Haswell-E Reviews

[RussianSensation]

Not too sure why people are so concerned about DDR4.

Cheapest 1 x 8GB DDR4 2133 is $102

http://pcpartpicker.com/parts/memory/#t=14&sort=a10

Cheapest 1 x 8 GB DDR3 1600 is $73

http://pcpartpicker.com/parts/memory/#t=11&Z=8192001&sort=a10

You are looking at about 40% more GB for GB assuming you are not going to buy the stupid expensive kits. Its $30 more.

If you want a 4x4 DDR4 Kit (quad channel) it will cost you $203

http://www.newegg.com/Product/Produc...D=3938566&SID=

The cheapest 4x4 DDR3 kit (1600 mhz) is $155.70

http://www.newegg.com/Product/Produc...D=3938566&SID=

Paying about $47 more. A 2 x 8 kit for DDR3 is about $127, you are paying $76 more.

Its more money but its not hundreds of dollars more unless you are paying hundreds of dollars (32-64 GB) of RAM anyway. For a full 64 GB you are looking at $631 for DDR3 and $816 for DDR4. $185 difference or 29% more. The $185 when you look at how much the build is going to cost anyway becomes much smaller.

http://pcpartpicker.com/parts/memory/#t=11&Z=65536008&sort=a10

In a few months DDR4 will have dropped much more.

Its bad if you already have DDR3 but for a new system the reduced costs of HW-E along with the performance increases cancels it out.

Excellent post. The negativity of DDR4 seems overstated for someone building a new rig from scratch. For gaming buying anything over 8GB is a waste anyway; so you can just start off with 2x4GB. Even in dual channel the platform will not be memory bandwidth bottlenecked unless we are talking about server loads.

5820K is only $299 at MC and the mobo and DDR4 premiums are easily justified in my eyes given how long modern Intel systems last (nearly 4 year old 2600K is still fast). A 6-core Haswell that will probably hit 4.4-4.5Ghz over 4790K that tends to hit 4.7Ghz is way safer bet over 4-5 years. Let's face it the extra 200-300mhz will do little for games but when future games take advantage of more than 4 threads, the 6 core will pull away by far more than 10%. If 4.4Ghz HW-E is too slow for games, 4.7Ghz version won't save the day. Those who ran X58 systems will really like these new CPUs.

I remember when Core 2 Duo E8400-8600 were really popular but in the long run they turned out to be a far worse choice than the Q9550. In that comparison, not only was the price difference greater to go with Q9550 but the clock speed difference was more than 200-300mhz and Q9550 was still the better buy. In addition, there are more PCI lanes for M.2 SSD + Tri-Fire/SLI.

As soon as DDR4 reaches parity with DDR3, it will be very hard to recommend the 4790K platform imo. I think 4790K is a better choice for non-overclockers but for overclockers, for me the 5820K is now far more preferable long term.

 

[Redstorm]

Changing over to DDR4 is the price for progress. Just because we have been static for 7 years on DDR3 means their is a ton of it in use.

If you want Haswell-E suck it up cream puffs its going to mean you need to buy new DDR4. then be happy for another 7 years (which with Haswell-E you could probably stretch the platform that long with a few GFX updates along the way).

 

[Redstorm]

As soon as DDR4 reaches parity with DDR3, it will be very hard to recommend the 4790K platform imo. I think 4790K is a better choice for non-overclockers but for overclockers, for me the 5820K is now far more preferable long term.

This is bang on.

It looks like the 5820k clocks to 4.5Ghz pretty easily, and a mild 4GHz overclock all day long, that's 2 more cores than 4790k at the same clocks on the new platform.

Long term 5820k makes more sense than buying more DDR3 at this time, which will be obsoleted before DDR4.

 

[Lepton87]

How much more expensive are faster DDR4 memory? If I'm changing my entire platform I'd like to keep the RAM for at least one more major upgrade routine. Right now I would buy just two sticks and another two when the price comes down or buy the cheapest RAM now if the price premium for faster memory is too big and sell the sticks that I bought now and buy 4 new faster sticks later on. What do you think? What's more sensible?

 

[know of fence]

I understand that overclocking these processors is mostly an academic exercise, given the massive rise in power consumption and cooling constraints. First I was surpised to see a relatively high base frequency (3.0/3.5 GHz) for an 8 core processor. Though upon further investigation Xeons of the previous generation go up to 3.7/4 GHz, at 150W. I guess E5 v3 Xeons is where the best bins end up anyway.
When it comes to hexa-cores the 5820K appears to be a bad 5930K bin, so what good is being unlocked if the voltages you're getting are high to begin with.
At least buying a Devil's Canyon quad core you don't feel like buying second rate. No, you feel like buying a year old processor late into the product cycle instead .

 

[ehume]

. . . When it comes to hexa-cores the 5820K appears to be a bad 5930K bin, . . .

Hadn't thought of that. So would the 5930k be a bad bin of 5960k? And the binning situation would explain why Intel is sending out 5960's for reviews, and not 5930's.

I guess we need to see a few more results of 5820k to know if they represent bad bins, or not.

 

[cbn]

How much more expensive are faster DDR4 memory? If I'm changing my entire platform I'd like to keep the RAM for at least one more major upgrade routine. Right now I would buy just two sticks and another two when the price comes down or buy the cheapest RAM now if the price premium for faster memory is too big and sell the sticks that I bought now and buy 4 new faster sticks later on. What do you think? What's more sensible?

If you are a gamer, I would just buy the cheapest 2 x 4GB DDR4 kit and call it a day.

Then if you feel the need to upgrade later on (for other non-gaming applications) I would buy opportunistically based on value.

 

[ShintaiDK]

Hadn't thought of that. So would the 5930k be a bad bin of 5960k? And the binning situation would explain why Intel is sending out 5960's for reviews, and not 5930's.

I guess we need to see a few more results of 5820k to know if they represent bad bins, or not.

If there is a core defect, you are down to 5930 nomatter what else the other 7 cores can clock and at what power consumption.

 

[frozentundra123456]

I understand that overclocking these processors is mostly an academic exercise, given the massive rise in power consumption and cooling constraints. First I was surpised to see a relatively high base frequency (3.0/3.5 GHz) for an 8 core processor. Though upon further investigation Xeons of the previous generation go up to 3.7/4 GHz, at 150W. I guess E5 v3 Xeons is where the best bins end up anyway.
When it comes to hexa-cores the 5820K appears to be a bad 5930K bin, so what good is being unlocked if the voltages you're getting are high to begin with.
At least buying a Devil's Canyon quad core you don't feel like buying second rate. No, you feel like buying a year old processor late into the product cycle instead .

Of course you picked the lower performance chip of the two to show. The other one clocked to 4.5 at 1.3 volts. So there is some luck of the silicon lottery involved. But these are for the 8 core processors. If it only overclocks to 4.2 or to 4.6, 8 cores will be a beast if you have the workload to use the cores. More relevant to most users will the the overclockability of the 6 core chips, which we need more information before making a decision. I really dont see why some posters are trying so hard to find negatives, since this is the first time intel has made what seems like a valid effort to make hex cores affordable for the upper mainstream, especially after DDR4 prices come down.

 

[know of fence]

Of course you picked the lower performance chip of the two to show. The other one clocked to 4.5 at 1.3 volts. So there is some luck of the silicon lottery involved. But these are for the 8 core processors. If it only overclocks to 4.2 or to 4.6, 8 cores will be a beast if you have the workload to use the cores. More relevant to most users will the the overclockability of the 6 core chips, which we need more information before making a decision. I really dont see why some posters are trying so hard to find negatives, since this is the first time intel has made what seems like a valid effort to make hex cores affordable for the upper mainstream, especially after DDR4 prices come down.

Maybe making 6 cores affordable, and 8 cores available for video production or some such, is a good perspective on things. My point is that these CPUs are bulkier working horses rather than prize thoroughbreds.

I picked this particular table because it included some power consumption numbers, given +144 W at the wall added or around +100 W under the heatsink, would someone really want to play games or encode video at with these kinds of overclocks?

 

[aigomorla]

the more and more I read the reviews for the 5960K, most of them are not doing very well in overclocking.

AT isn't the only one whose had problems overclocking past 4.5ghz...

It sort of reminds me of the old days with Q6600.
The original batchs were soso... it took a stepping revision to change that.
I am going to assume that the same might happen for the 5690K.

Intel has been pulling this type of stuff on the EE processor line anyhow with the exception of the 3960X + 4960X.

But you guys remember?
QX9650 -> QX9770
965X -> 975X
980X -> 990X

I still have the second stepping revision on all those cpu's, and yes they were significantly better then the predecessor.

I think they might pull a fast one on a stepping revision, so earily adopters might get shorted ended seeing how its not clocking as one would like them to clock.

 

[Lepton87]

I picked this particular table because it included some power consumption numbers, given +144 W at the wall added or around +100 W under the heatsink, would someone really want to play games or encode video at with these kinds of overclocks?

Emphatically YES. What's 144W? That's nothing, my PSU is yawning at the moment and waiting for some action. Although I won't get 8-core I might think about 6 core but ....

But you guys remember?
QX9650 -> QX9770
965X -> 975X
980X -> 990X

I still have the second stepping revision on all those cpu's, and yes they were significantly better then the predecessor.

I only remember the first stepping change you mentioned and it was a massive increase in clocking headroom like a node shrink. How much better were the later ones? I assume not as significant because I didn't hear about it or I didn't hear about it because it didn't trickle down like the first one with Q6600. How sure are you that there's gonna be a new stepping? It's been two generation without a stepping change it starts to seem like a trend I'm a bit pessimistic about this but I'd like to be wrong, I would buy a 5820 that OCs close to 5GHz in no time. Right now I might end up under 4.5GHz

UPDATE: now that I think about it I also remember Nehalem stepping change. It was significant, from what I remember it reduced power consumption whiled OCed more than clocking headroom but it was still a very good change but I can't remember a stepping change with 990X and I checked Intel ark and both 980X and 990X are B1
980X only B1 stepping
http://ark.intel.com/products/47932#@ordering
990X
http://ark.intel.com/products/52585...M-Cache-3_46-GHz-6_40-GTs-Intel-QPI#@ordering
also only B1
and now and 965 and 920 and 975

http://ark.intel.com/products/37149...M-Cache-3_20-GHz-6_40-GTs-Intel-QPI#@ordering

only C0
i7 920
http://ark.intel.com/products/37147...M-Cache-2_66-GHz-4_80-GTs-Intel-QPI#@ordering
Both revisions available
C0
D0
I remember that 920 D0 was more valued than 965X heh

and now 975X

http://ark.intel.com/products/37153...M-Cache-3_33-GHz-6_40-GTs-Intel-QPI#@ordering

only D0 revision
So what's with that?

 

[aigomorla]

So what's with that?

i dont recall what they exactly did.

however i do remember my sponsor telling me that there was a change in the fab process... or what he said his intel contact had told him that there was a change.

I do remember testing all the revisions and most definitely, the 990X clocked better then the 980X at a lower voltage.

The same can be said with the 975 to the 965.

The only time where i noticed almost nothing was the QX9650 vs QX9770.
Infact my QX9650 was probably a better clocker in the lower voltage spectrum then the QX9770... although the 9770 had a higher ceiling if i recall.

When i was a tester, i had the fortune to be able to play with all those cpu's.

 

[Enigmoid]

the more and more I read the reviews for the 5960K, most of them are not doing very well in overclocking.

AT isn't the only one whose had problems overclocking past 4.5ghz...

I don't know. IMO it seems like HW-E overclocks as well as the average 4770k, many of which can't go above 4.3 or even 4.2 ghz on the bad ones. Its not devil's canyon but certaintly comparable to LGA 1150 4770k.

 

[Lepton87]

i dont recall what they exactly did.

however i do remember my sponsor telling me that there was a change in the fab process... or what he said his intel contact had told him that there was a change.

I do remember testing all the revisions and most definitely, the 990X clocked better then the 980X at a lower voltage.

The same can be said with the 975 to the 965.

The only time where i noticed almost nothing was the QX9650 vs QX9770.
Infact my QX9650 was probably a better clocker in the lower voltage spectrum then the QX9770... although the 9770 had a higher ceiling if i recall.

When i was a tester, i had the fortune to be able to play with all those cpu's.

Are you sure your QX9650 was actually the older stepping because both CO and C1 were out in the wild both sole as QX9650. As for 990X that was probably something minor not made public. I wander if later 970 also receive this upgrade, I guess so. As for 980X, aren't EE CPUs always EOL when the next one takes its place?
ps. your situation with 965 might lend credence that this stepping change was more about keeping power in check then increasing clocking headroom, granted that's only 2 samples. In science statistically meaningless but fortunately we don't stick to scientific rigor in testing and drawing conclusions.

 

[aigomorla]

granted that's only 2 samples. In science statistically meaningless but fortunately we don't stick to scientific rigor in testing and drawing conclusions.

actually if u include the W3580 it was different 3 samples on 5 different cpus the 975.

 

[Lepton87]

actually if u include the W3580 it was different 3 samples on 5 different cpus the 975.

Even the worst 990X was better than the best 980X? How far did your CPU went? I assume 4.3-4.7GHz?

 

[psolord]

Did you guys notice the frame time variance in Tom's gaming tests?

The 5960X is consistently doing worse.

http://www.tomshardware.com/reviews/intel-core-i7-5960x-haswell-e-cpu,3918-6.html

Anyone can guess why?

 

[Idontcare]

Did you guys notice the frame time variance in Tom's gaming tests?

The 5960X is consistently doing worse.

http://www.tomshardware.com/reviews/intel-core-i7-5960x-haswell-e-cpu,3918-6.html

Anyone can guess why?

Would guess it comes down to clockspeeds, base and turbo.

 

[x3sphere]

Pulled the trigger on an X99 build. Honestly the z97 + 4770k combo I have now is fine but I'm thinking this will last a lot longer. Considering I can still sell that setup for a good amount I won't be taking too much of a hit either. I see PCI-E bandwidth limitations being more of a factor with the newer graphics cards coming out in SLI/CF, and I may need more than 32GB of RAM soon (not for gaming, server application).

 

[Dave3000]

This is bang on.

It looks like the 5820k clocks to 4.5Ghz pretty easily, and a mild 4GHz overclock all day long, that's 2 more cores than 4790k at the same clocks on the new platform.

Long term 5820k makes more sense than buying more DDR3 at this time, which will be obsoleted before DDR4.

The i7-4790k's stock turbo boost is 4.2 GHz when all it's cores are active though. So overclocking the 5820k to 4 GHz won't actually match the clocks of the 4790k. 4 GHz on the 4790k is just the stock base clock. You would need to overclock the all-core turbo to 4.2 GHz on the 5820k to match the stock all-core turbo of the 4790k.

 

[Dave3000]

Actualy Xplane 10, Anno 2070 and to a lesser extent F1 2013 are all running better without HT according to Hardware.fr review (excellent as usual, with tests that goes beyond the usual benches).


http://www.hardware.fr/articles/924...a-2011-v3-ddr4-core-i7-5960x-5930k-5820k.html

I currently have an i7-4930k at stock settings. The games I spend most of my time playing are X-Plane 10 and FSX. I wonder if I should switch to the i7-4790k as X-Plane 10 performs about 50% better than it does on the i7-4960x as what I can see from that benchmark on that posted link, so I estimate the 4790k would be around 55% better than my 4930k in that game. It appears that it's more than just IPC increase, clock speed increase, that give the advantage to the 4790k in Xplane 10 over the 4960x. I already saw the reviews on the 5930k and I don't think it's worth upgrading to the 5930k from a 4930k. On any game that fully loads up to 4 cores that would be like my 4930k running at 3.7 GHz vs the 4790k running at 4.2 GHz (about 4.4 GHz Ivy Bridge Performance).

 

[know of fence]

Emphatically YES. What's 144W? That's nothing, my PSU is yawning at the moment and waiting for some action.

First of all I'd recommend a new PSU to go along with a new build, and even significant graphics card upgrades. After all the prudent thing to do is to pick the PSU of the right size, rather than over-provision or leave room for expansion. This can make the PC significantly less wasteful during idle and low power states.
Also +144 W total (plus ~100 W) at the heatsink means at least another 200W for the actual PSU max wattage rating, if you want to stay in the efficient range.

 

[t1gran]

Doesn't i7-5960X have internal GPU (like Intel HD Graphics 4600)? If*not - what does perform graphics without separate GPU?

 

[DigDog]

these two don't seem to be a drastic improvement over my "single-gpu, 4670k" setup;
and not having the 5820k is stupid. most people can't afford to saturate x8 PCI-E 3.0, why would they spend $200 more for the same chip.

gibe 5820k OC review plox or gtfo