Core i7-3960X Extreme Edition vs. i7-990X Extreme Edition Benches Revealed

[Florinmocanu]

well, keeping the same 6 cores/12 threads and going 15% faster is for me a good thing.

My profession as a cg artist usually involves lots of FX work.

This is an explosion i created 2-3 months ago in my spare time

http://www.youtube.com/watch?v=3T2qX9IEiSA

It involved 200GB of simulation data, around 10-12 hours of total simulation time and then of course render time, which was around 11-12 hours (4 minutes per frame at around 200 frames, but the first ones where faster than 4 minutes, since 4 minutes where when the explosion is full on.

so that would amount to around 23-24 hours of computational time done on my 4Ghz 920. If i would have had a 15% faster CPU, that 24 hour time would go to around 20 hours, so 4 hours gained.

In my industry, gaining 4 hours means either faster productions or more detail in the same time frame, either outcomes are really good.

So, 15% is definitely really good for me, but i really need all the power in the world. But for a gamer, doesn't justify getting a 6core SB-E vs a 2600k.

[cx-ray]

In a professional working environment I'd rather upgrade to 1366 hex cores if you're still on quad cores. Switching to a completely new platform on initial release could potentially turn into a huge time sink. 1366 is pretty much proven now and would give you an instant ~50% boost.

Also, I'm not sure how much credence we should give to the Intel rumors. It would indicate that they would rather stick to their planned release schedule over delivering a complete product with next gen features. I don't see the point of that kind of time pressure, considering their market position in the high-end desktop market...unless the problems are much larger than I'm assuming they are.

[Florinmocanu]

For my type of work there is no downtime in switching platforms, all you need is more CPU power. so 15% more SB-E is definitely better. Depends on business and enviroment, i can imagine other work enviroments are more costly to changing the whole platform.

[Mungri]

And you can pick up a 6 core CPU if you ever need more cores.

If the prices ever come down, which I've been waiting ages for. It usually happens when the next gen stuff is released.

The I7 970 is still far too expensive, you can get a whole new sandybridge CPU + Motherboard for less cost.

[Macadamia]

well, keeping the same 6 cores/12 threads and going 15% faster is for me a good thing.

My profession as a cg artist usually involves lots of FX work.

This is an explosion i created 2-3 months ago in my spare time

http://www.youtube.com/watch?v=3T2qX9IEiSA

It involved 200GB of simulation data, around 10-12 hours of total simulation time and then of course render time, which was around 11-12 hours (4 minutes per frame at around 200 frames, but the first ones where faster than 4 minutes, since 4 minutes where when the explosion is full on.

so that would amount to around 23-24 hours of computational time done on my 4Ghz 920. If i would have had a 15% faster CPU, that 24 hour time would go to around 20 hours, so 4 hours gained.

In my industry, gaining 4 hours means either faster productions or more detail in the same time frame, either outcomes are really good.

So, 15% is definitely really good for me, but i really need all the power in the world. But for a gamer, doesn't justify getting a 6core SB-E vs a 2600k.

You could also just add render nodes- get a separate PC with a 2600k, speed your render time up to 75% instead of 15%, and can use commodity parts such as $99 Z68 motherboards, just 4GB of RAM etc because it's a render slave.
The saving grace is that you're using default scanline instead of a commercial renderer such as v-Ray or Brazil which charges by nodes (but the minimum license normally covers >2 PCs anyway)


As for 2011, I think it's a big joke as of now- especially since Z68 stuff has become cheap and proven, and you get an unlocked multiplier for less money. Remember how long it took for a decently priced X58 to show up? And AFAIK it was from MSI/Foxxconn instead of Asus/Gbyte.

[Florinmocanu]

You could also just add render nodes- get a separate PC with a 2600k, speed your render time up to 75% instead of 15%, and can use commodity parts such as $99 Z68 motherboards, just 4GB of RAM etc because it's a render slave.
The saving grace is that you're using default scanline instead of a commercial renderer such as v-Ray or Brazil which charges by nodes (but the minimum license normally covers >2 PCs anyway)


As for 2011, I think it's a big joke as of now- especially since Z68 stuff has become cheap and proven, and you get an unlocked multiplier for less money. Remember how long it took for a decently priced X58 to show up? And AFAIK it was from MSI/Foxxconn instead of Asus/Gbyte.

You cannot simulate fumefx on 2 pcs at the same time, you can simulate separate jobs ofc, that's another deal. of course it's good to have 1-2 slaves chugging the renders along your main workstation, but for a main workstation a SB-E looks really good.

[Macadamia]

going OT here, but have you tried Phoenix FD?
Given how it's from Chaos they'll make it really fast and efficient- I didn't use FumeFX much but I disliked Afterburn throughly.
Most particle-based sims should be required to distribute workloads- Realflow does that real good. If it only runs on 1 node it's normally a sign that you should stay away lol.

[Florinmocanu]

FumeFx is kind of the industry standard. I tried Phoenix FD, really good, but not there yet to the maturity/strength of FFX. you can render it multiple nodes, through backburner but actual simulation is one pc only, for consistency reasons.

[R101]

15% is not enough for upgrade (in my opinion), but if you are buying new server, more performance for the same amount of money is always a good thing .

[Tenknics]

Well im coming from a laptop and I want a hex and I need a hex. I need more cores and ram for my video editing work.

I personally can't invest in a 2+ year old platform when a brand new one is right around the corner. Im tempted to get a 2600k but I need a hex...Thats why sandy-e looks so good to me. Anyone wanna buy my laptop?

I've read Main stream Ivy's will be pin compatible with the current z68 boards/ be s1155. Has there been any info out about the Ivy-E's being s2011? or?

[zalbard]

Has there been any info out about the Ivy-E's being s2011? or?

We don't know anything about it yet...

[informal]

So on average 13% faster on desktop than 3.47Ghz Wetemere 6C,according to intel(disregarding physics). This puts it at around 257pts in hardware.fr chart,dangerously close(1% difference ) to what I speculated some time ago.

[Metroid]

I have seen many posts complaining about every new Intel architecture generation since Conroe, but what boils down to Intel is the competition in which has not caught up yet despite rituals and curses done. The truth is, this is just another cash cow and an excuse that they could not progress much further.

Me neither, I am sticking to 1366 until AVX is actually used.

Yeah, I don't see a reason either to upgrade as soon as it comes out.

[Macadamia]

True that, but take into perspective that it looks bad because Nehalem was workstation first, consumer later. And the dual core Arrandales had a hampered memory controller whilst P55 didn't have x16/x16; so 1366 was still a viable choice in 2009/10.

Sandy Bridge already had such strong consumer parts with little compromise, that's the thing.
The workstation part isn't really workstation grade, more like catered to the upper server segment where more cache and bandwidth for multi-socket systems would be useful.

Keep in mind too that this starts from a higher pricing level too when basically most of the 1366 platforms were 920/30/50/60 at <$350. Now you pay that price, and still no hexacore.

[rintamarotta]

Why not get Evga SR-2 and two X5690 Xeon's for rendering and simulation?
Would beat any i7 2600k and i7 990X machine easelly and simulation time would be cut to allmost half.

Its true that it is bit expensive platform but you're industry needs that kind of power, like you self said you can add eather more effects or cut time.

[Florinmocanu]

Time/money wise, it's good to have a good fast workstation, not extraordinarily expensive (atm a 6 core Westmere or 1 SB-E or 1 BD 8 core), maybe 1600-1700 USD investment, with 24-32 GB ram and than get 3-4 SB quads (2600k) systems, cheap, no gpu (600 USD each) etc.., just for slaving away at renders. with 12 GB ram, since for rendering you need less memory than simulating. This way instead of investing 5-6 k in 2 ultra high Xeons + SR-2 etc.., you get a more even investment but when you look at USD/CPU core, than your actually getting quite a bit more out of your money.
My studio where i work goes for 950 or 960 with 12 GB of ram for workstations, the same speed as older dual quad (penryn or conroe) but much more efficient with high ram situations (FSB was a big bottleneck) and then 1-2 machines are dual nehalems or dual westmere for heavy duty purposes, somewhat of an average of my ideal solution.

[SMTB1963]

Some solid discussion and some good insight going on here, yew

+1...this thread is why my no-nothing-ass comes 'round here...was waiting for next intel platform before buying new comps to go 24/7/365 WCG, but looks like 2600K's are/will be best bang for the (WCG) buck in the near term...

(at least til 9/19)

[Vapor]

Thread lightly cleaned. Again, linking to sites/posts of banned members is not allowed here....when someone is removed from XS, they no longer have a voice on XS.

[trn]

The most interesting CPU I think will be the K version of the SB-E (3960K I think it was.) Unlocked multi, 6c 12t and supposedly in the $600 range. What desktop applications (things we care about will benefit from the 15mb of cache in the 3960x vs the 12mb in the K version?