Skylake or Broadwell? (Spent 250 on z97 platform already)

[TechyGeek]

What would buying DDR4 + MB + CPU for $1200~1500 get me?
Would I rather settle for ddr3 haswell/broadwell, and dump 6-700 on gpu?
I have good 750 watt psu btw.

 

[Dave2150]

Skylake will probably be delayed to next year.

If you need an upgrade, I'd just go ahead.

 

[philipma1957]

wrong spot please delete

 

[R0H1T]

What would buying DDR4 + MB + CPU for $1200~1500 get me?
Would I rather settle for ddr3 haswell/broadwell, and dump 6-700 on gpu?
I have good 750 watt psu btw.

The latter would be a much better option IMO, get a better GPU now (or sometime later in the year) as DX12 is going to make the dGPU even more important than it is now.

Also I'd wait for Cannonlake with DDR4, PCIe 4.0 & USB 3.1 to show up because as things are today you need a ~600$ 5930K, which frankly is a ripoff as the 5820K (same number of cores but less PCIe 3.0 lanes) costs roughly 40% less, just to get 40 PCIe 3.0 lanes. Anything less than that is definitely not ideal for a build that needs to last some 3~5 years & support multi dGPU configs, perhaps PCIe/NVMe SSD's as well, & they'll definitely be bottle-necked by the lack of a good number of PCIe lanes on the 5820K.

I'd say by the time Cannonlake arrives DDR4 will definitely be cheaper, 6 (or 8) core systems might make it to the mainstream & we'll have PCIe 4.0 by then & widespread adoption of USB 3.1 & PCIe/NVMe drives.

So whatever the popular opinion is on this board my next upgrade will be Cannonlake & nothing before that since Skylake IMO will not be the magic elixir as some are predicting it'd be & it still lacks a good number of features I'd like to see in my next build, USB 3.1 & PCIe 4.0 being the most obvious.

 

[Lepton87]

The latter would be a much better option IMO, get a better GPU now (or sometime later in the year) as DX12 is going to make the dGPU even more important than it is now.

Also I'd wait for Cannonlake with DDR4, PCIe 4.0 & USB 3.1 to show up because as things are today you need a ~600$ 5930K, which frankly is a ripoff as the 5820K (same number of cores but less PCIe 3.0 lanes) costs roughly 40% less, just to get 40 PCIe 3.0 lanes. Anything less than that is definitely not ideal for a build that needs to last some 3~5 years & support multi dGPU configs, perhaps PCIe/NVMe SSD's as well, & they'll definitely be bottle-necked by the lack of a good number of PCIe lanes on the 5820K.

I'd say by the time Cannonlake arrives DDR4 will definitely be cheaper, 6 (or 8) core systems might make it to the mainstream & we'll have PCIe 4.0 by then & widespread adoption of USB 3.1 & PCIe/NVMe drives.

So whatever the popular opinion is on this board my next upgrade will be Cannonlake & nothing before that since Skylake IMO will not be the magic elixir as some are predicting it'd be & it still lacks a good number of features I'd like to see in my next build, USB 3.1 & PCIe 4.0 being the most obvious.

Nah, 5820K is fine for up to 3 graphics cards 3x8=24 and an PCI-E SSD 4X so unless you want 4 graphics cards 5820k is fine and even 40 lanes won't support all 4 graphics cards at 16x. Not even 3 as that would be 48 lanes. So no gain here really. Unless you would buy X99-WS I actually wanted to buy that mobo but I couldn't find it available and the additional latency could negate any BW improvement especially as 8X PCI-E 3.0 is not a bottleneck. Only two way would move up to 16/16x. I don't think having one card at 16x and the other at 8x is measurably better than both at 8x. I have 16x/8x but 8x/8x would work just as well. Trust me I overspend so much money on those titan that if moving up to 5930K from 5820K would do any good I would do it, it would be pennies really. I could have bought lower speed memory and 5930K instead but I read that 5930K actually are a bit worse overclockers which is strange. I had hopes that BW-E would be a real improvement that's why I didn't buy 5960X which I regret now. I thought that I would be replacing my CPU anyway in the not too distant future and in the meantime 6c/12t would do just fine.

ps. The only exception to that could be compute performance which is more sensitive to PCI-E bandwidth but I don't do that so I didn't care.

 

[R0H1T]

5820K is fine for up to 3 graphics cards

ps. The only exception to that could be compute performance which is more sensitive to PCI-E bandwidth but I don't do that so I didn't care.

For now, I doubt that'd be the case next year with HBM2 & 8 (or 16) GB dual GPU monsters with their bandwidth exceeding 1TBps. Also we already have PCIe 8x SSD's (that's probably only PCIe 2.0 ) so anyone with 3 or more top end GPU's & 2 or more high end SSD's will not be able to running his system at its optimal speed, that's with the 5820K & just looking at 2016.

Also now with DX12 incoming, GPU compute is going to matter even more & so is the number of PCIe lanes that the GPU(s) have access to.

 

[TechyGeek]

So bottom line please?
Assume that at most, I have 2 GPUs in sli/xfire, but ideally 1.
4790k or 5790k will not bottleneck next 2 generations of GPUs.
E.G. it won't bottleneck pascal, volta, or 490x, 590x.

Is this correct?

For now, I doubt that'd be the case next year with HBM2 & 8 (or 16) GB dual GPU monsters with their bandwidth exceeding 1TBps. Also we already have PCIe 8x SSD's (that's probably only PCIe 2.0 ) so anyone with 3 or more top end GPU's & 2 or more high end SSD's will not be able to running his system at its optimal speed, that's with the 5820K & just looking at 2016.

Also now with DX12 incoming, GPU compute is going to matter even more & so is the number of PCIe lanes that the GPU(s) have access to.