cache memory structures. If you go back to the -90s, then generally cpu´s had L1 and the motherboards had L2 and much less often L3.
Nowadays, most cpu´s have onchip cache from L1 to L3.

What cache memory DO, is to select data and instructions that are most used and store them, trying to keep delays for fetching stuff as low as possible.
The smaller the cache, the faster you can make it, so L1 tends to be tiny but have superfast transfer and reaction times while L3 tends to be large and relatively slow.


Now, what is/was FSB?
Oldstyle, the motherboard chipset consists of a northbridge and a southbridge.

Cpu connects to the northbridge, which is also what RAM and graphics(since AGP) connects to.

Southbridge connects network, PCI and ISA buses for addin cards, and it handles slow storage interfaces, IDE/PATA/SATA as well as USB, serial, paralell ports, keyboard and similar connections.

The FSB, is the link that connects the cpu to the northbridge.

So, EVERYTHING the cpu does, every piece of data that comes or goes to or from the cpu, regardless to or from where, goes through the FSB.

Now, a modern cpu changes things a lot.
Most importantly, it moves the RAM controller and connection onto the cpu itself. And this is THE main part of what the FSB needed lots of speed and bandwidth for.

And then, AMD switched to their Hypertransport in 2003 and Intel to their Quickpath Interconnect in 2008.

So, fastest Intel FSB ever used, managed 12.8GB/s(mostly theoretically and used in overclocked systems, like the one i´m typing this on). That was unidirectional and exclusive transfer, ie. the bus could only transfer a single packet of data at any one time and only in one direction. A severe drawback.

Fastest AMD used, was 3.2GB/s but was nonexclusive and bidirectional, so even though it sounds much slower, in reality AMDs fastest EV6 bus in 2003 was probably roughly equal to the fastest Intel had in 2008(better for some use, worse for some).

When AMD switched, their Hypertransport started out at 3.2GB/s. For a link that no longer needed to handle any transfers from the RAM. Today it runs at 12.8 up to 51.2GB/s.

Intel´s current QPI is roughly comparable(also in not being unidirectional and exclusive any longer(finally!)).


So in the end, the single most demanding transfer requirement of the FSB was removed, while what replaced the FSB has still been greatly speeded up.

Sure, gfx cards have speeded up and SATA and all, but it´s nothing even close to reaching the kind of bandwidth needed by the RAM memeory.

Hence, raising the speeds of the HT or QPI today becomes more or less irrelevant, as most things connecting through that link doesn´t need that much speed, nor does it have limitation issues that can be alleviated with greater speed ( like the old Intel FSB ).




So, need or not for FSB has nothing at all to do with any of the L1/L2/L3 caches. It has simply been replaced by a more dedicated link, while the RAM interface has been given it´s own separate link.

Talking about the cache structure is merely confusing and irrelevant.

Tenshinai wrote:

Spacekiwi wrote:I heard one guy use a restuaraunt as as an example of why pcs dont need a fsb anymore. he called the L1 cache a table you sat at, the L2 cache a serving trolley, and the L3 cache the maib buffet table in the middle of the room. the fsb was rollerskates with jets. he said it used to be you had food on your table, or he would skate to the kitchen and make you an order. that would take a minute, as opposed to half a second for having it on the table to start with. with the introduction of the L2, if it wasnt on the table, and it was on the trolley, slightly longer than from the L1, but quicker then going to the kitchen. L3, slower than L2, but quicker than kitchen. So it used to be 100% went through FSB at slow speeds. now, >80% goes through L1, >80% of remainder goes through L2, >80% of remainder goes through L3. result? going to the kitchen not required except in <1% of cases. so nowdays, improving the speed of the rockets does bubkus.....

Lol... Completely incorrect and totally confusing, but definitely funny description!


First of all, L1/L2/L3 are cache memory structures. If you go back to the -90s, then generally cpu´s had L1 and the motherboards had L2 and much less often L3.
Nowadays, most cpu´s have onchip cache from L1 to L3.

What cache memory DO, is to select data and instructions that are most used and store them, trying to keep delays for fetching stuff as low as possible.
The smaller the cache, the faster you can make it, so L1 tends to be tiny but have superfast transfer and reaction times while L3 tends to be large and relatively slow.


Now, what is/was FSB?
Oldstyle, the motherboard chipset consists of a northbridge and a southbridge.

Cpu connects to the northbridge, which is also what RAM and graphics(since AGP) connects to.

Southbridge connects network, PCI and ISA buses for addin cards, and it handles slow storage interfaces, IDE/PATA/SATA as well as USB, serial, paralell ports, keyboard and similar connections.

The FSB, is the link that connects the cpu to the northbridge.

So, EVERYTHING the cpu does, every piece of data that comes or goes to or from the cpu, regardless to or from where, goes through the FSB.

Now, a modern cpu changes things a lot.
Most importantly, it moves the RAM controller and connection onto the cpu itself. And this is THE main part of what the FSB needed lots of speed and bandwidth for.

And then, AMD switched to their Hypertransport in 2003 and Intel to their Quickpath Interconnect in 2008.

So, fastest Intel FSB ever used, managed 12.8GB/s(mostly theoretically and used in overclocked systems, like the one i´m typing this on). That was unidirectional and exclusive transfer, ie. the bus could only transfer a single packet of data at any one time and only in one direction. A severe drawback.

Fastest AMD used, was 3.2GB/s but was nonexclusive and bidirectional, so even though it sounds much slower, in reality AMDs fastest EV6 bus in 2003 was probably roughly equal to the fastest Intel had in 2008(better for some use, worse for some).

When AMD switched, their Hypertransport started out at 3.2GB/s. For a link that no longer needed to handle any transfers from the RAM. Today it runs at 12.8 up to 51.2GB/s.

Intel´s current QPI is roughly comparable(also in not being unidirectional and exclusive any longer(finally!)).


So in the end, the single most demanding transfer requirement of the FSB was removed, while what replaced the FSB has still been greatly speeded up.

Sure, gfx cards have speeded up and SATA and all, but it´s nothing even close to reaching the kind of bandwidth needed by the RAM memeory.

Hence, raising the speeds of the HT or QPI today becomes more or less irrelevant, as most things connecting through that link doesn´t need that much speed, nor does it have limitation issues that can be alleviated with greater speed ( like the old Intel FSB ).




So, need or not for FSB has nothing at all to do with any of the L1/L2/L3 caches. It has simply been replaced by a more dedicated link, while the RAM interface has been given it´s own separate link.

Talking about the cache structure is merely confusing and irrelevant.

Thank you for the update and clarification. As it is, I still stand by what I said earlier about what to purchase/spend your money on when buying a PC/building a PC. Buying the latest and greatest CPU will simply cost you far more than you'll gain.

You're better off buying more RAM, an SSD hard drive and making sure your BIOS is up to date. If you've the money left over after that, step up the video card a notch or two.

I leave video cards for last because, like CPU's, the cost is exponential. The latest and greatest I have seen is about $700, for a kick ass card. But a few steps down, you can get a real good one for $350.

Fireflair wrote:Thank you for the update and clarification. As it is, I still stand by what I said earlier about what to purchase/spend your money on when buying a PC/building a PC. Buying the latest and greatest CPU will simply cost you far more than you'll gain.

Oh certainly. I built this system just as Intel was busy promoting their next generation cpu:s, just a month after the first i7 got out.

And with a mild, 20%, overclock, this E8400 has shown itself to be just as good as i expected it to be. Even outrunning i5 cpu:s for years in some types of software due to the huge L2(6MB shared between cores) and 4 ALUs(Nehalem ( 2008 ) and derivatives had 3 up until Haswell came out in 2013, taking it back again to 4 ALU(and also added a 3rd AGU(adress generation unit))).

Sadly, Intel works hard at keeping people from truly exploiting this, reducing the number of cpu models at each level and keeping old cpu:s artificially expensive(VERY expensive, i had hoped to be able to upgrade this to a quad Q9650, but it´s still so expensive that i´m better off getting a completely new system with an i5).

And until AMD manage to get a good cpu back out for sale again, Intel sets pricetags hideously high.

Ack, how i miss the days of K7 and K8 success for AMD, so damn good that Intel was forced to both start serious R&D AND cut prices...

Fireflair wrote:You're better off buying more RAM, an SSD hard drive and making sure your BIOS is up to date. If you've the money left over after that, step up the video card a notch or two.

I leave video cards for last because, like CPU's, the cost is exponential. The latest and greatest I have seen is about $700, for a kick ass card. But a few steps down, you can get a real good one for $350.

You can get a quite excellent gfx card nowadays for much less than that.
My current gfx card is a AMD HD 6770 (with passive cooling), and it essentially provides me with all the 3D performance i could ever want.
And the pricetag for it was something more like $120-150.

The basic question you should use is first of all wether you actually use software that runs heavily in 3D, i see lots of people buying $200+ gfx cards, and then never run a single app or game that even begin to stress the 3D of the card.
They often could have used even just integrated gfx and be happy at the much cheaper system.
3D acceleration is useless for a 2D app or game after all.

Well, as an example....

GEFORCE GTX Titan with 6GB of GDDR5 is probably the single best performing card on the market. It also retails for almost a $1000.

Where as the GeForce GTX 760 with 2GB of GDDR5 is still quite a nice card, and retails for $250. It's also three or four models down from the Titan.

For my money, I'd got with the GTX 770 at $300. I run a multi-monitor setup, streaming video while running online games at max graphics. Driving 2 monitors off of one of these cards is quite possible, with out killing your FPS, because the card is quite nice.

I run 3 monitors, currently. So I run dual video cards. FPS stays high on all 3 monitors, regardless of what I'm doing. And it minimizes any loading on the CPU for video needs.

Of course, if you don't have a large monitor at 1080p (or better) resolution, you're not gaming or running any other video intensive programs, then you don't even have to spend more than $50 on a video card.

actually, the new graphics cards by amd are relatively cheap even over here now. a radeon r7 260x is at 220 or so, at launch prices. it will probably drop by 30 over xmas, giving a pretty good card a pretty good price. All the Nvidia cards are way overpriced here though. A 780 costs around $1k here.
But i'm quite impressed at how wel my intel hd 3000 does. I can play far cry 2 and borderlands 1 in 1024 by 720 with medium settings and get playable frame rate out of it. An HD 4000 is apparently 60% better, and compared to the HD 4000, the hd 4600 and 5000 are apparently 60% more powerful, so upgrading to a Hd 5000 would give me around 260% of my current performance, from a integrated graphics. Im happy with that.

Fireflair wrote:Thank you for the update and clarification. As it is, I still stand by what I said earlier about what to purchase/spend your money on when buying a PC/building a PC. Buying the latest and greatest CPU will simply cost you far more than you'll gain.

You're better off buying more RAM, an SSD hard drive and making sure your BIOS is up to date. If you've the money left over after that, step up the video card a notch or two.

I leave video cards for last because, like CPU's, the cost is exponential. The latest and greatest I have seen is about $700, for a kick ass card. But a few steps down, you can get a real good one for $350.

Unfortunately for me and my frineds, a 770 is $600 here. However, the new 280x is $450, and according to several tech sites, outcompetes it by a fair amount, So I know what side im sticking with for now.

although I would say that these days, you dont even need the $50 card, with how well the intel graphics performs.

I cant wait to see how well intels next gen graphics do.

Fireflair wrote:Well, as an example....

GEFORCE GTX Titan with 6GB of GDDR5 is probably the single best performing card on the market. It also retails for almost a $1000.

Where as the GeForce GTX 760 with 2GB of GDDR5 is still quite a nice card, and retails for $250. It's also three or four models down from the Titan.

For my money, I'd got with the GTX 770 at $300. I run a multi-monitor setup, streaming video while running online games at max graphics. Driving 2 monitors off of one of these cards is quite possible, with out killing your FPS, because the card is quite nice.

I run 3 monitors, currently. So I run dual video cards. FPS stays high on all 3 monitors, regardless of what I'm doing. And it minimizes any loading on the CPU for video needs.

Of course, if you don't have a large monitor at 1080p (or better) resolution, you're not gaming or running any other video intensive programs, then you don't even have to spend more than $50 on a video card.