i7 5820

[NTMBK]

Texture loads you mean how long it takes a level to load? I'd say that it's almost entirely reliant on the SSD's speed even with PCI-E 2.0 8X or lower which is still miles ahead of consumer SSDs in terms of speed and it cannot be saturated by SSDs barring some ultra-fast PCI-E solutions that utilize many controllers and cost in the thousands of dollars.

I was more meaning time taken to see full res texture- I'll often see very low res textures, improving through several levels of quality until it finally reaches high quality as the texture streams in. As you point out though it could be down to disc performance, I have a terrible old hard drive connected by SATA 1.0 I was assuming the texture would be in main memory, but of course it's doubtful most games would fit all of their texture caches into my tiny 4GB...

(God, I really need a new PC...)

 

[BonzaiDuck]

When using a 5820k for tri sli you will need to check the motherboard manual to see if it runs 8x/8x/8x or 16x/8x/4x, since the latter does not support tri sli as nvidia requires at least 8x for all cards in sli configurations. My motherboard does not support tri sli with a 5820k.

I thought it was dependent entirely on the processor. Why would various mobo makers, all offering X99 boards, offer any variation in multi-GPU capability when someone could easily prefer a 5960X or 5930K over the 5820K? Release of the i7-5820K is the first time Intel chose to limit PCI-E lanes with the processor.

 

[Lepton87]

I thought it was dependent entirely on the processor. Why would various mobo makers, all offering X99 boards, offer any variation in multi-GPU capability when someone could easily prefer a 5960X or 5930K over the 5820K? Release of the i7-5820K is the first time Intel chose to limit PCI-E lanes with the processor.

The configuration of PCI-E lanes is different for 28 lanes CPUs and 40 lanes CPUs depending on the MOBO so for for examples a mobo may offer 16x/8x/4x for 28 lanes CPU and 16x/16x/8x for 40 lanes CPU, the former is less than ideal for 28 lanes CPU, 8x/8x/8x is a better option for three cards on a 28 lanes CPU.

 

[BonzaiDuck]

The configuration of PCI-E lanes is different for 28 lanes CPUs and 40 lanes CPUs depending on the MOBO so for for examples a mobo may offer 16x/8x/4x for 28 lanes CPU and 16x/16x/8x for 40 lanes CPU, the former is less than ideal for 28 lanes CPU, 8x/8x/8x is a better option for three cards on a 28 lanes CPU.

Hmm . . . I think you may be right. Different models of mobos allocate limited resources differently, even with the same chipset.

Listen. I have to ask you this, and hope you aren't irritated by the question.

You seem to be in a hurry to build this Haswell-E. And that's OK. I'm just wondering about your Sandy Bridge system.

Are you having troubles with it? I'm curious as to why you have the processor volted to 1.45V to get 4.77Ghz. I can do 4.70Ghz on mine without touching the bCLK (=100), and "turbo" voltage bottoms out at 1.368/1.37V, or unloaded just around 1.39V. Truth be told, I can lower the voltage settings so the loaded value is between 1.355 and 1.36V, while still passing stress-tests of marathon length.

I've never seen much wide variability in how the 2600K over-clocks. Maybe my observations will be useful for you -- maybe not. If not -- sorry for asking.

 

[Lepton87]

Hmm . . . I think you may be right. Different models of mobos allocate limited resources differently, even with the same chipset.

Listen. I have to ask you this, and hope you aren't irritated by the question.

You seem to be in a hurry to build this Haswell-E. And that's OK. I'm just wondering about your Sandy Bridge system.

Are you having troubles with it? I'm curious as to why you have the processor volted to 1.45V to get 4.77Ghz. I can do 4.70Ghz on mine without touching the bCLK (=100), and "turbo" voltage bottoms out at 1.368/1.37V, or unloaded just around 1.39V. Truth be told, I can lower the voltage settings so the loaded value is between 1.355 and 1.36V, while still passing stress-tests of marathon length.

I've never seen much wide variability in how the 2600K over-clocks. Maybe my observations will be useful for you -- maybe not. If not -- sorry for asking.

It's more like 1.4V with spikes to 1.424V. Also everything works fine, those software measurement are useless anyway, I don't have a multimeter so I don't know the true voltage.

 

[Geeksmirage]

This is an absolutely brilliant thread. Thanks OP for asking this question.
The issue is that I go one way and then the other with the 4790 / 5820 decision.
Im not a big OC guy.

For me, the diff is that I dont care about the minute texture differences, but I run so many apps that I run my comp like a beast.

I dont know if its allowed, but Im having issues trying to figure out my own build as well. "Urgent (Black Fr): $2k High End Gaming / Heavy Use Build"
http://forums.anandtech.com/showthread.php?t=2410205