8Packs Ambient OC on Haswell E and 4ghz comparison with 4930K.

OcUK Staff
Joined
20 Feb 2012
Posts
10,163
Location
John Smiths Stadium
When I set out to do this initial look at Intels new X99 CPU's I wanted it to be all about how efficient the CPU is and how does it compare to the last gen. I also wanted to know how far it would Overclock on good Air cooling such as the Megahelms with two fans attached and basic custom water cooling XSPC dual rad (240mm) with raystorm cpu block setup.

For once I picked the least overclockable chips I tested from system OEM stock so all my results should be easy to replicate by all customers. Most customers want cool operating hardware so if I had to use voltages where temps go above 80C I backed off.

I ran the memory speed at DDR2400 C10 for all my initial ambient overclocking testing.

For stability testing I used Intel XTU and Prime 95 ( Not normally a choice of mine but guys on here love it!!) I ran both for one hour each.

My motherboard of choice was my ASUS Rampage V Extreme which performed flawlessly throughout with some insane DDR4 overclocking possible.

First I tested the 5960X 8c 16t CPU on Air cooling 3.9-4.0 ghz was possible with 1.1v and uncore at 3.5. For this I used Cache volts at 1.1 vccin 1.9. During loading temps where in the mid to late 70's.

The 5960x definitively benefited from water cooling with 4.2-4.3ghz being possible on water with 3.5 uncore. The volts I could use on water where 1.2 core, 1.2 cache with temps topping out at 78C.

I am sure with a quad rad on CPU even more mhz will be possible. At 4.3 this chip is a true power house.

Next I looked at the 5930K 6C 12T on good air I got this one to 4.0-4.1ghz with 1.15v Uncore was at 3.5ghz with the same volts. Temps on load peaked at 78C.

Like the X the K benefited from water cooling. I managed 4.4ghz with this sample at 1.225v with uncore at 3.5ghz on the same volts. Temps peaked at 79C.

Finally I looked at the cut down 5820K which offers 6C 12T just like the 5930K but has less PCI X lanes 28 to be precise. This means its only really suited to single GPU setups. The 5820K I expected to clock the best of all as it does have these cut down features.

This was not to be on Air the 5820K reached 4ghz at 1.125v with 3.5ghz Uncore temps around 77 C peak and on water it reached 4.2 at 1.215v with 3.5ghz uncore.Temps where again in the late 70's C.

My initial conclusion about how overclockable these chips are is they are very similar to there X79 counterparts volts for volt. But they get warm much quicker which is hardly surprising given there DDR4 onboard memory controller massive on board Cache and extra PCI Ex lanes. End users need the best cooling they can to get the most from this silicon.

After testing what ambient clocking was possible I then tested the 5930K and 5960X clock for clock performance at DDR4 2400 and DDR4 3000 against the 4930K at DDR 2400. For all these tests I used GSKILL highly overclockable and tunable 3000mhz kit. I used several benches including 3D mark, 3D mark 11, Heaven, Aida and Cinebench. Below you can see the results.



I have shown screenshot from aida further down the page so you guys can see detailed scores of how DDR4 compares to DDR3.

AIDA 4930K


AIDA 5930K DDR2400


AIDA 5930K DDR4 3000


AIDA 5960X DDR4 2400


In conclusion X99 has a lot to offer. Its DDR4 bandwidth is top notch and will only improve as DDR4 matures. As is its clock for clock performance. This translates into great speed on rendering tasks such as Cinebench and I am sure all Photo editing / sound editing type applications the same. The 8C 16t 5960X ragging through this stuff faster than any desktop part before.

I also saw excellent physics benching scores so games relying on this type of compute power will gain from X99 even if only using a single GPU.

X99 comes into its own with multi GPU setups with not only massive bandwidth but also many PCI X lanes to use.

Some games will undoubtedly get a boost from X99 but not all. Haswell on Z97 for many will be enough with single GPU but this is enthusiast level so I expect early adopters to be those gamers with multi GPU setups or professional users with a need for high and efficient core count. For these two groups X99 is great and does offer real benefits.

For me its the best benching platform so far such fun tuning GSKILL DDR4 and the potential for massive scoring. 80K plus Vantage CPU test yes please ;)!!! if your a bencher X99 is also a must.
 
Last edited:
@8 Pack

Thanks for doing the post above it explains a lot.

In a couple of the bench threads I have been critical of the numbers as I suspected what was going on is what you have explained here.
 
Finally I looked at the cut down 5820K which offers 6C 12T just like the 5930K but has less PCI X lanes 28 to be precise. This means its only really suited to single GPU setups. The 5820K I expected to clock the best of all as it does have these cut down features.
.

Then all these people that are running dual cards at PCEX3.0 x8/x8 are better off suited to a single card.

x8/x8 is more then enough for 2 cards.
 
Good read cheers Ian, two questions for you...

With the 5820k being 28 lane, does this mean it will still manage tri cards at 8/8/8? Or even quad at 8/8/8/4? Or is my logic on how lanes are distributed completely wrong, 28 lanes, 8x3=24 + 4 =28.

Secondly how are motherboard temps? Would you expect better results with it on water?
 
Perfect post bud - EXACTLY what I wanted to see.

I see your also suffering the same X79 bug using 2400DDR3 (poor write speeds) - wish the motherboard peeps would sort that out!

Otherwise it's more or less exactly what we expected I think. Top marks for a very concise post bud.
 
Last edited:
Then all these people that are running dual cards at PCEX3.0 x8/x8 are better off suited to a single card.

x8/x8 is more then enough for 2 cards.

16x 16x is the ideal equation and anything less especially with todays high end GPU and low cpu clock speeds is creating bottleneck.

8x 8x is ok but not ideal.
 
MB temps fine 99% of the time as you cant use much volts before cooling on CPU is overpowered. You only need 1.25 or so to max out teh chip on dual rad at this the VRM etc cope easily. I dont anticipate any ambient cooling needing extra MB stuff as you simply cant plough 1.4v into these chips.

As for disti of lanes on 5830K I did not test that. I will try three and four Monday and respond. I tried others with this config but did not think it of interest on the budget CPU.
 
I'm just curious as to how it handles 3 and 4 way thats all, if I were going X99 (I still might!) it would likely be the 5930K.

Glad the boards are holding up well, no repeats of the hot rear mounted fets we saw on a few X79 boards :)

The 5960X looks a hot beast of a chip, a shame it seems to be choking a little with the heat, definitely one for sub-ambient cooling.

Any word on clockable Xeons + dual socket boards? Would love to see something along the lines of the SR-2 roar back to life, the crippled X79 Xeons ruined the SR-X.
 
Good read cheers Ian, two questions for you...

With the 5820k being 28 lane, does this mean it will still manage tri cards at 8/8/8? Or even quad at 8/8/8/4? Or is my logic on how lanes are distributed completely wrong, 28 lanes, 8x3=24 + 4 =28.

Secondly how are motherboard temps? Would you expect better results with it on water?

28-LANE will do 8x8x8 :)
 
Xeons are crippled. No clocking other than BLK no strap to use either.

No the table is correct. I was showing effect of memory speed vs 4930K in first three columns
 
Yeah sure will speeds of 4200mhz planned for future.

What sort of timings and speed should someone be looking for from DDR4 for a PC that will do some benching on X99.

Or putting it another way what is the equivilent DDR4 memory to the Cas 9 2400mhz Corsair DDR3 used on X79
 
C10 2400 is equivalent if not slightly faster. Ideal benching around DDR4 3000 C12-15-15-15 1T with TRFC 270 and TWCL 9.
 
Back
Top Bottom