Haswell-E 5960X first OC impressions

[Dadealus]

Prepare your wallets boys...

AND this is only on ENG samples/ Beta boards/ beta bios

 

4.5 @ 1.29vcore 

4.3 @ 1.25vcore

EDIT( the forum would not let me link the pictures in directly. I had to attach them? Pics are below)

Cooling used was H100

 

Sauce

 

 

 

managerman:

Was only able to test the 5960x so far, but it took forever trying to dial in the settings on the motherboard I am using. Obviously still beta bios and beta motherboard drivers...It will be good when someone a lot smarter than me comes out with an X99, Haswell-E Overclocking guide.

1.) I was able to hit 4.3Ghz (stock 3.0Ghz) ...not bad considering it was not as easy as just raising the multiplier and core voltage....lots of weirdness in power settings that I had to play around with to finally get "stable" at 4.3Ghz

2.) Had issues with having one or both Speedstep and C States being disabled..caused a crazy "Phantom" load on the CPU when no process was actually causing the load...I ended up keeping them enabled and setting windows to "performance" in the power options.

3.) HOT HOT HOT!! My Corsair H100i was no match for the heat that the 8 cores was producing. I was only at 1.25 volts on the vcore and my temps, while running small FTT in Prime 95, were easily in the 90's on some cores. Adding anymore voltaage would shoot temps to over 100C and shutdown would occur. I'm sure with more robust cooling higher voltage and speeds can be achieved.

4.) Adding voltage and speed to the ring bus adds more heat! I was however able to get the ring bus up to 3.5ghz at 1.17v...

5.) The 16GB of DDR4-2666 would only run at 2133, anything higher and the machine was not stable. I assume better MB bios will help this issue...

6.) Lots of temp variation between the 8 cores. Almost 15C observed in some cases....

7.) Disclaimer: This was only a quick test. Take the information with a grain of salt......not to the bank....

Overall I am pretty excited about the 5960x....When BIOS' mature and the experts come out with more and better overclocking guides for Haswell-E...it should be pretty interesting....

Next up 5820k...

-M

 

EDIT

update about  5820K
 

Thanks for the positive comments from everyone!

I had to go out of town for business at the last minute on Monday...so that delayed my testing. Before I left I was able to run a quick overclock on the 5820k. I did not take any screen shots..

I was able to get the 5820k up to 4.5ghz at 1.25v....

I plan on doing some more testing this weekend with the H100i, but am also going to hook up a quad radiator, D5 pump, and Swiftech block and see if there is any more headroom on the 5960x...I will post screenshots of the 5820k results too..

.....and to fuel the debate on DDR4 memory.....here is a quick comparison of DDR3 on X79 and DDR4 on X99...Interesting...again take these with a grain of salt....

X79 DDR3-1866 Cas 9 Passmark memory bench = 3355
X99 DDR4-2666 Cas15 Passmark memory bench = 3028

Each with 4x4gb sticks

I sure hope mature MB bioses are waiting in the wings....

Only 7 days before launch!!

-M

 

 

 

Think about it .... a 5960X @ 4.4  is the same as having TWO 4790Ks.

 

Given mantle, openGL, directX12, xbone, and PS4 are all going to be pushing the gaming industry to multicore.  I think its just a matter of time before we see some performance gains with multicores. do you want to be the guy with just 4c/8t when that shift happens? or do you wanna be the with  6c/12t or 8c/16t.

 

side thought.... DAT DDR4 prices.

 

Although I had a retailer talking about DDR4 prices and was stating there would be up to a 20% decrease in DDR4 prices after September. He states this is due to Intel bumping Haswell-e release up one month. This has caused DDR4 supplies for launch to be EXTREMELY LIMITED and pricey.

I'll find the source on this too and post it here.
EDIT
memory sauce

 

Gibbo, Staff @ OverclockersUK store
As with all DDR4 the product is in severe shortage and as such the early adoption price on the G.Skill is rather high, we predict these prices to fall by around 20% in late September. We are being as honest and upfront as possible so please bear this in mind. DDR4 is so short simply because Intel moved the launch date forward by over a month which has caused DDR4 memory to be in shortage.

 

 

 

Gskill's price isn't looking as bad as the dominator platinum's 500$ for 4x4GB

G.SKILL Ripjaws Series 32GB (4 x 8GB)  2666 (PC4-21300)  15-15-15-35

[AlwaysFSX]

I need new pants now.

[dragoon20005]

open wallet...

 

 

damn

[patrickjp93]

God that heat... That can't be a well-done solder job. WTH Intel? Haswell is thermally efficient in ways Ivy Bridge couldn't touch and yet your 8-core is pushing 90s at frequencies much lower than the AMD 9370...

 

I get the feeling even an EK or AquaComputer Cuplex Kryos water block won't do the job without a naked mount, and that's a soldered chip...

[LukaP]

hotdamn if he can get the 5820 to 4.5 i shall prepare me wallet

[Dadealus]

 Think about it .... a 5960X @ 4.4  is the same as having TWO 4790Ks.

 

Given mantle, openGL, directX12, xbone, and PS4 are all going to be pushing the gaming industry to multicore.  I think its just a matter of time before we see some performance gains with multicores. do you want to be the guy with just 4c/8t when that shift happens? or do you wanna be the with  6c/12t or 8c/16t.

 

side thought.... DAT DDR4 prices.

 

Although I had a retailer talking about DDR4 prices and was stating there would be up to a 15% to 30% decrease in DDR4 prices after September. He states this is due to Intel bumping Haswell-e release up one month. This has caused DDR4 supplies for launch to be EXTREMELY LIMITED and pricey.

I'll find the source on this too and post it here.

[patrickjp93]

 Think about it .... a 5960X @ 4.4  is the same as having TWO 4970Ks.

 

Given mantle, openGL, directX12, xbone, and PS4 are all going to be pushing the gaming industry to multicore.  I think its just a matter of time before we see some performance gains with multicores.

4790K*

 

And, not quite, no iGPU for quick heterogeneous computing.

[APrettyCoolWalrus]

Those temps, I hope this gets sorted out. Is this as hot as the 9590? I'm excited for the 5820k, it's got me saying:

dayum 

dayum

DAYUM

[Dadealus]

Just updated some DDR4 gossip

[patrickjp93]

Just updated some DDR4 gossip

I didn't think G.Skill would let Crucial and Kingston both show it up...Waiting on the Trident X series before I judge. It better have a CL of 14 or less or I'm gonna be pissed at G.Skill for the first time in my life.

[Real_PhillBert]

 

*throws money at monitor*

[Najuno]

Man, if these OC results are real, just if this is true, then there is no doubt in my mind that I will buy the 5960X CPU!

 

But it's kind of hard to believe since the base clock is 3.0GHz and Turbo will or is supposed to be at 3.5GHz. To push it up to 4.5GHz with that Voltage looks unreal haha .

[TBoneSan]

Silicon Lottery. Hopefully these ES's are similar to the OEM's. 

I'm thinking ill go 5820k if they can break 4.5 with decent temps/volts.

[FellTheSky]

i was expecting much much less since its an 8 core

[vm'N]

God that heat... That can't be a well-done solder job. WTH Intel? Haswell is thermally efficient in ways Ivy Bridge couldn't touch and yet your 8-core is pushing 90s at frequencies much lower than the AMD 9370...

heat-density is a big issue. Especially with Intels setup (having all cores places next to eachother), where AMD place the modules in each corner.

These core are extreme big. A haswell core is around the same size as a piledriver module, but is on 22nmFF where piledriver is on 32nm SOI.

Try to overclock a Opteron 6386 SE.

[patrickjp93]

heat-density is a big issue. Especially with Intels setup (having all cores places next to eachother), where AMD place the modules in each corner.

These core are extreme big. A haswell core is around the same size as a piledriver module, but is on 22nmFF where piledriver is on 32nm SOI.

Try to overclock a Opteron 6386 SE.

I'm aware, but Haswell was built to be efficient volt for volt, lowering heat output. 8 cores or not at just 1.3 volts 90C is nuts!

 

I might just have to buy the Cuplex Kryos (never thought I'd say that) and a better heat gun to pry the heatspreader off for a naked mount if this crazy trend keeps up.

[vm'N]

I'm aware, but Haswell was built to be efficient volt for volt, lowering heat output. 8 cores or not at just 1.3 volts 90C is nuts!

That was mostly done by turning of units that are not been used more efficiently. But that doesn't help under heavy workload.

[patrickjp93]

That was mostly done by turning of units that are not been used more efficiently. But that doesn't help under heavy workload.

There's a lot more to it than that, but yes, I know when workloads are maxed out the ALU and FPU densities are not kind to thermal design. Hopefully Broadwell's extra electrical efficiency can solve a little of that heat problem, although increasing the ALU count to 4 might counteract that a bit (a lot).

[vm'N]

There's a lot more to it than that, but yes, I know when workloads are maxed out the ALU and FPU densities are not kind to thermal design. Hopefully Broadwell's extra electrical efficiency can solve a little of that heat problem, although increasing the ALU count to 4 might counteract that a bit (a lot).

The major difference between ivy and haswell in power consumption is due to a more effecient way of shutting down units and of course the integrated VRM and a more mature FinFET.

EDIT: An ALU is considered small and easy to implement. The hard job is utilizing it (which require alot of diespace to play around with).

[patrickjp93]

The major difference between ivy and haswell in power consumption is due to a more effecient way of shutting down units and of course the integrated VRM and a more mature FinFET.

EDIT: An ALU is considered small and easy to implement. The hard job is utilizing it (which require alot of diespace to play around with).

And using 4 on one core, especially simultaneously, is expensive.

[juretrn]

By the main question is, can it go to 5 GHz? Would be funny to see the 9590's 8 "cores" vs similarly clocked proper cores. The result will probably be something along the lines of "suck iiiiit AMD"

[patrickjp93]

By the main question is, can it go to 5 GHz? Would be funny to see the 9590's 8 "cores" vs similarly clocked proper cores. The result will probably be something along the lines of "suck iiiiit AMD"

Like I've said, without the BEST water block on the market on a NAKED mount, a dense, thick quad 140ml radiator, and really high-performance fans, there's no freaking way a general consumer could do it. Maybe with chilled liquids or Peltier, but I don't think you can get above 4.6 without those higher-end methods.

 

Edit: it would be a slaughter. AMD can't compete on the highest end and the whole world knows it. With the completion of HSA on Carrizo, and with a process shrink and FinFET on its successor working with DDR4, AMD might be able to break into the server/supercomputer market.

[juretrn]

Like I've said, without the BEST water block on the market on a NAKED mount, a dense, thick quad 140ml radiator, and really high-performance fans, there's no freaking way a general consumer could do it. Maybe with chilled liquids or Peltier, but I don't think you can get above 4.6 without those higher-end methods.

Doesn't matter.

[wpirobotbuilder]

Doesn't P95 pull a lot more voltage on Haswell than it's supposed to? I remember that P95 would overwhelm my NH-D15 at stock settings.

[CatCloud]

Damn... Those are some nice numbers aside from the heat..

Can't wait to see how Skylake-E will be now!