More Cache - What good is it?

All about them.

Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee

nutball
*Lifetime Patron*
Posts: 1304
Joined: Thu Apr 10, 2003 7:16 am
Location: en.gb.uk

Post by nutball » Thu Apr 15, 2010 1:51 pm

ces wrote:You may be able to read and interpret the benchmarks. I have tried and have not been able to.
:?:

:?

The answers are in this thread and the other threads you've started here about your micro-techno-angst. In very plain language as far as I can discern.

The benchmarks you see aren't relevant to your use-case. They aren't intended to be relevant to your use-case, the reason being that your use-case is almost entirely trivially satisfied by any semi-competent modern hardware configuration. That includes all the configurations you're considering and many more cheaper too. There's no point benchmarking your use-case, for this reason [just like everyone stopped benchmarking 2D VGA performance a decade ago]. This has been stated in this thread. By several people.

For your use-case you are most likely to be I/O-bound, and mildly I/O bound at that. This has been stated in this thread along with a solution - spend your money on an SSD not on a CPU, if you feel you have to spend your money at all.

So, please excuse me, but can I turn the question around. What answer are you actually looking for? Would you be happy if we all just clubbed together and made up an answer for you? We can probably all manage to sound authoritative if you want.

Sorry if I'm coming across like an arse here, but you really do seem to be being almost intentionally filtering out answers that don't back up some unstated pre-conceptions you have about what you want to buy.

Mats
Posts: 3044
Joined: Fri Dec 26, 2003 6:54 am
Location: Sweden

Post by Mats » Thu Apr 15, 2010 3:57 pm

I think you should use the Q9550, unless you have sold it. SPCR's undervolted system used 117 W with all four cores at full load, and it never exceeded 95 W during benchmarking. Imagine how often that kind of situations would happen to you, probably rarely going above 80 W with something similar to SPCR's system. If you don't think that's enough you can always underclock a bit.

I'm just saying, yes, a dual core would be enough and use less power, but if it's not that easy to sell, keep it and use it

The test system:
# Intel Core 2 Quad Q9550 processor - 2.83 GHz, 45nm, 95W
# Asus P5Q3 motherboard - P45 chipset
# Zerotherm Zen FZ120 CPU cooler with Nexus 120mm fan
# Corsair XMS3 DHX memory - 2x2GB, DDR3-1600 @ 1333MHz, 9-9-9-24
# Asus EN9400GT Silent Edition graphics card - 512MB
# Western Digital VelociRaptor hard drive - 300 GB, 10,000 RPM, 16MB cache, SATA
# Seasonic SS-400ET ATX power supply

Mats
Posts: 3044
Joined: Fri Dec 26, 2003 6:54 am
Location: Sweden

Post by Mats » Thu Apr 15, 2010 4:19 pm

Microsoft Security Essentials is probably ok, but I haven't seen any signs of it being fast. Do you guys have a link to some comparative review?

Avira is free, three times faster and detects more malware.
It also happens to find a bit more false positives but that's ok with me, it didn't detect more more of those on my HDD than any other AV I've used.

ilovejedd
Posts: 676
Joined: Tue Oct 21, 2008 4:14 pm
Location: in the depths of hell

Post by ilovejedd » Thu Apr 15, 2010 4:26 pm

Mats wrote:Microsoft Security Essentials is probably ok, but I haven't seen any signs of it being fast. Do you guys have a link to some comparative review?
Lol, the report before that. :P
http://www.av-comparatives.org/images/s ... port24.pdf

ces
Posts: 3395
Joined: Thu Feb 04, 2010 6:06 pm
Location: US

Post by ces » Thu Apr 15, 2010 5:21 pm

nutball wrote: So, please excuse me, but can I turn the question around. What answer are you actually looking for? Would you be happy if we all just clubbed together and made up an answer for you? We can probably all manage to sound authoritative if you want.

Sorry if I'm coming across like an arse here, but you really do seem to be being almost intentionally filtering out answers that don't back up some unstated pre-conceptions you have about what you want to buy.
I have no preconceptions. I am just trying to learn. And I am learning.

I didn't mean to irritate you in the process. But it is important to me not only to know the answer, but how it was arrived at. What may be simple and obvious to you, is not yet so for me.

Well maybe I am filtering out things. If I am unable to understand the basis of a recommendation I need to continue to probe it until I do understand it.

Instead of being given a fish, I want to learn how to fish - sort of. Is it that bad that I am trying to learn and understand.

So far what I have learned is that there are three key factors to CPU performance:
1. GHz
2. Cache
3. Number of Cores


They each contribute different attributes to the performance of the computer. And I am beginning to get a sense of what each does and doesn't do.

For each person talking here, there are probably a hundred lurkers, some of whom are learning along with me. Show a little patience for those less knowledgeable than you.

Jay_S
*Lifetime Patron*
Posts: 715
Joined: Fri Feb 10, 2006 2:50 pm
Location: Milwaukee, WI

Post by Jay_S » Thu Apr 15, 2010 6:37 pm

ces wrote:Instead of being given a fish, I want to learn how to fish - sort of. Is it that bad that I am trying to learn and understand.

So far what I have learned is that there are three key factors to CPU performance:
1. GHz
2. Cache
3. Number of Cores

They each contribute different attributes to the performance of the computer. And I am beginning to get a sense of what each does and doesn't do.
1. Some slower-clocked CPUs outperform faster-clocked CPSs.
2. Some CPUs with less cache outperform CPUs with more cache.
3. Some CPUs with fewer cores outperform CPUs with more cores.

I don't think CPU performance can be reduced to bullet points. Please ignore these three attributes for the time being and read up on basic CPU Microarchitecture concepts (Wikipedia).

ces
Posts: 3395
Joined: Thu Feb 04, 2010 6:06 pm
Location: US

Post by ces » Thu Apr 15, 2010 6:54 pm

Jay_S wrote: 1. Some slower-clocked CPUs outperform faster-clocked CPSs.
2. Some CPUs with less cache outperform CPUs with more cache.
3. Some CPUs with fewer cores outperform CPUs with more cores.
That I understand. And I understand that performance has multiple measures. And each of the attributes contributes different things to different measures. And those relationships are not necessarily simple. And there are other variables as well.

I just went and looked at that link. That is too much for me to grasp and turn it into a practical understanding of what to expect out of any particular CPU. It seems to me that the nature and detail of that information, by itself, is not sufficient for anyone to make practical application of it.

The missing information is the interrelationships of these component pieces of knowledge - combined with knowledge of how various specific software programs are programmed to work.

One important piece of info is whether and under what circumstances, if any, a faster CPU has any effect whatsoever on speeding up disk access. And how different programs running on different cores interact and perhaps even interfere with each other when accessing the disk.

At some speed 0.0001 Mhz CPU speed is important. But perhaps in the GHz range, even with an SSD it is of little consequence. It wonder what the point is?

judge56988
Friend of SPCR
Posts: 455
Joined: Sun Jul 27, 2008 10:39 am
Location: England

Post by judge56988 » Thu Apr 15, 2010 9:36 pm

Ces, maybe you should be asking these questions on a more general site? There are plenty of them out there.
I'm pretty sure that there must be some information on the web that can explain the mechanics of a cpu in a basic way; how about this to start with?

I don't imagine that anyone here begrudges giving you advice on silent computing or indeed on one or two general issues, but what you seem to want is for someone to spend a good chunk of time writing what amounts to a beginners guide to the cpu. That's too much to ask and it's part of the reason why you are getting a negative reaction from some people.
If you want to learn rather than just follow the advice, you have to be prepared to do the legwork yourself.

ces
Posts: 3395
Joined: Thu Feb 04, 2010 6:06 pm
Location: US

Post by ces » Thu Apr 15, 2010 9:55 pm

judge56988 wrote:Ces, maybe you should be asking these questions on a more general site? There are plenty of them out there.
I'm pretty sure that there must be some information on the web that can explain the mechanics of a cpu in a basic way; how about this to start with?

I don't imagine that anyone here begrudges giving you advice on silent computing or indeed on one or two general issues, but what you seem to want is for someone to spend a good chunk of time writing what amounts to a beginners guide to the cpu. That's too much to ask and it's part of the reason why you are getting a negative reaction from some people.
If you want to learn rather than just follow the advice, you have to be prepared to do the legwork yourself.
This is a unique site. I think the people here are fairly smart and thoughtful people. That is not a plentiful commodity on the internet.

Using as little computing power to accomplish as much as you can is related to silent computing. It seems evident to me that there are a lot of people out there using more CPU than they need, with applications that are running slower as a result (think of a slow quad vs a fast dual core).

Certainly no is forced to express their opinions on this thread. I don't demand it of any one.

When I can, I help people here as best I can. It is voluntary. As is any commentary anyone might have for me.

Anyone contributing to this thread should only do so if it feels good. That would not stop me trying to help them on other things that I might be able to.

I have learned a lot out of this thread and you and the others for your consideration and time in assisting me in that education. I will try to return that goodwill elsewhere in these forums on subjects that I am a bit more knowledgeable about.

So thank you for your time and consideration.

judge56988
Friend of SPCR
Posts: 455
Joined: Sun Jul 27, 2008 10:39 am
Location: England

Post by judge56988 » Thu Apr 15, 2010 10:05 pm

ces wrote: This is a unique site. I think the people here are fairly smart and thoughtful people. That is not a plentiful commodity on the internet.
That's why this is the only site I go to on a regular basis - and thank you for your excellent input on fans.

Jay_S
*Lifetime Patron*
Posts: 715
Joined: Fri Feb 10, 2006 2:50 pm
Location: Milwaukee, WI

Post by Jay_S » Thu Apr 15, 2010 10:28 pm

ces wrote:I just went and looked at that link. That is too much for me to grasp and turn it into a practical understanding of what to expect out of any particular CPU. It seems to me that the nature and detail of that information, by itself, is not sufficient for anyone to make practical application of it.

The missing information is the interrelationships of these component pieces of knowledge - combined with knowledge of how various specific software programs are programmed to work.

One important piece of info is whether and under what circumstances, if any, a faster CPU has any effect whatsoever on speeding up disk access. And how different programs running on different cores interact and perhaps even interfere with each other when accessing the disk.
The meaning is in the markup. "Microarchitecture" is not presented by itself; the "missing information" is just a hyperlink away.

From one of your posts in this thread:
If you are thinking about multitasking, a clock cycle is a clock cycle. Maybe there is a slight difference between a one core chip and a two core chip. But not between a 2 core and a 4 core - and the i3-530 with multi-threading is more like a 3 core chip.
Clock for clock, the Athlon64 was faster than the Pentium 4. Clock for clock, Intel's Core2 microarchitecture was faster than Athlon64. And clock for clock, Core i# is faster than Core2. These are radically different microarchitectures. To a lesser extent there were minor changes within: moving from Conroe through Wolfdale, for example, SSE4 debuted with Penryn. Making comparisons on the basis of core count, cache size, and clock rate - while ignoring microarchitectural differences - will only hinder your understanding.

I cut my teeth on the Pentium Pro in the mid 90's for a research paper in college. At that time, I could spell "computer" and play Doom. So the learning curve was steep.

I guess what we're all saying is, at some point you gotta jump in the pool. Yes, it will suck in the beginning, but you'll be forced to learn how to swim.

Core concept: programs don't run on cores:
http://msdn.microsoft.com/en-us/library ... 85%29.aspx
http://en.wikipedia.org/wiki/Thread_%28 ... science%29

There have been several new posts while I've been composing this, lol. I'll end with perhaps the wisest thing anyone ever said to me WRT compute performance. This was my Cisco CCNA instructor, when asked why he still used an ancient PC: "As long as it's faster than I can type, it's fast enough for me."

ces
Posts: 3395
Joined: Thu Feb 04, 2010 6:06 pm
Location: US

Post by ces » Fri Apr 16, 2010 10:11 am

If you are thinking about multitasking, a clock cycle is a clock cycle. Maybe there is a slight difference between a one core chip and a two core chip. But not between a 2 core and a 4 core - and the i3-530 with multi-threading is more like a 3 core chip.
I have been around longer than you and I remember the big schtick about not all clock cycles being equal and how AMD thumped that drum - you can pick away at bits like that - the main thrust of my questioning was that they are not equal - but in what manner are they not equal. And really how much difference in a clock cycle within a core among all the 775 dual cores or even quad cores. And even with the i3-530, how many % more punch does one of its core clock cycles have? Maybe 20-30%? Maybe not?

Trying to prove my ignorance on this subject isn't difficult. Bit picking away to prove it??????? Why?

The best way to test someone's expertise is to ask questions, if they stumble and can't explain the what and why of what their assessments - if they are not able to explain the what and why simply. Well they are not quite the masters of the domain they claim to be.

Not being able to explain something in a simple and cogent manner. Then getting gruff and sending them out on a journey to learn it on their own, has a certain defensive "Wizard of Oz" element to it - in my opinion anyways.

Vicotnik
*Lifetime Patron*
Posts: 1831
Joined: Thu Feb 13, 2003 6:53 am
Location: Sweden

Post by Vicotnik » Fri Apr 16, 2010 10:57 am

ces wrote:I have been around longer than you and I remember the big schtick about not all clock cycles being equal and how AMD thumped that drum - you can pick away at bits like that - the main thrust of my questioning was that they are not equal - but in what manner are they not equal. And really how much difference in a clock cycle within a core among all the 775 dual cores or even quad cores. And even with the i3-530, how many % more punch does one of its core clock cycles have? Maybe 20-30%? Maybe not?
CPUs are very complex and the only way to answer that question is to read a lot of reviews to get sort of a sense of what CPU is best for what kind of task.

What are you looking for exactly? Are you trying to determine what CPU to buy or are you looking to know more about CPUs in general?

ces
Posts: 3395
Joined: Thu Feb 04, 2010 6:06 pm
Location: US

Post by ces » Fri Apr 16, 2010 12:07 pm

Vicotnik wrote:CPUs are very complex and the only way to answer that question is to read a lot of reviews to get sort of a sense of what CPU is best for what kind of task.
I think you are right. The problem is that the reviewers are interested in writing about only certain aspects of interest to their readers - mostly enthusiasts interested in very specific types of performance scenarios.
Vicotnik wrote:What are you looking for exactly? Are you trying to determine what CPU to buy or are you looking to know more about CPUs in general?
I am interested in both and perhaps helping to educate others like me. I think I now have enough information to make my purchasing decision. I am still interested in this subject and anyone who has anything to contribute to it.

lm
Friend of SPCR
Posts: 1251
Joined: Wed Dec 17, 2003 6:14 am
Location: Finland

Post by lm » Fri Apr 16, 2010 12:20 pm

ces: Car analogy - you are getting a shopping bag for an 1km grocery store trip twice a week and you are analyzing horse power, torque, and even timings of the cylinder valves.

ilovejedd
Posts: 676
Joined: Tue Oct 21, 2008 4:14 pm
Location: in the depths of hell

Post by ilovejedd » Fri Apr 16, 2010 1:21 pm

ces wrote:I think you are right. The problem is that the reviewers are interested in writing about only certain aspects of interest to their readers - mostly enthusiasts interested in very specific types of performance scenarios.
There's a reason for that - they'll be the most affected by changes in CPU performance.

Again, current processors are fast enough that even buying a $50 Celeron E3300 or AMD Athlon II X2 245 would be more than what regular users would need. For light tasks (I think the ones you mentioned can be categorized as this), there wouldn't be any noticeable difference between a Celeron E3300 or a Core 2 Duo E8400.

There's a reason why the Atom is still around despite the abysmal performance - it works fast enough for most people. That should at least explain why most tech sites cater to the enthusiast crowd in their reviews.

ces
Posts: 3395
Joined: Thu Feb 04, 2010 6:06 pm
Location: US

Post by ces » Fri Apr 16, 2010 2:05 pm

ilovejedd wrote:
ces wrote:I think you are right. The problem is that the reviewers are interested in writing about only certain aspects of interest to their readers - mostly enthusiasts interested in very specific types of performance scenarios.
There's a reason for that - they'll be the most affected by changes in CPU performance.
Of course there is a reason for it. That doesn't alter the fact that it causes certain aspects of performance to be covered, and certain other aspects of performance to be disregarded.

grazzhoppa
Posts: 27
Joined: Sun Apr 01, 2007 1:31 pm

Re: More Cache - What good is it?

Post by grazzhoppa » Fri Apr 16, 2010 2:46 pm

ces wrote:What attributes of performance does extra cache help?
  • Extra cache on the CPU is only useful for programs that repeatedly do [a few types of computation] on [very limited amount of sets of data] in a [very short period of time].
    In the realm of consumer-level software, these are virus scanners, file archivers (WinZip, WinRAR, and the like), real-time games, photo manipulation filters, rendering 3D images, rendering antialiased vector images (PDF, Flash). Any program that hits the CPU hard with a limited amount of code will benefit from extra cache.

    That said, extra cache in the real world does not help much at all in typical multitasking environment i.e. the "office worker" or "family" or "personal use" OS environments.
    • The cache is constantly getting overwritten by the many programs that are allowed to run their code on the CPU. In a multitasking environment, by the time a program gets to run its code on the CPU again, any cache of its code or data from its pervious run has most likely been overwritten by the multitudes of other processes/threads/OS-kernel that had been given a chance to run between the program's two runs.

      Extra CPU cache can really shine on single-use environments like web-servers or file-servers that run only the essential processes/programs that will fulfill their function as a server. There is less "cache thrashing" or overwriting of the cache in these environments.
Here are Newegg's Dual Core CPU prices:
E3300 Wolfdale 2.50GHz 1MB L2 Cache $51.99
E5400 Wolfdale 2.70GHz 2MB L2 Cache $69.99
E6500 Wolfdale 2.93GHz 2MB L2 Cache $79.99
E6600 Wolfdale 3.06GHz 2MB L2 Cache $97.99
E7400 Wolfdale 2.80GHz 3MB L2 Cache $124.99
E8400 Wolfdale 3.00GHz 6MB L2 Cache $167.99

1. Low Cost - What uses would cause me to not choose the e3300?
  • As far as how much L2 cache it has, there is no reason to avoid it because it has the least amount of cache.

    But it does have the slowest clock speed, which is somewhat significantly lower than the average clockspeed (~2.9GHz) in that lineup of CPUs. But if you take into account the huge price in crease for higher clock speeds, then the e3300 seems to be the best value in that lineup.
2. More Cache - What uses would cause me to choose the E8400 for the extra cache? How much extra $ is extra 5MB pf cache worth?
How about the E7400? How much extra $ is extra 2MB pf cache worth? What good is it really?
How about the E5400? How much extra $ is extra 1MB pf cache worth? What good is it really?
  • Running a single-purpose machine that has little to no direct interaction with human beings (i.e. a server).
    In a heavy multitasking environment, that 5MB of extra CPU cache is worth jackshit.

    The differences in performance in a heavily multitasking environment between a Core 2 Duo CPU with 1MB and another one with anywhere between 2MB-6MB is negligent. It really shouldn't be a factor in your purchasing decision.
3. More GHz - What uses would cause me to choose the E6500 for the extra GHz? How much extra $ is the extra 0.43GHz worth?
  • If you multitask with many real-time programs, then that extra 430MHz may make a difference. Nowadays most people do.
    For example, you could be playing a video in your webbrowser, while it runs website advertisements using Flash animation, while you are involved in a VOIP phone call or a video chat, while you are downloading files that your virus scanner is verifying in real-time that they are not malicious, while playing music softly in the background, while scrolling through a PDF document, and periodically exploring your file system.
    • I remember the days when an extra 50Mhz in CPU speed could allow me to play an MP3 file without it skipping. Those days are long gone, but if you do saturate the CPU with enough real-time programs, you may see unsatisfying results like the inability for some programs to keep up in real-time.
    deciding [2.5GHz for $52] vs [2.93GHz for $80] is a gray, muddy process. In my opinion the 50% increase in price for an extra 430MHz is not worth it; the performance jump going from 2.5GHz to 2.93GHz will not be an ostentatious display of power.
Like other people have said, I suggest getting the cheapest quad core available today. Unfortunately since you're boxed into an Intel platform, the quad cores available to you are hideously more expensive compared to the quad cores (and triple-cores) AMD offers. You don't need "extra cache" nor "extra mega/gigahertz" - for your operating environment, you'd find the best value in "more cores" among today's CPU offerings.

Now that I think about it, for the last few years I've made my decisions to buy a CPU based on power consumption/transistor tech. and feature set (# of cores, virtualization support, hardware acceleration (SSE, AES acceleration)).
  • CPU speed and cache size are the absolute last things I use to decide between CPUs nowadays.

ces
Posts: 3395
Joined: Thu Feb 04, 2010 6:06 pm
Location: US

Post by ces » Fri Apr 16, 2010 8:10 pm

grazzhoppa - Wow. I am going to have to come back and read your message a few more times. It seems to contain a lot of practical wisdom.

danimal
Posts: 734
Joined: Mon Jun 08, 2009 2:41 pm
Location: the ether

Re: More Cache - What good is it?

Post by danimal » Fri Apr 16, 2010 8:57 pm

grazzhoppa wrote:Like other people have said, I suggest getting the cheapest quad core available today. Unfortunately since you're boxed into an Intel platform, the quad cores available to you are hideously more expensive compared to the quad cores (and triple-cores) AMD offers. You don't need "extra cache" nor "extra mega/gigahertz" - for your operating environment, you'd find the best value in "more cores" among today's CPU offerings.

Now that I think about it, for the last few years I've made my decisions to buy a CPU based on power consumption/transistor tech. and feature set (# of cores, virtualization support, hardware acceleration (SSE, AES acceleration)).
  • CPU speed and cache size are the absolute last things I use to decide between CPUs nowadays.
that could be a sensible approach for people who don't have a need for serious computing.

but there is no way that amd can deliver the horsepower that i need, so i abandoned the platform a couple of years ago.

i agree that spending more for a few gigahertz of stock speed doesn't make much sense, because overclocking is so easy these days, at least for intel cpus... you should be buying the slowest cpu, in the class you want, i.e., i7 920 over i7 930.

ces
Posts: 3395
Joined: Thu Feb 04, 2010 6:06 pm
Location: US

Re: More Cache - What good is it?

Post by ces » Fri Apr 16, 2010 10:06 pm

danimal wrote:that could be a sensible approach for people who don't have a need for serious computing.
What does "serious computing" encompass and what does it not encompass?

danimal
Posts: 734
Joined: Mon Jun 08, 2009 2:41 pm
Location: the ether

Re: More Cache - What good is it?

Post by danimal » Sat Apr 17, 2010 11:51 am

my i7 920 effectively has 8 virtual cores, it will do everything that he listed there with ease... that's not serious computing.

you'll know that you are doing serious computing when you see the cpu utilization of all 8 of your cores pegged continously at 80-100%, with the cpu speed manually overclocked by, say, 25-35% or so.
grazzhoppa wrote:For example, you could be playing a video in your webbrowser, while it runs website advertisements using Flash animation, while you are involved in a VOIP phone call or a video chat, while you are downloading files that your virus scanner is verifying in real-time that they are not malicious, while playing music softly in the background, while scrolling through a PDF document, and periodically exploring your file system.

ces
Posts: 3395
Joined: Thu Feb 04, 2010 6:06 pm
Location: US

Re: More Cache - What good is it?

Post by ces » Sat Apr 17, 2010 1:09 pm

danimal wrote:my i7 920 effectively has 8 virtual cores, it will do everything that he listed there with ease... that's not serious computing.

you'll know that you are doing serious computing when you see the cpu utilization of all 8 of your cores pegged continously at 80-100%, with the cpu speed manually overclocked by, say, 25-35% or so.
If I have learned anything here, it is not that simple. For instance. for some applications an i5-670 running at its stock 3.46Ghz will handily outperform an i7 920 running at its stock 2.66GHz. How about Gaming?

And I don't think that there are many applications which i5-670 overclocked to 5.5Ghz is not at least as fast, or faster than an i7 920 running at stock 2.66GHz. Can the 920 be overclocked. Sure. But at around 4GHz you start to be able to cook eggs on it.

At least that's what it appears to me seeing benchmarkers using that to stress CPU heatsinks. Take the same super coolers that xbits uses to reach 4GHz on its 1366 socket testing system and it seems not unreasonable for that same cooler to support 5.4 to 5.5 Ghz on the 670. (I will admit that maybe this balance of horsepower changes if you are talking about water cooling - but that is going a bit too far off the bell curve for me)

So I guess if you are doing some heavy encoding or some such, a 920 with all its cores cooking has marginally more horsepower. Even for users of this site, despite all the talk about high end computing, that represents only the smallest sliver of time to which they put their computers.

That seems to make the 1366 chips look as practical as a 600 hp street Ferrari. Sure you can rationalize why you need it. But unless you have a convenient race track - it serves no purpose. And if you do have one, there are dedicated racing machines (ie database servers) that will kick its but.

An over-muscled over-heated quad seems, for almost all non-commercial users, to serve more of an emotional need than a practical need. And that is likely to remain so until the majority of commonly used software is reprogrammed to take advantage of multiple cores.

I think what I have just said is consistent with the learning espoused in this thread. I would never have imagined saying any of the above before the start of this thread. But it is hard to think otherwise after absorbing the knowledge laid out here.

Am I drinking koolaid, or is the above valid thinking?

danimal
Posts: 734
Joined: Mon Jun 08, 2009 2:41 pm
Location: the ether

Re: More Cache - What good is it?

Post by danimal » Sun Apr 18, 2010 2:30 pm

ces wrote: If I have learned anything here, it is not that simple. For instance. for some applications an i5-670 running at its stock 3.46Ghz will handily outperform an i7 920 running at its stock 2.66GHz.
and in some applications the i7 will blow it away... but it's a pointless comparison, because stock clock with one type of cpu vs. overclocked in another model of cpu serves no practical purpose.
ces wrote:So I guess if you are doing some heavy encoding or some such, a 920 with all its cores cooking has marginally more horsepower. Even for users of this site, despite all the talk about high end computing, that represents only the smallest sliver of time to which they put their computers.
translation: you have never had a need for serious computing.

if you had, it would all become very clear, lol

~El~Jefe~
Friend of SPCR
Posts: 2887
Joined: Mon Feb 28, 2005 4:21 pm
Location: New York City zzzz
Contact:

Post by ~El~Jefe~ » Sun Apr 18, 2010 3:08 pm

All chips are faster in every way with a larger cache

the only reason they dont put big ones in is because they want to screw the consumer and make a huge profit

Every notch bigger in cache sizes equates to more performance. If a piece of programing/instruction whatever can fit in the cache and not be called from memory, it speeds up. the larger the cache the larger the TYPE of thing can be in there and not yanked from ram.

Gaming rocks with extra cache. I wish AMD would give bigger caches on the x3-4 and new x6's. it does help.

ALSO, you can write your software to take a big leap of performance if you know the chip has a lot of cache in it.

the 1366 socket is the real socket and is also made better for business and industry. there are no QPI links on 1156. It's a homebrew socket, thats about it. The 920, now 930, can handle certain things much better than the 1156 860, but hardly anyone around utilizes the chip and motherboard potential in a home setting.

My next rig is an unlocked 3.2ghz 6 core amd anyways :)

ces
Posts: 3395
Joined: Thu Feb 04, 2010 6:06 pm
Location: US

Post by ces » Sun Apr 18, 2010 5:04 pm

~El~Jefe~ wrote:All chips are faster in every way with a larger cache

the only reason they dont put big ones in is because they want to screw the consumer and make a huge profit

Every notch bigger in cache sizes equates to more performance. If a piece of programing/instruction whatever can fit in the cache and not be called from memory, it speeds up. the larger the cache the larger the TYPE of thing can be in there and not yanked from ram.

Gaming rocks with extra cache. I wish AMD would give bigger caches on the x3-4 and new x6's. it does help.

ALSO, you can write your software to take a big leap of performance if you know the chip has a lot of cache in it.

the 1366 socket is the real socket and is also made better for business and industry. there are no QPI links on 1156. It's a homebrew socket, thats about it. The 920, now 930, can handle certain things much better than the 1156 860, but hardly anyone around utilizes the chip and motherboard potential in a home setting.

My next rig is an unlocked 3.2ghz 6 core amd anyways :)
I read something interesting, I think at the anand site. My recollection is that it was in interview with a an Intel chip designer.

Bottom line is the reason the i7 line has less cache than the Q9000 series, is that they found they were losing too many clock cycles accessing larger cache. There is a diminishing benefit and what they are trying to do is find ways to reduce the clock cycles it takes to access cache, thereby permitting them to use more without diminishing returns.

I am certain the impact of the cache must vary with the application. And while you can never be certain what the software is doing, certainly grazzhoppa's analysis makes logical sense. If the software keeps accessing the same routines over and over, of course cache will help. Is it doesn't, well how could the cache be of any help.

The best way to determine if software will be helped is to test the software with large cache and then with small cache, on the same series of chips (that would need to be a low end 775 chip and a high end 775 chip as all the i7 chips have the same cache). At least I can't think of any better way to determine that.

ces
Posts: 3395
Joined: Thu Feb 04, 2010 6:06 pm
Location: US

Post by ces » Mon Apr 19, 2010 4:11 am

~El~Jefe~ wrote:My next rig is an unlocked 3.2ghz 6 core amd anyways :)
I am thinking I want to see what the new Sandy Bridge is or isn't. Then get a dual chip CPU that I can overclock the heck out of. It might be the Sandy Bridge, or it might be the new i5-680 Clarkdale.

It appears that for my use, that is going to provide the crispest user experience.

~El~Jefe~
Friend of SPCR
Posts: 2887
Joined: Mon Feb 28, 2005 4:21 pm
Location: New York City zzzz
Contact:

Post by ~El~Jefe~ » Mon Apr 19, 2010 5:12 pm

I can't stand the idea that every intel setup i have cannot be ever upgraded.

if i was SMART i would have gotten the black edtion 720 triple core from AMD instead of this e8400 chip. This cant be upgraded. ram cant be either without downclocking it severely so its just a thing that sits there.

plus, the board was like 119 dollars. I know that isnt a lot but the same board with a few more ports like 1394a would be around 79 dollars for amd. I would then just snap in my 6x core amd.

I can't imagine how snappy that will be as a windows7 or linux experience. I could even have gotten an am2+ setup and walked it for 3-4 years up through the 2010 series chips. Rumor has it am2+ can use 6 core!

yuu
Posts: 132
Joined: Sun Aug 26, 2007 5:03 pm
Location: eu

Post by yuu » Mon Apr 19, 2010 10:42 pm

Intel made a mistake by giving 6MB L2 cache to E8400 and it clocks to 4.0Ghz, it is just too good to need a replacement yet and for the next 2 years.

Of course with the with the unlocked 6-core amd you get 3 times better performance in rendering, cost the same, on 50$ Asrock motherboard, and in my case I have to just add the DDR2 8GB, as for DDR3, I can't begin to imagine how much it costs now days. Wow, intel price politics gets more and more inadequate. That E8400 should cost 50$ and be called E5200 or probably i2-200.

alecmg
Posts: 204
Joined: Thu Mar 13, 2008 5:56 am
Location: Estonia

Post by alecmg » Tue Apr 20, 2010 3:05 am

My rule of thumb is
Cache is good for gaming
Nr of cores is good for encoding
GHz is good for responsiveness.

Lets not forget memory either, mem frequency/timings are important in all three :) Thats why I'm quite sad that todays notebooks are 95% same slow memory setup.

Memory amount only matters when its not enough.

Post Reply