Depending on the amount of rendering, you could save a fair chunk of cash going for an i5.
In rendering an i7’s extra cache and the HT cores can make a substantial difference (10-25%), for single or weakly threading applications though not so much.
I’m wondering if I should be going with the i7 3930K or if I should be considering a Haswell chip instead…?
I don’t want to overclock anything.
If you do NOT want to overclock then DO NOT buy a K CPU. The K is for unlocked multiplier, AKA an OCing CPU.
Which to go for depends on price. Haswell is a tock (a minor improvement), but it’s also not much more expensive than an ivy bridge.
Personally, since there’s a socket change and for resale value and upgradability, given the small difference, I would consider a non K haswell if you don’t want to overclock, or a K if you decide to.
You don’t, in general, need water cooling. K series overlock fine by a couple notches on air with no side effects, and these days they are well protected. Even for more extreme clocking all-in-one solutions like the small Corsair H hardly make much of a difference.
On a non K CPU the stock Intel cooler is perfectly fine and has good noise levels.
If you decide to go for a K and bump it to 4Ghz a better cooler is indicated. If you decide to try and shoot for 4.2 and up, then water cooling, at least for noise levels, might be indicated.
In that case though make sure your case is well ventilated, as the move from on-socket air to liquid often takes away some cooling from the areas surrounding the socket.
Currently for non multithreading heavy the best bang for buck looking at the future and with high quality components in mind is probably an Asus z87 Plus with a Haswell i5
Bump that to i7 if HT is of interest (frequent rendering or encoding), and bump it again to a K if you change your mind and decide to overclock.
The memory is fine. These days you simply don’t overclock the base frequency, so 1600-1866 reputable brands are all you need regardless of OCing or not. Higher frequency memory at this point is as much of a con as mineral water in bottle
You could even go for the older vengeance 1600 cas9 and not notice a difference (personal experience). No harm in getting a 1866 though.
I don’t know if the 3930K is still worth its price, I can’t see it leaving a 4770K behind much even with the two added cores, not enough to buy old gen and completely kill any upgrade possibility for more money. That said, it will be far from slow, so if you have a good deal on one, go for it. Core for core Haswell is only very marginally faster at the same clock anyway, and there’s hardly any thermal difference.
Stock cooler that should come with it, or whatever intel sells as recommended, will be plenty, and you will probably get to bump the clock a little bit too even air cooled if you’re so inclined.
I probably still wouldn’t pick a 4770 over a 3930k despite it being a newer platform. Cost-wise I think you’re just getting what you pay for when you factor in the cost of the whole system. If you’re just looking at CPU and motherboard costs alone, then 4770 is better.
single threading performance is about 13-16% better than the 3930k while the 3930k is 25-35% faster than a 4770 at multithreading.
the 2011 socket isn’t dead yet. Isn’t that what the next gen xeons are using? The xeon sockets seem to change less often than the consumer sockets which have reliably changed every year. There’s a possibility you’d just need BIOS update and you could have up to a 12-core xeon to drop into a 2011 board
I think the sandy bridge-E chips were a mistake on intel’s part to begin with and just their way to deal with selling the xeons they were making that weren’t up to spec. I doubt we see anything more than 4-6 cores in the next couple years for the consumer grade chips. I know if I was forced to choose between a 4 or 6-core consumer chip (even with overclocking) and a 8-12 core xeon chip with the possibility for dual sockets, I personally wouldn’t hesitate to buy xeons.
That said, if intel continues to use the 1150 socket well into the future (doubtful), the chip you’ll upgrade to will likely still be a 4-core or 6-core chip.
You might be overrating the 3930K IMO.
Everywhere I’ve seen it benched it barely, and I mean just barely, scores a 3-8% better than the 4770 in the heaviest multithreaded benchmarks that aren’t extremely artificial and insular (including some relevant ones such as cinebench and various CPU pump ala SETI and Folding).
I don’t know over there, but here it’s almost twice the price of a 4770K (bought my 4770K for 264AUS$, about 250 US, the cheapest 3930K was 525 without cooler).
At that price you can buy the 4770K now, and a short while from now get a 4930 and resell and STILL come out on top or on par money wise and have much better future proofing and resale value.
Do you have anything around that actually shows the 3930k being 25-30% faster at multithreading in actual scenarios?
It would imply that the two additional cores are literally scaling linearly (something that never happens), and that in such case the individual cores line up in performance (also something that doesn’t happen when comparing Haswell and Ivy, however minor and “tock-y” Haswell is, not to mention the 4770 is faster to begin with).
The added cache, even if it wasn’t so frequently starved anyway, is mostly core split, so that’s just level ground.
I have trouble imagining the 3930 pulling ahead by such margin in any scenario that isn’t extremely artificial, or such scenarios being frequent enough to be relevant. I can believe it being, in practice, that 3, 5, even rarely 10% faster in some MT scenarios. 25-30? Nope, can’t imagine it.
The 3930k’s speed improvement gets more extreme when you factor in overclocking since the 3.5ghz 4770k tops out around 4.7ghz (1.2ghz improvement) while the 3.2ghz 3930k also goes to 4.7ghz (1.5ghz improvement or 25% more overclocking headroom)
I know cinebench isn’t the most perfect benchmark, but it’s what everyone uses as a measuring stick and is runs the same consistent scene. It depends on your software and scene files. Our maya mental ray scenes use a lot of SSS and approximation smoothing which can have a lot of single-threaded functions at render time making you lose the benefits of more cores. Cinebench multithreading is a near perfect multithreaded benchmark. It always kills me that more websites don’t also include the single-threaded benchmark because you’re otherwise not getting the whole story of how the chip performs.
I have a collegue who has a conservative 4.6ghz 2600k that renders our maya mental ray scenes about 68% as fast as our 4.7ghz 3930k machines. Given what we’ve seen from the 4770k’s performance, a 4.6ghz 4770k would probably render our files 77-78% as fast as our 3930k’s which is means the 3930k is still 28-29% faster overall.
The whole thing depends on what you need to buy to make your computer. If you’re just looking at the price of motherboard and CPU, the 3930k is a rip off since it’s twice as expensive. If you’re buying an entire system and going to have to spend $2000-3000 anyway, then going with a 3930k is about 10-15% more expensive for about a 25-30% rendering speed upgrade, but a minor 10-15% penalty in single-threading. IMO 10-15% is a barely noticeable difference.
3.2Ghz * 6 = 19.2
3.5Ghz * 4 = 14
A 3930 37% faster than a 4770 would imply a perfectly linear scaling.
Given all factors involved in managing the tasks and all, 30% performance gain is bleeding close to a linear scaling of the task, 35 would be better than linear unless you were literally just cranking the handle on the number crunching or something equally artificial.
Then consider it over the actual, effective amount benefitting MTing while waiting for traffic lights, lock release, forking and joining and sorting, and it’d mean not only a linear scaling of the activity by the core, but also that an older CPU would be faster, cycle for cycle, than a newer one, or at the very least evenly matched (which is disproven when tested core for core).
I know cinebench isn’t the most perfect benchmark, but it’s what everyone uses as a measuring stick and is runs the same consistent scenes.
I actually find it a much better bench than most, but it’s still miles away from actual cases IMO.
Both frybench and cinebench still give the 980x as faster than the 4770k though, and often enough faster than the 3930. In practice though I still have to see that 1 grand three and half years old CPU topping everything that ever followed the way many benchmarks would have it
At home though I get something in the low 9s (4770k) on cinebench (no idea how the reviewers are getting low 8s, I’ve seen more than one like that, and only one or two in the 9s), and in actual rendering tests the gap closes further from the only 3930 I have access to (Arnold, PRMan doesn’t count because it still scales like a dog).
I’m not calling you out on something or anything like that, just I still have to be impressed by the hexacore offerings, especially in value terms.
Benchmarking aside they don’t seem, in actual use, to be fast enough to warrant double or quadruple the price tag.
The whole thing depends on what you need to buy to make your computer. If you’re just looking at the price of motherboard and CPU, the 3930k is a rip off since it’s twice as expensive. If you’re buying an entire system and going to spend $2000-3000, then going with a 3930k is about 10-15% more expensive for about a 30% rendering speed upgrade.
I don’t disagree with that, the numbers are sound reasoning, I just disagree about the 30% in actual practice, and I think it a bit shortsighted when it comes to future proofing, not to mention it leaves out all things that are not intensive rendering which will be a lot more comfortable on a 4770K.
If I had to put a home render mule together for a quick project and following resale, I would consider a 3930 for sure, and I would still consider one if I found a particularly good deal for it.
For a longer term investment though I honestly can’t recommend it, it’s just both terrible value for money and only narrowly advantaged in rendering, and disavantaged in everything else.
oh ok, I missunderstood what you meant by “linear scaling”
the 3930k and 4770k are two levels apart in architecture so of course there isn’t a direct linear correlation between the two since there are no 6-core haswell chips yet, and none planned for the next 1.5 years.
When not overclocked, everything is a different story and I’d probably pick a 4770k over a 3930k in that case, so I have to agree with you on that. Low 9’s is pretty good cinebench for a 4-core chip at stock speeds. It’s odd that everywhere else seem to report low 8’s. Maybe you’re doing some minor overclocking?
Anyway, you have a 4770k which is a great chip that just recently came out. It has a brand new socket and features, with an obvious lead in single-thread performance over other existing chips.
I’ve had a 3930k at home for over a year now, and also have a renderfarm of them at work. I get a 13.62 in multithreaded cinebench and 1.84 in single-threaded cinebench. I obviously overclock.
4 core haswells are certainly better than a 3770k or 2600k. The 2.5 year old 2600k could do 5-5.2ghz and almost keep up with an overclocked 4770k, achieving low 10’s in cinebench. It surpassed overclocked 3770k’s even though the 3770k was better at stock speeds. It’s only recently with the new 4770k, IMO that the 2600k has finally been soundly beaten across the board.
I think what appealed to me the most about the 3930K system I built a year ago was the value proposition. What you are essentially buying is a Xeon processor without the kid gloves on it. Spending hundreds of dollars for thousands of dollars of performance is a win-win situation to me. On a whim I can click a button and have a Cinebench X-CPU score over 14; when I want a whisper quiet system that sips power I can run it so slowly my fans all turn off or become inaudible. It’s a very flexible system and configurable CPU that you can’t get now with the 4xxx series of processors, since Intel, with its dominant market position, sees fit to further stratify product ranges by artificially crippling certain features. I’m also not interested in paying extra for the built-in graphics chip on such processors. I can’t be sure how much of a problem it is not having ECC but I needed a powerful system without the “pro” price and so really had little choice, but so far I don’t think it has affected the system stability.
As has already been mentioned, there’s a potential upgrade path that doesn’t exist with the mainstream consumer line. The trend is that CPUs for consumers will eventually come soldered to the motherboard and purchased as a unit; this is already happening and I see no reason for Intel to stop that any time soon as it saves them money and shifts the burden of warranty charges further on to the shoulders of the motherboard manufacturers. It’s the same reason why Intel started using lands instead of pins - they didn’t want to incur the costs involved with broken or bent pins on their processors.
Arghhh, sorry, I lied horribly, however unintentionally. It’s the 4670k I got for 265, the 4770k I paid 370 for.
Motherboard is an Asus z87 pro, which for the record I’m hella impressed with and wouldn’t hesitate recommending.
Online I’m finding somewhat horrible prices the last couple weeks, e-Bay has particularly gone to crap.
Recent bout of shopping I did at the Capitol Square centre on George st. Just know you have to haggle and pitch them against each other every inch of the way, but you can get good prices that way. Ironically, most of the shops there, after you repeatedly haggle them out of a few bucks (possibly angrily) will become a lot friendlier.
It’s largely a cultural thing, but the first few times I went there I used to get lots of cold shoulders and the listed price. After a few times you walk out without buying anything (and make a point of slowly walking by with another shop’s bag later) an odd form of respect starts coming through. If they can do it quietly they will often enough sell at cost anything that isn’t selling like hot cakes if they know you are, otherwise, a lost sale.
For the memory right now only MSY has some decent enough stock variety.
The plus is apparently just as good, just misses the wi-fi and has a slightly shoddier audio unit. Ethernet chipset might also be one notch down, but that would make bugger all difference.
If you care for neither consider it, might save you another 30 bucks. Read some reviews though and make sure it has the same TPU/EPU management and overclocking capabilities (which to my knowledge it should).
There are a few spots where brick’n’mortar stores actually are on par or better than online, and therefore better if you want to browse, haggle or ever lay claim to a warranty without spending more in shipping and rolling the dice with RMAs and tests.
You also save shipping costs (which when free are simply offset somewhere else).
Even the few online shops that are truly competitive often have a store-front for order pick-ups and you end up ordering online only because they happen to be in another state.
Capitol Square in particularly is literally 30 shops crammed into a very small space owned by a dozen different small enterprises who fiercely compete with each other. You very rarely pay more than 1 or 2% more than you would online there, frequently I found things for less.
Honest online-only shops in Australia are few and far apart, most of them are very, very dodgy forelogging pre-pay, foreign owned front-ends where “in-stock” means you’ll get it once they have twenty items ordered by different buyers and can place a bulk order with no storage/backlog risks.
This is very peculiar of Australia.
I lived in several different countries, and only here in Sydney I found such situation.
I live in the UK where we have very few shops that sell a wide range of components, and those that sell components at all often have a horrendous mark-up. For me, online is the only choice. Luckily we have very consumer-friendly laws when it comes to online shopping, meaning for example you can return any product within a week, no questions asked. I do a lot of research before buying so fortunately returns of a faulty product are very rare.
Yeah. London was the polar opposite of Sydney for computer parts shopping.
Local Asian owned shops in the cbd here are frequently the best deal, with rare exceptions for a couple Queensland and Melbourne retailers that have online outlets.