Intel Haswell Overclocking Clubhouse.

[TheHunter]

What's interesting is the huge differences between boards. Not sure what's up with that, actually, might be the T-Topology stuff on the faster boards paying off. I finished testing the ASUS Maximus VI Hero last night and it put up quite similar numbers as the Gigabyte Z87X-OC.




X79 reports avg 60k MB/s, slightly higher latency. The bench is very different now. Sub-timing tweaks make for the huge differences with the Z87 you and I see, IMHO. What I report in board reviews for memory performance is "out-of-the-box" performance, with just XMP enabled. Cache may play a role as well, although the impact is minimal that I have seen. CPU speed also seems to have little impact on Z87, where as with SNB and IVB the differences between stock CPU and OC to 4.6 GHz were rather pronounced.

Yeah Aida64 3.0 is now multi threaded like sisoft sandra which utilizes x79's quad channel ram bandwidth, I wish z87 would have such speeds xD

 

[Womper]

Is overclocking your cpu multi/cache with a 1:1 ratio really a necessary requirement for daily type usage? I find running higher cache speeds also require greater cache voltages to stabilize... What about dropping cache multi down some? I'm currently testing (dialing in) a 46x/43x overclock for possible use while "crunching" Rosetta/WCG... etc. I might even consider going lower to 41x/42x. I believe that the performance hit for running lower cache speeds is rather minimal?

No, 1:1 ratio isn't magical. I tested the new 3dmark at 4.8Ghz with a 3.9GHz cache and a 4.6Ghz cache, but didn't see anything close to decisive- the normal test to test deviation was too large. This isn't surprising since increasing the cache speed, from the CPU's point of view, is no different than using faster RAM in principal. And we all know that most, if not all, games see very little performance improvement after DDR 1600. If a higher cache speed were required, it would really be more of a CPU design flaw/limitation.

Anyways, why am I running 4.8GHz core (+0.225v offset) with a 4.6GHz cache (+0.250v offset)? Well, I haven't noticed any significant increase in temperatures while playing with cache speed and voltage, so why not? It will be the first thing I turn down if I uncover stability issues. And if your cache can't go past 4.3GHz or whatever, then don't sweat it, it's not worth the trouble.

I also just realized that I should stop using "Sync all cores" and start using "Per Core." With that same +0.225v offset I'm going to shoot for 5GHz for 1 & 2 cores active, 4.8GHz for 3 & 4 cores active. I found that 1.41v manual is required to boot all cores sync'd at 5GHz. Damn man, Haswell is really fun to tweak. So many options, so little time.

P.S. Per your guys' suggestion I have a Swiftech H220 on the way. Let's see if it can take 4.8GHz under (non-AVX) load.

 

[radrok]

Thanks Dave and Earthdog.

Picked the G.Skill kits

Gonna be fun to see if I can get that speed on 64GB

 

[PolRoger]

No, 1:1 ratio isn't magical. I tested the new 3dmark at 4.8Ghz with a 3.9GHz cache and a 4.6Ghz cache, but didn't see anything close to decisive- the normal test to test deviation was too large. This isn't surprising since increasing the cache speed, from the CPU's point of view, is no different than using faster RAM in principal. And we all know that most, if not all, games see very little performance improvement after DDR 1600. If a higher cache speed were required, it would really be more of a CPU design flaw/limitation.

I'm curious to know when you tested 48x/39x... How much less ~voltage for CPU Cache did your sample require than when you are running/testing it at 48x/46x?

I'm not really sure... But I've been thinking lately that at higher multi(s)... successfully stablized Haswell overclocks are actually impacted more by increases in cache speed (ratio) than perhaps even bumping up to run higher memory speeds?

Anyways, why am I running 4.8GHz core (+0.225v offset) with a 4.6GHz cache (+0.250v offset)? Well, I haven't noticed any significant increase in temperatures while playing with cache speed and voltage, so why not? It will be the first thing I turn down if I uncover stability issues. And if your cache can't go past 4.3GHz or whatever, then don't sweat it, it's not worth the trouble.

I've run higher cache speeds at 44x/45x but I'm much more aware of the "voltage tweaking" required to dial in higher cache speeds as compared to lower cache speeds while running at 46x/47x. I've done very little testing at 48x... As I'm waiting to return home when I'll be able to access my custom "420" water cooling setup.

 

[cadaveca]

But I've been thinking lately that Haswell overclocks are actually impacted more by increases in cache speed (ratio) than perhaps even bumping up to run higher memory speeds?

not in my findings, cache means maybe 2-5% performance boost for most apps. Some see no difference.

Memory, still working on. I really got to stop killing CPUs

push cache voltage a touch higher than CPU voltage to go a bit further, perhaps, works for me.

I have 2933 MHz G.Skill set, review will be done very soon, these benchmarks I am doing now are the "baseline" benchmarks with other kits. Having issues getting some older sets to pass 3DMark Cloudgate Physics test (can prime, wprime, anything, just not this 3DMark and sometimes Metro:LL will crash).

5 hours in on testing, still can't get this bloody benchmark to pass at anything. I WILL HAVE SUCCESS!!!

 

[PolRoger]

not in my findings, cache means maybe 2-5% performance boost for most apps. Some see no difference.

Memory, still working on. I really got to stop killing CPUs

push cache voltage a touch higher than CPU voltage to go a bit further, perhaps, works for me.

I don't think I was clear with my intent on the previous post... (See below).

I'm not really sure... But I've been thinking lately that at higher multi(s)... A successfully stablized Haswell overclock can actually be impacted more by increases in cache speed (ratio) than perhaps even bumping up to run higher memory speeds?

 

[cadaveca]

I don't think I was clear with my intent on the previous post... (See below).

I got that. Still no difference, and yes, my chips are fully stable when testing.

I have not tested many 3d apps yet, just all the tests you see in my memory/board reviews, but in real-world use the difference is very small.

Sure, some benchmarks show big differences.. but that's benchmarks. the same could be said of memory as well.


Does one have more impact than the other? Probably, and yeah, I would think cache would be more important as well. But really, it's going to depends on how well the cache is utilized.

 

[Womper]

I'm curious to know when you tested 48x/39x... How much less ~voltage for CPU Cache did your sample require than when you are running/testing it at 48x/46x?

I'm not really sure... But I've been thinking lately that at higher multi(s)... successfully stablized Haswell overclocks are actually impacted more by increases in cache speed (ratio) than perhaps even bumping up to run higher memory speeds?

I've run higher cache speeds at 44x/45x but I'm much more aware of the "voltage tweaking" required to dial in higher cache speeds as compared to lower cache speeds while running at 46x/47x. I've done very little testing at 48x... As I'm waiting to return home when I'll be able to access my custom "420" water cooling setup.

For 48x/39x, "Auto" cache voltage rule wasn't stable. I hit the cache with 1.15v manual voltage for that. I didn't try any other voltages; it probably only needs a small bump. 48x/46x needs manual 1.35v on the cache. 47x cache needed north of 1.4v manual, and I wasn't comfortable with that. 45x cache needs a manual 1.25v. So ya, you need a small bump to cache voltage when you really push the core.

Performance wise between increasing cache speed vs DDR speed, I think it would be workload dependent. Each can offer an additional couple % improvement, independent of each other. Those % will be different for each workload.

Stability wise, who knows. Every chip is different. The cache might be the more likely candidate; I would guess that forcing 4 * 2MB caches interconnected on a ring buffer to all operate at a higher speed is probably tougher to achieve than telling the memory controller to use DDR 2800- just from a gate-count perspective.

 

[cadaveca]

For 48x/39x, "Auto" cache voltage rule wasn't stable. I hit the cache with 1.15v manual voltage for that. I didn't try any other voltages; it probably only needs a small bump. 48x/46x needs manual 1.35v on the cache. 47x cache needed north of 1.4v manual, and I wasn't comfortable with that. 45x cache needs a manual 1.25v. So ya, you need a small bump to cache voltage when you really push the core.

Performance wise between increasing cache speed vs DDR speed, I think it would be workload dependent. Each can offer an additional couple % improvement, independent of each other. Those % will be different for each workload.

Stability wise, who knows. Every chip is different. The cache might be the more likely candidate; I would guess that forcing 4 * 2MB caches interconnected on a ring buffer to all operate at a higher speed is probably tougher to achieve than telling the memory controller to use DDR 2800- just from a gate-count perspective.

What I have been finding is that memory itself it quite important for stability, and sticks that were great on IVB.. not so much any more. Looking at other forums and such, other people are reporting the same as well, and I noticed that secondary timings for Haswell are really different, so XMP settings for some users are going to pose stability issues.

When it comes time to overclock memory, CPU differences are huge, and it seem to be playing out weird for me. The chip that boots the highest divider is not my best memory clocking chip...it's one that cannot even boot 2933 divider. If that's the CPU, the board, how the ICs and memory interact..I am not sure.

Cache performance has been something I've ALWAYS looked at. Early benchmarks of Bulldozer showing low L2 cache performance was how I knew to not expect much from those chips...cache speeds didn't support the level of performance that some were expecting.


It ahs definitely got a lot more complicated with Haswell, and having the Cache speed adjustable is very nice, since it was tied to CPU speed with 1155. And the lack of impact on performance I've noticed from this change really surprises me.

 

[PolRoger]

I decided to try out the ASUS "4-Way Optimization" software in the new AI Suite III.

http://api.viglink.com/api/click?fo...=BoaAT5TkXc4&jsonp=vglnk_jsonp_13723406137336

I thought it did a pretty decent job overclocking my CPU... It went from complete stock (auto/defaults) to a 4.8G(2-core)/4.7G(4-core) overclock at DDR3-2400 10-11-11-31-2T (my 4x4GB kit's XMP).

I've gone and tweaked the ASUS settings a little more... Now running 4.8G(3-core)/4.7G(4-core) and bumped CPU Cache from 39x to 41x (may go higher) and I tighten ram to 10-11-11-28-1T (1.635v). I've also customized some of the Digi+ Power Control settings to my preference. It is running now on Adaptive vcore settings and uses ~1.275v VID for 4.8G and ~1.255v VID for 4.7G. I also thought the Fan Xpert 2 optimization and controls are pretty nifty... It is now running quieter but still with acceptable cooling.

 

[Womper]

No luck with Per-Core modes. 48x on all cores doesn't boot, while Sync All Cores has been quite stable. I tried switching to Adaptive and using Additional Turbo voltage instead of an offset, but that doesn't help either. Maybe it needs Adaptive Offset or less aggressive cache speeds to do per-core.

 

[cadaveca]

3400 MHz ram:

http://valid.canardpc.com/show_oc.php?id=2846038

 

[sneekypeet]

Just saw someone on FB posting about this, might be entertaining at the very least...http://www.scan.co.uk/shops/intel/webcast

Also since it is from Scan.UK, you may need to adjust the times a bit

 

[Jstn7477]

Hope you didn't start reviewing the ASRock boards yet as I just got a P1.90 UEFI update and ASRock pulled all the previous versions off their website, so this one better be good. They finally added a Vcore reading to the H/W monitor page and mine is 1.002-1.007v after the UEFI flashed (still no actual indication of what it is set to). There's also a setting called CPU OC fixed mode that supposedly disables throttling but I'm not going to use it because I don't want to fry my hot chip. Hopefully I can get a better OC now as I was literally crashing at 4.2GHz today like my chip has been degrading.

 

[dumo]

Testing Asrock OCF-M board on air

4X4GB 3300 straight boot up from bios



3000 CL11, TWCL6 1T

 

[springs113]

Member Result Rankings

Help everyone else find out where their CPU ranking in the "Silicon Lottery". Results Rankings require CPU-Z screenshots of both CPU and memory pages. It's best to run a light application like SuperPi 32M to make sure CPU and Cache multis are running at their maximum soeeds for your screenshot. Please report CPU batch, clocks, the cooler used, the voltage you used for CPU, memory and, CPU cache, and anything else modified that might help others get the most out off their chips too.

Please follow this format in order to have your results added to our list. Unstable "maximum" speed screenshots are just as great as fully-tested clocks, just be sure to note which type of OC your are posting. If some details are missed, that's fine, but screenshot using the proper version of CPU-Z is mandatory. CPU-Z reviewer kit with other CPUID softwares like Haswell-ready CPU-Z and other included can be found HERE:


Format:

user name|CPU/batch|Stock Volt:CPU/Cache|OC CPU Clock|OC CPU Voltage|OC Cache/Ring Multi|OC Cache/Ring Voltage|Memory Speed|cooling|notes



*Batch information can be found both on the chip itself, and on the box. Pictures of either are greatly appreciated!

infos:| USER NAME | CPU/BATCH | Stock VOLT:CPU/Cache(Ring) | OC CPU CLOCK | OC CPU VOLT | OC CACHE/RING MULTI | OC CACHE/RING VOLT | MEM SPEED | COOLING |notes
1.|v12dock|4770K/L307B198|1.072 V/-|4500(45x100 MHz)|1.285 V|x39|1.150|1600 MHz|-|ASUS Z87 PRO/MicroCenter
2.|cadaveca|4770K/L309B318|1.040 V/1.015 V|4600(46x100 MHz)|1.280 V|x39|1.140 V|2800 MHz C11|CoolerMaster TPC812|boardtestchip
3.|cadaveca|4770K/L311B405|1.020 V/1.040 V|4625(37x125 MHz)|1.250 V|x36(4500 MHz)|1.150 V|3000 MHz C12|Corsair H90|ASUS Maximus VI Extreme/memtestchip
4.|15th Warlock|4770K/L311B411|1.000 V/1.024 V|4700(47x100 MHz)|1.280 V|x40|1.150 V|1866 MHz|Corsair H100|MSI Z87-GD65 GAMING/Newegg
5.|DOM|4770K/L311B411|1.053 V/1.053 V|5500(55x100 MHz)|1.725 V|x39|1.200 V|2400 MHz C9|Single-Stage Phase|MSI MPower MAX/Newegg
6.|PolRoger| 4770K/L310B492 | 0.992 V/0.999 V |4500(45x100 MHz)|1.150 V|x43|1.150 V|2400 MHz C10|Corsair H110|ASUS Z87 Deluxe/Microcenter
7.|Womper|4770K/L312B318|1.008 V/-|4800(48x100 MHz)|1.350 V|x45|-|2133 MHz C9|Zalman CNPS10x |ASUS Z87-PLUS/Intel EPP.
8.|Allen337|4770K/-|-/-|4800(48x100 MHz)|-|-|-|1600 MHz C-|Corsair A50|ASUS Z87 PRO
9.|Justin7477|4770K/L312B326|-/-|4400(44x100 MHz)|1.220 V|x41|1.220 V|2133 MHz |CoolerMaster TPC812|Z87 Extreme6
10.|TheHunter|4770K/L308B202|0.967/-|4400(44x100 MHz)|1.150 V|x42|1.210 V|2133 MHz C9|Corsair H90|ASUS Z87 DELUXE
11.|springs113|4770K/L311B411|-/-|4400(44x100 MHz)|1.220V|-|-|2000 MHz C-|Corsair H100|MSI MPower/Newegg
12.|FireKillerGR|4770K/L312B508|1.020V/1.000V|4700(47x100 MHz)|1.350 V|x41|-|-|-|-
13.|HammerON|4770K/L311B411|-/-|4400(44x100 MHz)|1.170 V|x39|1.000 V|2666 MHz C11|Custom Water|MSI Z87-GD65 GAMING/Newegg
14.|
15.|

Hey Dave my stock Cpu voltage is 1.056, current ring voltage 1.040, OC...ring x39, 1.144...swiftech h220

 

[cadaveca]

Testing Asrock OCF-M board on air

4X4GB 3300 straight boot up from bios

[url]http://img853.imageshack.us/img853/7941/x3y.png[/URL]

3000 CL11, TWCL6 1T

[url]http://img849.imageshack.us/img849/258/dgzm.png[/URL]

NICE!!!


I'll be working on this board myself very soon.

Loving your ram clocks, too.

 

[MxPhenom 216]

Wasn't the Asrock Z77 OCF pretty good with ram overclocking too?

 

[dumo]

NICE!!!


I'll be working on this board myself very soon.

Loving your ram clocks, too.

Thanks

As long as not CFR then latest bios will be smooth sailing.

Gskill 2933 clocked the same if not better.

 

[springs113]

Thanks

As long as not CFR then latest bios will be smooth sailing.

Gskill 2933 clocked the same if not better.

Is it worth it for me to get rid of my mpower for this board? Originally i wanted this board they just took too long to bring it out.

 

[DOM]

flashed new beta bios and got 170 bclk

http://valid.canardpc.com/2849067

 

[dumo]

Is it worth it for me to get rid of my mpower for this board? Originally i wanted this board they just took too long to bring it out.

I haven't test any Z87 MSI board, but saw good things on Cadaveca's review.

Keep your Mpower board unless you want to clock memory competitively and benching 2Ds

 

[TheHunter]

Hey Dave my stock Cpu voltage is 1.056, current ring voltage 1.040, OC...ring x39, 1.144...swiftech h220

Cadaveca could you please update my post too?


Missing data: Default ring cache voltage is 1.049v



And my new cpu OC atm 4.5Ghz @ 1.170v, Ring cache 41x @ 1.115v

 

[DOM]

idk if any knows but heres the MSI BETA & MOD BIOS Section

think you have to register to be able to dl

 

[EarthDog]

Or know MSI people. LOL!