The Gigabyte G1.Sniper 5 appeals to me since it's already set up for chipset water cooling. Not enough of a reason for me to upgrade though. I'm on a Sniper 3 with a 2600K.
CPU speed is still the #1 thing to strive for. After that is memory speed and then cache speed. Memory speed is more important if using the iGPU. Most, but no all, CPUs will do 2400MHz memory speed. This is dependent on CPU speed though. Few processors are able to achieve max CPU and memory speeds at the same time. I think the best price to performance right now for 16GB of memory is the G.Skill Trident X 2400MHz sticks.
The Gigabyte G1.Sniper 5 appeals to me since it's already set up for chipset water cooling. Not enough of a reason for me to upgrade though. I'm on a Sniper 3 with a 2600K.
OC force has that as well. I think I'm past the point in my life where I want to build custom watercooling. I might get bit by that bug again in the future though. Who knows.
Haswell's at best a fractional jump from SB-E; the same as SB-E's a fractional jump from Westmere. Two generations and maybe a 10% difference at the extreme. Since AMD's relegated themselves to "used to be an also-ran" and Intel's decided to switch to low-power, low-margin junk chips, there's no pressure to provide more than tiny improvements while obsoleting the socket every year. Seriously - the E3-1220v3 is less than 5% 'faster' than the E3-1220v2. In most situations, it's well within the margin of error. Meaning yeah - all new socket, supposedly a 'tock,' but essentially zero performance gains.
Oh, and 4K UHD is 8,294,400 pixels. 99,532,800 bits per second (presuming 12-bit color.) It just isn't an issue of bandwidth. Purely an issue of GPU horsepower and local memory. PCIe 3.0 doesn't really give you any significant gain over PCIe 2.0 as far as 4K UHD.
If you wanna go absolute top end, the winner is still SB-E 2011, period. Gigabye GA-X79S-UP5-WIFI for mild and stable overclocks with 2-4 top-end cards, ASRock X79 Extreme11 if you want to go crazy with GPUs and don't mind a buggy high risk board with no warranty or customer service that costs over $600. (Has dual PLX PEX8747 48-lane/5-port switches. And ASRock == Asus' downmarket lower quality brand.)
I'm running X58 1366 Bloomfield right now. I've waited this long, I can wait longer
of course my luck, Haswell-E will still be 16 lanes
Pretty much exactly - I'm on X58/ICH10R i7 950 Bloomfield. The only reason I'm going SB-E 2011 is necessity - board's pretty much dead, can't get a replacement. Sad when a new C602 board costs less than an old stock X58 - by a huge margin. The expected performance gain from going i7 950 to i7 3820? Well, see for yourself. Bloomfield's well ahead in some. Either way, it ain't a whole hell of a lot. It's even worse when you consider the i7 3820 is $305 MSRP (box, no heatsink - recommended Intel liquid cooling system, so add another $90) and the i7 950 was $294 MSRP with heatsink.
Basically, at this point, Intel's changing the socket constantly to force 'upgrades' that really aren't. The only real advantage of 2011 over 1366 comes down to pins. 2011 gave enough pins to go quad-channel on memory and bump the voltage to 1.5 from 1.35. (Which also, BTW, is why 1150 isn't likely to see more lanes. Could do a QPI->PCIe switch in theory, but don't hold your breath.) But otherwise, despite a 3 year gap? Fractional speed improvements, no real TDP improvement (130W vs 130W,) and less in the box for the same money.
EDIT: Oh, and lest you think it's just Intel and CPUs that's getting this way.. consider this fact. NV is blatantly doing same-product renames to fake new generations again (GTX680 = GTX770, and yes you can BIOS flash straight to a 770.) AMD's latest and greatest is two HD7970's on one card at PEG+8x each via a PLX bridge. The benchmark performance differences from both NV and AMD between generations have been narrowing and the real world differences have been evaporating more quickly. Heck, the 7990 gets beat by CF'd 7970's.
Sigh. Got to love TechPowerUp's atrociously bad reporting and complete lack of clue. The first Intel 8 core processor? NOPE. And high end UP and MP workstations have been at 6+ for ages. If going past 4 cores offered a real benefit for gaming, they would have done it long ago. But it doesn't, because games don't actually multithread worth a crap, because multithreading is genuinely hard, and because lock stalls are too hard to not screw up and very visible when you do screw it up in a game. Because it's not independent, it's all one big clump that has to interlock very tightly.
Will be quite pissed if they introduce socket incompatibility again, since at this point, it's beyond transparent that the only reason they'd do it on Haswell-E would be to screw over customers and force them to buy new motherboards. Even though the pin count is the exact same, and a socket tolerant of 1.5V DIMMs is most definitely tolerant of 1.2V DIMMs.
Comments
CPU speed is still the #1 thing to strive for. After that is memory speed and then cache speed. Memory speed is more important if using the iGPU. Most, but no all, CPUs will do 2400MHz memory speed. This is dependent on CPU speed though. Few processors are able to achieve max CPU and memory speeds at the same time. I think the best price to performance right now for 16GB of memory is the G.Skill Trident X 2400MHz sticks.
http://www.newegg.com/Product/Product.aspx?Item=N82E16820231589
neither Haswell or Ivy-E offer what I want (40 lanes and 6 SATA 6Gb/s ports)
as I told @mertesn when he visited, my next PC will be 4k gaming
Seriously - the E3-1220v3 is less than 5% 'faster' than the E3-1220v2. In most situations, it's well within the margin of error. Meaning yeah - all new socket, supposedly a 'tock,' but essentially zero performance gains.
Oh, and 4K UHD is 8,294,400 pixels. 99,532,800 bits per second (presuming 12-bit color.) It just isn't an issue of bandwidth. Purely an issue of GPU horsepower and local memory. PCIe 3.0 doesn't really give you any significant gain over PCIe 2.0 as far as 4K UHD.
If you wanna go absolute top end, the winner is still SB-E 2011, period. Gigabye GA-X79S-UP5-WIFI for mild and stable overclocks with 2-4 top-end cards, ASRock X79 Extreme11 if you want to go crazy with GPUs and don't mind a buggy high risk board with no warranty or customer service that costs over $600. (Has dual PLX PEX8747 48-lane/5-port switches. And ASRock == Asus' downmarket lower quality brand.)
of course my luck, Haswell-E will still be 16 lanes
The expected performance gain from going i7 950 to i7 3820? Well, see for yourself. Bloomfield's well ahead in some. Either way, it ain't a whole hell of a lot. It's even worse when you consider the i7 3820 is $305 MSRP (box, no heatsink - recommended Intel liquid cooling system, so add another $90) and the i7 950 was $294 MSRP with heatsink.
Basically, at this point, Intel's changing the socket constantly to force 'upgrades' that really aren't. The only real advantage of 2011 over 1366 comes down to pins. 2011 gave enough pins to go quad-channel on memory and bump the voltage to 1.5 from 1.35. (Which also, BTW, is why 1150 isn't likely to see more lanes. Could do a QPI->PCIe switch in theory, but don't hold your breath.) But otherwise, despite a 3 year gap? Fractional speed improvements, no real TDP improvement (130W vs 130W,) and less in the box for the same money.
EDIT: Oh, and lest you think it's just Intel and CPUs that's getting this way.. consider this fact. NV is blatantly doing same-product renames to fake new generations again (GTX680 = GTX770, and yes you can BIOS flash straight to a 770.) AMD's latest and greatest is two HD7970's on one card at PEG+8x each via a PLX bridge. The benchmark performance differences from both NV and AMD between generations have been narrowing and the real world differences have been evaporating more quickly. Heck, the 7990 gets beat by CF'd 7970's.
I set a GPU budget, and I buy the best that budget can get at the time of purchase.
The rumors I've seen is Haswell-E will be a new socket, LGA 2011-3
http://www.techpowerup.com/185719/haswell-e-intels-first-8-core-desktop-processor-exposed.html
Will be quite pissed if they introduce socket incompatibility again, since at this point, it's beyond transparent that the only reason they'd do it on Haswell-E would be to screw over customers and force them to buy new motherboards. Even though the pin count is the exact same, and a socket tolerant of 1.5V DIMMs is most definitely tolerant of 1.2V DIMMs.