Cores and threads?

[CliveG]

Can someone explain for the less technically minded about some of the logic behind what seems to be the advice here that a good I7 should be a better investment for 3d and Cad work than a dual xeon?

 

My old I7 is about to be upgraded, it has I believe 4 cores and provides 8 buckets when rendering (8 threads?).

 

I'm interested in a dual xeon 6 core arrangement, thinking that this will provide me with multiple buckets when rendering and I am lead to believe cooler running CPUs. Is this right would I get 12 threads per processor so 24 buckets when rendering and can 8 buckets in a new I7 then outstrip the performance of those 24 buckets?

 

Heat has been a big issue for me with the I7, I'm on my 3rd mobo and 2nd heat sync. I thought that getting threads and lower heat would be a better choice, or have I got it completely wrong?

 

Any advice for the technically challenged appreciated!

 

Cheers

[numerobis]

There are some other aspects to consider beside cores and threads...

Especially multi-CPU xeons can have very different clock speeds - from 1,7GHz to 3,5GHz with different turbo speeds for the Ivy Bridge Xeons ( http://en.wikipedia.org/wiki/List_of_Intel_Xeon_microprocessors#.22Ivy_Bridge-EN.22_.2822_nm.29_Entry_2 )

And the high clocked Xeons are very expensive ($2000-$3000 for one CPU). The i7's are all running in the 3,2-3,7GHz range, with all core turbos up to 3,7GHz and single core up to 4,0Ghz.

So depending on the apps you're using the performance can be very different. This applies to multi core performance and maybe even more to single core performance - which is very important for workstation use where most tasks are still not multithreaded (almost anything other than rendering and simulation).

 

And there is another point to take into consideration... the IPC (Instructions Per Cycle) of a processor. ( http://en.wikipedia.org/wiki/Instructions_per_cycle )

At the same clock speed a Haswell based i7 has a ~5-7% higher IPC than Ivy Bridge for instance (maybe even a bit more compared to Ivy Bridge -E). The new Xeon E5-V2 are based on Ivy Bridge-E like the i7 48xx/49xx. And the i7 47xx quad cores are based on Haswell.

 

So concerning your i7 quad vs dual hexa core Xeon comparison... the 12 cores (24 threads) will be faster for rendering even if you take the smallest, 2,1GHz hexa core xeons. But the single thread performance of the i7 at 3,9GHz will be much higher (IPC + clock speed)

But you're comparing one quad core to two hexa cores here... There are also hexa core i7, like i said above (i7 39xx/49xx) which would be the more appropriate i7's for such a comparison...

 

So coming to your very first question... It all depends on your needs and on your budged. If you want a high multicore performance in one box and you're willing to spend several thousend dollars only for two Xeon processors then this can be the right choice.

But if you want a best bang for the buck with the highest possible single thread performance then nothing beats a quad or hexa core i7 (depending on your budged). And even more if you consider to overclock your system to 4,5GHz or more - there simply is nothing comparable at the moment on the Xeon side, because the current Xeons are locked and can't be overclocked.

And for the budged of one dual Xeon setup you can build several i7 (maybe overclocked) systems for network rendering - if this is an option and you don't want it in one box.

Edited September 10, 2013 by numerobis

[CliveG]

Wow thanks for a very comprehensive answer!

 

Any thoughts on the heat output, is it fair to think that the xeon s will produce less?

 

Thanks again

[numerobis]

Basically Xeon and i7 are the same silicon (within the same CPU generation). But there are always chips on a wafer that need less voltage or will stay cooler than others. So it is possible that for example the low voltage Xeons are selected chips. But if you have to combine two of them to get the same power like a higher clocked i7 i don't think this will help much in terms of energy consumption and heat.

And if they are clocked similar i don't think there will be much difference.

[CliveG]

Thanks again for taking the time to provide such thorough answers.

 

I think as the only work I do which requires outright power is rendering, then subsequently the Xeons aren't a bad option for me, 24 buckets is very tempting.

 

Also the price differential certainly isn't 2 or 3 I7's to one dual Xeon here in Aus (Instead they're all expensive!) so I7's aren't perhaps the bargain in comparison - for me anyway.

 

Thanks again.

[Dimitris Tolios]

Some things mentioned many times before around this is:

 

1) Most programs are NOT multithreaded, or only certain portions of them are.

24 and 36 "buckets" sound fun and all, but for most CG Architects, rendering is only a small portion of the actual production workflow.

Meaning you spend more of hours modeling and testing and setting up something, than actually rendering the final image.

And before you hit the "render" button, almost all of the operations are single threaded = rely on a single, fast thread, regardless of you having a quad core, a quad core with 8 threads or a 2P Xeon...you will be using one and only thread (at least your main app i.e. 3DS will).

 

2) Xeons "don't run cooler" comparing apples to apples.

Xeons don't run cooler "per core / per GHz". Xeons might be running cooler when you under-utilize their assets (as explained above, you have 6-8 cores and you are using 1x), but this is not happening without a sucrifice: intel designed them to be running @ full speed and remain within specs @ full speed. Thus, the maximum core speeds of Xeons with more cores are slower, as they have to have "more cylinders" firing while producing the same heat as their "i7" counterparts.

 

3) Temperature doesn't mean the chip dies or wears out prematurely.

Reciting Led Zeppelin: lots of people talking and few of them know.

 

4) 6C/12T are not created equal...

Those E5-2620s or w/e people are thinking are a great deal for hex-cores...well...those are pretty slow...in fact a single FX-8350 is already 66-67% as fast as a couple of 2620s together when rendering. And you can have an FX 8350 + basic mobo for $250 US or so....while a 2P s2011 mobo alone is $450~500, and the pair of 2620s another $900. Mediocre investment.

 

On the other hand, we have a 6C/12T like the 3930K and soon the 4930K.

3930Ks routinely overclock above 4.5GHz (my 3930K is @ 4.9GHz with water cooling). Usually you can do 4.0GHz without Vcore boosts. That is enough for a single 3930K 6C/12T to match the performance of 2x E5-2620s in rendering, while blowing them off in single threaded contenct creation (At 4.5~5GHz ofc it is even faster). A 3930K, a mobo capable of boosting it @ 4.5GHz easily and a CLC cooler than can cool it (A sub $100 CLC like the H100i is "ok" for 4.5 in the 70s. Lower temps means either lower clocks, or much much more expensive water cooling solutions.) will set you back a bit less than just the 2620 pair, saving you a 2P mobo and a pair of coolers.

Xeons don't overclock.

 

5) Turboboost evolves with generations.

Haswell > Ivy > Sandybridge > Westmere. Each generation turboboosts differently, with more cores / under heavier load capable to do it with every generation. Thus Haswell (4770K) is 3-4% faster than Ivy bridge (3770K) with turboboost disabled (clear IPC advantage is 3-4%), but with turboboost enabled it is more than double that - it just turboboosts itself more effectively throughout. Otherwise the 2x i7s have identical clocks (base = 3.5 + turboboost = 3.9).

Edited September 10, 2013 by dtolios

[komyali]

can somebody put list of xeon processors that support dual or quad cpu configuration

If I overclock my i7 (noctua cooler) is it going to short CPU life or create any other problem ?

[AJLynn]

Overclocking can cause overheating, and overheating can shorten CPU life. But it takes an unusually high temperature to cause damage. (You can find a lot of opinions on the web about what a safe temperature is, I'm not making any endorsements.) You have a CPU cooler upgrade, which is good. Also make sure that there are working case fans and room for air to move in the case and outside where the in and out air flow is.

 

As for Xeons, the list is long, but it's something like this:

 

E3-XXXX, E5-1XXX: Single CPU only

E5-2XXX: Up to two CPUs

E5-4XXX: Up to four CPUs

E7: These are mostly for server use, I think they go up to 8 per motherboard

[Dimitris Tolios]

can somebody put list of xeon processors that support dual or quad cpu configuration

If I overclock my i7 (noctua cooler) is it going to short CPU life or create any other problem ?

CPUs don't die with high temperatures nearly as easily as some people imply. Actually most of them cannot die or really degrade due to temperatures at all.

 

Overheating prevention methods are built into modern CPUs from both AMD and intel, and the CPUs will simply throttle down (decrease Vcore + frequencies) if temperatures go past a certain threshold.

 

Intel's threshold for throttling is above 100oC.

What would stop them putting it @ 90 or even 80oC if that was a real issue?

Is it a conspiracy to "kill your CPUs" faster for you to replace them?

 

What demonstratively kills or degrades CPUs fast is very high V-Core for prolonged time.

 

There are easy to follow "rules of thumb" for maximum voltage used for each architecture / nm manufacturing process, above which it is not recommended to run your CPU for 24/7.

 

Those settings are pretty conservative, and not even close to the "harmful" levels crazy or careless over clockers use to break records. Most of these would require liquid nitrogen or extreme combinations of phase change or TEC + water cooling to be sustained anyways.

 

An example for Ivy Bridge (i.e. 3770K & 3570K)

 

I would say 1.4V for Sandybridge / 1.25~1.3V for Ivy + Haswell are pretty "safe" on air and CLCs (plz, don't confuse the closed loop coolers - i.e. CLCs - like the H100 with full-blown water cooling, those barely match or beat high end air coolers like the Noctua or Thermalright twin towers).

[komyali]

tnx for fast reply, by my logic, number in front tells you how many cpus you can connect, ah I found at last "Max CPU Configuration"

http://www.intel.com/content/www/us/en/processor-comparison/comparison-chart.html

 

So you think I need to overclock my i73930K (noctua cooler bigest one) how for to go I am scared I am looking for 2-4 cpu configuration because in vray precalculation primary and secundary bounce like light cache and Irradiance map sometimes take time and you cant render them on render farm because they doesnt support net rendering, so that is the only reason I am looking for 2-4 cpu configuration. I dont know maybe new vray 3.0 support net rendering but what I saw till now it is 2 times faster than vray 2. Also does vray Indirect ilumination GI support multiple cupu configuration ? If chaos group solve this problem I would buy new 3930K

[Dimitris Tolios]

Been running my i7-3820 @ 1.425V / 36x125 = 4.5GHz all its life under a SilverArrow SB-E, topping 80oC in my un-conditioned appartment with no issues. Was nearly 100% load all the time (Folding@Home).

 

The replacement, my current 3930K, was also under air (same cooler) @ 4.7GHz with 1.425V, would be in the 83oC range full load.

 

Currently under a $350+ water loop, it does much lower temps, but i hope you see the pattern.

NO MERCY within reason!

 

Will it degrade? Could be the case. So what does that mean? How fast?

Had been overclocking for ages. My dad has my previous desktop with a clocked Q6600 pushed all its life...like what? 6-7 years? (guess never more than 1 core under load, he just browses).

 

I don't believe you can kill a CPU with a "mild" OC and proper cooling within any timeframe that would beat the obsolesence of the technology (i.e. would you really be using your 3930K in 10 years from now?)

Edited September 11, 2013 by dtolios

[komyali]

You've convinced me, I will overclock it but not now I dont have time for that, So you say overclocked @4.7GHz working 2 days at full load wont create problems if cooler is good(noctua nh-d14 se2011)?

[Karl Larsen]

The simple fact that you MUST add more cooling to an overclocked chip should be enough to tell you something, and pumping water through a computer is a short-circuit waiting to happen...

 

I prefer the comfort of knowing that if, God forbid, anything ever went kooky with my chip, I'm not going to have to lie to the manufacturer and say I didn't overclock it, presuming it is still under warranty.

 

And with the speed of chips today, what is the point really? if you are rendering a large scene and you're impatient, send it to a render farm - if you are trying to be Industrial Light and Magic on a workstation, well you have more problems than overclocking is going to solve

 

Overclocking can cause overheating, and overheating can shorten CPU life. But it takes an unusually high temperature to cause damage. (You can find a lot of opinions on the web about what a safe temperature is, I'm not making any endorsements.) You have a CPU cooler upgrade, which is good. Also make sure that there are working case fans and room for air to move in the case and outside where the in and out air flow is.

 

As for Xeons, the list is long, but it's something like this:

 

E3-XXXX, E5-1XXX: Single CPU only

E5-2XXX: Up to two CPUs

E5-4XXX: Up to four CPUs

E7: These are mostly for server use, I think they go up to 8 per motherboard

[Dimitris Tolios]

You've convinced me, I will overclock it but not now I dont have time for that, So you say overclocked @4.7GHz working 2 days at full load wont create problems if cooler is good(noctua nh-d14 se2011)?

 

Go 4.5...it is still great improvement (1.3GHz above base speed or more than 40%, don't be greedy).

I would say if you are lucky and your chip can do taht with less than 1.43V you will be golden. Temperatures less than 85oC are "easy", no worries.

Remember tho, that the power consumption and heat generation will sky-rocket, more than doubling the 130W TDP rating...

My CPU @ 4.7 and 100% utilization would draw around 310W or power (plug load, kill-a-watt measurement minus PSU efficiency and 20-30W for the peripherals/fans,HDD/SSD etc).

[komyali]

definetly overclock these xeon processors are too expencive, I can build 2 rigs i7 for price of one xeon cpu... dont want to watch them anymore, till I get rich xD

[AJLynn]

I'm all for fast clocks, but if this is your first time overclocking a box do it in steps. Base clock is 3.2, try running a temp monitor while rendering for 20 minutes (turbo will mess with the clock speed, but don't worry about that yet). Then jack it up to 3.6 and repeat, 4.0 and repeat, 4.3 and repeat, 4.5 and repeat. If CPU temp gets much above some threshold (decide what this is going to be by surveying forum comments by people who have the same CPU) something's going wrong - check for airflow obstructions, check that your fans are spinning, make sure the CPU cooler is really well correctly seated, etc.

[numerobis]

As Dimitris said, what lets a CPU degrade faster is more a too high voltage than a high temperature. At high temps the system will first become instable or begin to trottle.

But concerning save values for voltages there are also some other values to watch besides the vcore, especially the vccsa and vtt for the Sandy Bridge-E. There have been reports about some dead 3930K in the past which seem to be a result of too high vccsa (and vtt) settings produced by the auto overclocking features of the boards! ( http://rog.asus.com/forum/showthread.php?19926-3930K-C2-Degradation )

So this is something to watch and to be save i wouldn't trust in the auto overclock, because normally the voltages are set too high - not only in this special case of the vccsa but also the vcore or vtt which might be not that dangerous but will result in higher temps (and power consumption). So it would be better to set them manually or at least to control the settings.

I try to keep my vcore under 1.375v and the vccsa at 0.95v. (air cooled and water cooled)

Edited September 11, 2013 by numerobis

[komyali]

maybe it is better to overclock my old i7 920... so if I failed I still have better one

[AJLynn]

Don't worry too much, just proceed a bit conservatively until you develop some familiarity with it and you'll be fine. This is waaaaay easier than it was a few years ago. It used to be that the multiplier was locked and the front side bus speed wasn't, and you'd need to use above-spec RAM and overclock the RAM to make the CPU overclock. There was a lot more to get wrong.

[numerobis]

some infos for Sandy Bridge E overclocking...

 

http://www.overclock.net/t/1189242/sandy-bridge-e-overclocking-guide-walk-through-explanations-and-support-for-all-x79-overclockers

http://archive.benchmarkreviews.com/index.php?option=com_content&task=view&id=834&Itemid=38

 

btw... the i7 920 is more difficult to overclock

Edited September 11, 2013 by numerobis

[Dimitris Tolios]

The simple fact that you MUST add more cooling to an overclocked chip should be enough to tell you something, and pumping water through a computer is a short-circuit waiting to happen...

 

I prefer the comfort of knowing that if, God forbid, anything ever went kooky with my chip, I'm not going to have to lie to the manufacturer and say I didn't overclock it, presuming it is still under warranty.

 

And with the speed of chips today, what is the point really? if you are rendering a large scene and you're impatient, send it to a render farm - if you are trying to be Industrial Light and Magic on a workstation, well you have more problems than overclocking is going to solve

Overclocking adds performance that is un-available to your given budget or even the industry in general.

The performance attainable through overclocking a 3xxx/4xxx CPU, will be available to consumers "out of the box" in 2-3 years - in some cases maybe more.

 

Companies know that before hand, but they won't release 4.5+ GHz chips, if the market "allows" them to milk consumers for 2-3 years by gradually introducing revisions of pretty much the same thing (what Sandy Bridge, Ivy Bridge and Haswell are as an example, and the intel "Core" architecture in general which is still based on principals set during the design of the Pentium III architecture).

 

Also, as they try to hit multiple birds with one stone, companies design for a given thermal density that is based on requirements set by large computer companies, datacenter specs etc. You cannot introduce 200-250W parts, when the industry has set a cap @ 130-150Ws, and really prefers sub-100W for consumer PCs.

 

As far as water cooling goes, yes, there are issues and things that need extra attention, but nothing is "waiting to happen" necessarily.

Unless you stop yourself from driving or riding in a car (the #1 way do die, far more dangerous than any mass-transit medium), taking a bath in your bathroom (the #1 place for serious or fatal accidents in your house) etc etc.

 

Also, your reasoning around "reliability" and extra heat within the CPUs would probably fail within Intel's and AMD's marketing departments, just like with nVidia etc, as all these companies not only provide products that are marketed as "overclocking friendly", but also guidelines and even special warranties for overclocking enthusiasts, much like car companies offer "official tuning packages" with warranties on some of their vehicles.

 

That doesn't make factory tuned EVOs more appealing to the masses than the Camry or the Corolla, but for some is a legit option, and for few a "necessity", despite the shortcomings, the dangers, the lower reliability, the increased running and maintenance costs.

 

The average reliability remains pretty high, water cooled PCs are not for everyone, but CLCs (which are liquid coolers, even if I look down upon them in my previous post) are extremely effecting for their volume, and trusted by both intel and AMD which have introduced factory branded CLCs, as well as respected boutique manufacturers like BOXX and even Dell (alienware) which routinely use them in their pre-overclocked (read "factory") systems.

 

So, yeah, personal preference and personal fears can keep you off doing some stuff - that is understandable. But neither the overclocking-phobic or the overclocking enthusiast can be dogmatic about what boils down to be an opinion.

 

I have been fiddling and building my own PCs the better half of my life.

I have done silly things. Never killed a chip with overclocking, even going back to Pentium 133MHz being pushed to 200MHz for all their life with no cooling capacity increase.

 

Does that make it a "law"? Does it guarantee anything?

No. You just read it in a forum.

Still more "real" than many of the "stories someone I know" get transmitted through broken telephones daily all around.

 

Anything that plugs into a wall plug can die at any time, overclocking involved or not (I actually have lost PCs to PSUs blowing up, no clocking involved), just like being on the street or even your house kitchen might kill you.

C'est la vie.

Edited September 12, 2013 by dtolios

[Karl Larsen]

Overclocking adds...

 

I am glad that all works for you but I already have more than enough to think about each day and overclocking just isn't going to make enough of a difference in my life to get me all crazy about doing it!

 

Back in the olden days when PC's were slow - maybe...

[AJLynn]

Actually my home PC is a bit underclocked. Right now I've got it optimized for low power consumption. (But that's one thing that's cool about the current hardware - you can have a low power PC and if something comes along and you need to use your home box for a render, reboot and change a few things and get 30% more speed.)