Why not pay 1000 dollars for a CPU, yet we see TRI and QUAD SLI/XFIRE.

[lavaheadache]

Sorta wondering about this. What I said in subject...

A tri is gonna cost about avg 1k depending on video card model.

Imagine 3x 680 GTX. that is 1500 dollars.

My question or wonder is, why do we never see anyone with the top of the line chip. The 1k chip the E chip. Sandy E or Ivy E ........ and do a single video card setup or no more then SLI... I think most Anand peepz are gamers.....thx gl

I'm sure it more of a dimishing returns reason.

If $200 cpu's performed like $200 cpu's and couldn't come close to the performance of that $1k cpu, I'm sure there would be more $1k cpu's sold.

 

[tweakboy]

Amazing responses, thank you soo much guys. You are all lovely guru anand peeps.....

Yeah one thing tho, even tho you can OC and make it same speed as 1k chip. WHat about the 15MB cache compared to 12MB cache ..... Or that doesn't make much difference. thx

 

[AdamK47]

Thanks and good luck!

 

[Ajay]

I'd only build something like that if I had allot of cash burning a hole in my wallet. When F@H new fast return bonus kicks in for GPUs, it would be nice to have an x79 board and 4 GTX 680s folding. At present, my wife would faint just looking at the electric bill! Forget about the ~$3K cost of the system!!

 

[kache]

I'd only build something like that if I had allot of cash burning a hole in my wallet. When F@H new fast return bonus kicks in for GPUs, it would be nice to have an x79 board and 4 GTX 680s folding. At present, my wife would faint just looking at the electric bill! Forget about the ~$3K cost of the system!!

You can always give her a present to compensate:
http://forums.anandtech.com/showthread.php?t=2287212

 

[Ajay]

You can always give her a present to compensate:
http://forums.anandtech.com/showthread.php?t=2287212

Yes, that does work wonders :thumbsup:

But I'm back to school for my MSCS and don't have the dosh to drop >$1K on her (which I've always been happy to do). It'll be a different story next Christmas, after I finish in May. My salary will be all gravy, since she's paying the bills right now. Most of it will go to savings and to the tax man, but we'll both have a nice Christmas in 2013

 

[kache]

Yes, that does work wonders :thumbsup:

But I'm back to school for my MSCS and don't have the dosh to drop >$1K on her (which I've always been happy to do). It'll be a different story next Christmas, after I finish in May. My salary will be all gravy, since she's paying the bills right now. Most of it will go to savings and to the tax man, but we'll both have a nice Christmas in 2013

Nice.

 

[taltamir]

I never got the whole "back to school" thing.
Nothing they sell is actually useful for education. In fact the vast majority of it counteracts learning... its like a "resume your diet with great prices on ice cream!"

 

[Spaceme1117]

Definitely because of overclocking.

 

[Ajay]

I never got the whole "back to school" thing.
Nothing they sell is actually useful for education. In fact the vast majority of it counteracts learning... its like a "resume your diet with great prices on ice cream!"

I'm finishing a Master's latter in life to improve my career prospects. I'm learning very useful things - I suppose it depends on the school you go to. Many job openings for software engineers lately are preferentially looking for Master's and PhD grads.

 

[taltamir]

I'm finishing a Master's latter in life to improve my career prospects. I'm learning very useful things - I suppose it depends on the school you go to. Many job openings for software engineers lately are preferentially looking for Master's and PhD grads.

I was clearly referring to advertisements for products being sold as a "back to school gear" not the education itself

 

[Ajay]

I was clearly referring to advertisements for products being sold as a "back to school gear" not the education itself

Sorry, it wasn't so clear to me, especially after the post I had just made. NP though.

 

[Idontcare]

I never got the whole "back to school" thing.
Nothing they sell is actually useful for education. In fact the vast majority of it counteracts learning... its like a "resume your diet with great prices on ice cream!"

Have you given equal consideration to the relevance of the commercial activity associated with Christmas, Halloween, Valentines day, Cinco De Mayo, St. Patrick's Day, Easter, July 4th, Black Friday, Cyber Monday, Mother's Day and Father's Day?

There is no shortage of retail-driven "holidays", back to school week is just another weekend that has been hijacked by commerce and it wouldn't actually happen if we didn't like it that way.

What is awesome here in PA is we have what is called "tax free weekend" where for one weekend every year there are zero state sales taxes on a whole list of items (clothing, school supplies, etc), sanctioned by the state government no less.

 

[taltamir]

Have you given equal consideration to the relevance of the commercial activity associated with Christmas, Halloween, Valentines day, Cinco De Mayo, St. Patrick's Day, Easter, July 4th, Black Friday, Cyber Monday, Mother's Day and Father's Day?

I actually have . And I find it silly as well.

What is awesome here in PA is we have what is called "tax free weekend" where for one weekend every year there are zero state sales taxes on a whole list of items (clothing, school supplies, etc), sanctioned by the state government no less.

tax holidays are actually tremendously stupid... If taxes are too high they should be lowered permanently, it discourages shopping on the rest of the year.

 

[Lepton87]

When I bought my rig there was no SB-E out or 28nm cards, so the only logical choice was 1155 SB. From a purely gamers perspective, paying 1k dollars for a CPU is illogical if you didn't already max out your GF set-up. 3960X(3970Xnow) with tri-fire 7970 versus 3930K with quad-fire 7970. The cost is about the same. I would vastly prefer the second rig.

 

[Concillian]

All about what you get for your money.

tri-fire / 3 way SLI is many times more powerful than a $250-400 GPU.

$1k CPUs are not many times more powerful than a $250-400 CPU.

In a case where you can get advantage from GPUs, they will make the most sense. This is also why the guys at my work who are doing physics modeling are doing it with GPUs rather than CPUs.

 

[Puppies04]

This can be answered with one simple question.

Can you find me a benchmark where a 1155 unlocked quad + GPU(s) results in unplayable frame rates while a 2011 hex + the same GPU(s) results in playable frame rates?

 

[Lepton87]

This can be answered with one simple question.

Can you find me a benchmark where a 1155 unlocked quad + GPU(s) results in unplayable frame rates while a 2011 hex + the same GPU(s) results in playable frame rates?

Maybe in another 5 years and I mean 4c/4t versus 6C/12T. 4c/8t should last quite a bit more then 4 threaded CPUs. When I bought mine over 1.5 years ago there wasn't a single game on the market that would benefit from additional threads. Someone quoted some Russian benchmarks that it is slowing begging to change.
Upon seeing those benchmarks once again, I revise my statement, those extra threads do squat. Easy to see comparing 760 to 930. 2600K is almost only faster then 2500k due to extra clock speed and cache. Maybe those games are extra sensitive to the cache size like WoW where 3.1GHz Xeon E-5 with 20MB of L3 outperforms 3770K. Also when we compare X4 to X6 there doesn't seem to be much difference in performance with 50% cores. I wouldn't draw any conclusions about the state of multi-threading from those benchmarks. They are just strange.

 

[BrightCandle]

This can be answered with one simple question.

Can you find me a benchmark where a 1155 unlocked quad + GPU(s) results in unplayable frame rates while a 2011 hex + the same GPU(s) results in playable frame rates?

Both will produce playable frame rates. But in some games the hexa core produces higher frame rates, and sometimes its the difference between 60 fps and 50 fps. Sometimes its also the difference between the dip down to 20 fps instead of 30 fps as a minimum. Most of the time Ivy bridge is the same performance or slightly better than Sandy Bridge E, it certainly can clock higher.

I don't think gaming is currently a compelling reason to get a hex core. The only cases I think hex core are worth it are:
1) You need lots of PCI-E bandwidth such that the additional slots are worth it. 4 GPUs it might be worth having, 2 certainly isn't.
2) You need more RAM than 16GB.
3) You have an application that is embarrassingly parallel and has a task you know will show +50% boast for the two additional cores.

Other than that its not honestly worth it.

 

[Idontcare]

This can be answered with one simple question.

Can you find me a benchmark where a 1155 unlocked quad + GPU(s) results in unplayable frame rates while a 2011 hex + the same GPU(s) results in playable frame rates?

I suspect the answer here lies beyond the simple statistic that a singular metric like average FPS will capture and convey.

Take the techreport's methodology of capturing and reporting the average frame rate for just the 99th percentile, or the time spent rendering frames that exceed 33.3ms.

These are ways to analytically capture and analyze game response factors that matter to the end-user above and beyond a flat "what is my average framerate?" because average fps does not speak to stuttering and lag in ways that the experiential result does.

I have no doubt that a hexcore is unlikely to improve on the gameplay from a flat average framerate perspective over that of a quad, but I have every expectation that a user playing a game on a hexcore is more likely to find the gameplay more fluid with less noticable random minimums in the framerate and so forth.

In many ways this is no different than what Anand uncovered and highlighted years ago regarding SSDs and the lockups they would do with random read/write of small files and so forth. That was a very real issue that was not captured by benchmarks designed to measure sequential bandwidth with large files.

For all the reasons that mattered for SSD users, I expect the same to manifest with gamers and their video cards versus cpu core count (since core count reduces contention on the backend where system processes are transpiring simultaneously).

 

[bunnyfubbles]

Both will produce playable frame rates. But in some games the hexa core produces higher frame rates, and sometimes its the difference between 60 fps and 50 fps. Sometimes its also the difference between the dip down to 20 fps instead of 30 fps as a minimum. Most of the time Ivy bridge is the same performance or slightly better than Sandy Bridge E, it certainly can clock higher.

I don't think gaming is currently a compelling reason to get a hex core. The only cases I think hex core are worth it are:
1) You need lots of PCI-E bandwidth such that the additional slots are worth it. 4 GPUs it might be worth having, 2 certainly isn't.
2) You need more RAM than 16GB.
3) You have an application that is embarrassingly parallel and has a task you know will show +50% boast for the two additional cores.

Other than that its not honestly worth it.

that's actually not exactly true

while Ivy should be able to clock higher, the non-solder TIM holds back Ivy's potential

my 3570K's temps start to skyrocket if I give the volts required to push it beyond the 4.5GHz I have it at

So while SB-E hexcores will produce a ton of heat that can be hard to dissipate with cheaper coolers (simply due to the sheer number of transistors it packs) Ivy will produce temps that can be even harder to dissipate because of that damn heatspreader TIM.

Basically, of all the Ivy chips I've worked with, they've all been performance side-grades because I couldn't clock them as high as all the Sandy chips I've replaced them with (its typically been ~4.5GHz avg vs. ~4.7GHz avg, including my SB-E), and thus the advantage comes from less heat/power/noise it takes to run Ivy.

 

[aaksheytalwar]

I suspect the answer here lies beyond the simple statistic that a singular metric like average FPS will capture and convey.

Take the techreport's methodology of capturing and reporting the average frame rate for just the 99th percentile, or the time spent rendering frames that exceed 33.3ms.

These are ways to analytically capture and analyze game response factors that matter to the end-user above and beyond a flat "what is my average framerate?" because average fps does not speak to stuttering and lag in ways that the experiential result does.

I have no doubt that a hexcore is unlikely to improve on the gameplay from a flat average framerate perspective over that of a quad, but I have every expectation that a user playing a game on a hexcore is more likely to find the gameplay more fluid with less noticable random minimums in the framerate and so forth.

In many ways this is no different than what Anand uncovered and highlighted years ago regarding SSDs and the lockups they would do with random read/write of small files and so forth. That was a very real issue that was not captured by benchmarks designed to measure sequential bandwidth with large files.

For all the reasons that mattered for SSD users, I expect the same to manifest with gamers and their video cards versus cpu core count (since core count reduces contention on the backend where system processes are transpiring simultaneously).

This

Every time I try to point this to somebody, nobody seems to believe me though.

 

[Lepton87]

I have no doubt that a hexcore is unlikely to improve on the gameplay from a flat average framerate perspective over that of a quad, but I have every expectation that a user playing a game on a hexcore is more likely to find the gameplay more fluid with less noticable random minimums in the framerate and so forth.

If that were true, that would also mean that Hyper-threaded i7s CPUs are better for games then I5s. So far, I only saw claims that HT can actually hamper the experience. But go ahead and buy a 3970X and do your meticulous testings and report your findings. I think that most improvements in fluidity over a same clocked sandy would be only due to the larger cache, which would be easy to find by locking affinity to only four physical cores. At the same clocks architectural improvements in IVY would easily offer more advantage than larger cache in SB-E but not always. 3.1GHz 20MB Xeon E-5 was faster in WOW then 3770k and desktop hex-core in tomshardware test.

 

[Idontcare]

If that were true, that would also mean that Hyper-threaded i7s CPUs are better for games then I5s. So far, I only saw claims that HT can actually hamper the experience. But go ahead and buy a 3970X and do your meticulous testings and report your findings. I think that most improvements in fluidity over a same clocked sandy would be only due to the larger cache, which would be easy to find by locking affinity to only four physical cores. At the same clocks architectural improvements in IVY would easily offer more advantage than larger cache in SB-E but not always. 3.1GHz 20MB Xeon E-5 was faster in WOW then 3770k and desktop hex-core in tomshardware test.

I have no idea how you got any of that out of my post.

When did hyperthreading come into the discussion?

 

[Puppies04]

I have no idea how you got any of that out of my post. When did hyperthreading come into the discussion?

I understand what he is alluding to, if you claim having 6 cores would reduce the random minimum frame rates due to having more threads available why would having 8 threads on a quad core not produce the same results when in has been shown in more than one situation to actually hinder FPS results. In a perfect world a hex would be better than a quad with HT which would be better than a quad with 1 thread per core but we aren't talking about a perfect world we are talking about game engines and windows task scheduling.

Edit. On a side note I understand why HT sometimes has a negative impact on games as you can end up in a bulldozer like situation where 2 threads are vying for the same core specific resources which was why HT was never really designed with gaming in mind.

I have no doubt that a hexcore is unlikely to improve on the gameplay from a flat average framerate perspective over that of a quad, but I have every expectation that a user playing a game on a hexcore is more likely to find the gameplay more fluid with less noticable random minimums in the framerate and so forth.

Are you talking specifically about games that scale past 2-4 cores or all games? Because if you aren't the only advantage 2011 chips have in that situation is increased L3 cache and greater memory bandwidth unless I am missing something. I think the point Lepton87 was trying to make was while you say the increased core count on a hex chip would be beneficial surely this is entirely dependant on the game in question.

As a side point, without messing around with core affinity for individual programs/games how can you be sure that windows won't try to run some background process on a lightly used core on a hex core at the same time a game decides it needs more system resources and cause a momentary amount of lag in game. Ok the law of averages dictates this is less likely on a hex core but it certainly wouldn't remove the possibility of it happening.