Intel Core i9-7900X (14nm Skylake-X)

[Glock24]

It's better than I expected, but not good enough. It's obviously out of it's optimal power/frequency curve o reach those clocks, resulting in high power draw. Still, the Ryzen 7 is not the competitor here. Let's see how it fares against AMD's HEDT platform.

[m_maric]

i9 7900x $1000 thermal paste.
i9 7900x $1100 soldered.

Is that to much to ask INTEL?
Learn from auto industry, give people what they want.

[whatif]

Huh, how long has it been since we've been able to say "Intel cack their pants and rush to launch a product that isn't really ready"?

Impressive work to get such a large processor to clock like the consumer equivalents, mind you - it gives them back the big advantage they had over AMD before Ryzen. OTOH that power draw 50W more than 1800X and 100W more than 1700. It's going to lose that multithreading performance advantage to a moderately clocked Threadripper 16-core that'll probably draw less power too....!

Exactly what I was thinking when I saw the TDP's.
If the middle of the road is sucking 140 TDP without overclocking, how much will the higher end ones suck.
Although my thought went more to whether the new Core i9-7980XE will come with a standard water cooling kit given it has all the bells and whistles and almost twice the cores/threads - especially since it will be overclockable. Besides, for around US$2000 (be closer to $3000 Australian when converted and the extras added on), I would expect some sort of water cooling kit thrown in for free. I would definitely class this as an enthusiast processor.
Be interesting to see how AMD's new Threadripper processors stand up (when they are both sorted out properly of course), hopefully well enough to create some major competition and price drops.

One other thing, the Core i7-7740X only has 16 PCI lanes, makes it hard to run 1x or 2x M.2's with a video card / or try to run 2x video cards, or am I getting something wrong?

[watercooled]

Efficiency was one of the first things I was interested in for these CPUs. No doubt it's an impressive chip (albeit nothing really unexpected), but even against Intel's best, Ryzen is still holding its ground when it comes to actual power draw (BTW Hexus how come you still seem to use TDPs for your bang4watt ratings, they can be wildly inaccurate - some chips do massively better than their TDPs would imply like the i3s) - an area where AMD were pretty much expected to come behind Intel even if they came close in performance.

In an ideal world, it would be nice if AMD could lift per-core clocks for HEDT use, but given their aims to re-use the same die and therefore saving on production costs, it probably didn't look practical for this generation at least. To lift clocks, they'd likely have to re-do the physical design of the chip and have an entirely separate die for just one market segment which defeats the purpose of the whole single-die approach. And of course it's a trade-off, the approach they've taken seems to concentrate more on efficiency at decent clocks for all segments, and right where the server parts will be at.

A part of me wonders what sort of clocks they could get out of TSMC's 16nm, but they'd likely struggle to get any significant volume out of them for a reasonable price, and then they'd have the WSA to worry about again. AFAIK they've had a *lot* of capacity waiting for them at GloFo which they're probably filling now what with Ryzen and their GPUs, to the point I wonder if it makes more sense for Vega to be made at TSMC a) For presumably higher clocks to make it more competitive and b) because they already have experience with HBM. In fact I think I'd be more surprised if Vega wasn't made at TSMC!

But anyway, back to the topic - very good performance and all, but Intel just can't back away from their pricing structure can they? Sure, it performs better than Ryzen but nearly three times the price of the 1700X??? Threadripper could make things really interesting! While I imaging there are some use cases where Intel's on-die bandwidth will shine, in others it really doesn't matter so much...

[Xlucine]

One other thing, the Core i7-7740X only has 16 PCI lanes, makes it hard to run 1x or 2x M.2's with a video card / or try to run 2x video cards, or am I getting something wrong?

The 7700K has 4 extra lanes for the chipset, which can't be used for graphics. So the 7740K will probably have a similar setup

Efficiency was one of the first things I was interested in for these CPUs. No doubt it's an impressive chip (albeit nothing really unexpected), but even against Intel's best, Ryzen is still holding its ground when it comes to actual power draw (BTW Hexus how come you still seem to use TDPs for your bang4watt ratings, they can be wildly inaccurate - some chips do massively better than their TDPs would imply like the i3s) - an area where AMD were pretty much expected to come behind Intel even if they came close in performance.

The 'actual power' column is platform power draw while video encoding (handbrake), as hexus doesn't give power consumption in cinebench so actual bang4watt may differ. All the efficiencies I looked at went down with this modification (understandable with the inefficiency of PSUs, and all the systems tested used the same PSU so it should be fair), but the 1700's lead over the 6950X completely disappeared.

Any chance we could get future multithreaded bang4watt given with handbrake performance, so you can compare actual power consumption in the benchmark? It'd leave the singlethreaded bang4watt the odd one out, but with the intelligent power gating and boost on modern chips it isn't useful anyway. True bang4watt in cinebench would be nice, but it'd mean re-testing a lot of chips - going to handbrake is easier, as the data needed is already recorded.

[watercooled]

So.. what exactly are you measuring in your table? You absolutely cannot use power consumption from one benchmark and performance from another as power draw can vary drastically between applications.

[CAT-THE-FIFTH]

Handbrake power consumption is included in the review.

The Ryzen 7 1700 only has a power delta of 66W to produce 76FPS in Handbrake.

The Core i7 6950X has a power delta of 85W to produce 78.5FPS in HandBrake.

The Ryzen 7 1700 absolutely destroys the Core i7 6950X in this test! Even if you take the absolute power of 113W and 132W respectively its more of the same.

The Core i7 7900X is terrible - a delta of 163W to produce 112.7FPS or an absolute value of 214W.

Even the overvolted Ryzen 7 1800X has a delta of 114W to produce 86.9FPS.

[watercooled]

It does suggest that it really wasn't intended for such high clocks, but given the competitive pressure they've pushed above what they typically would and sacrificed some efficiency in the process - it seems less efficient in real terms than Broadwell. I wonder how much of that is down to clocks and how much is other uncore changes like the new mesh? I remember listening to a podcast where David Kanter was talking about the trade-offs between a crossbar design like Ryzen, and a mesh and IIRC hinted efficiency and die size can take a hit in order to keep latency consistent across the chip.

[Xlucine]

So.. what exactly are you measuring in your table? You absolutely cannot use power consumption from one benchmark and performance from another as power draw can vary drastically between applications.

It's not ideal, but it's more relevant than TDP & the best I can do without actually testing the chips myself

[watercooled]

As CAT pointed out, there are actual handbrake figures presented in the article.

Depending on the benchmak/program it might very well be a worse estimate than the TDP, so you have to be careful. E.g. just on my 7700, 7zip's miltithreaded benchmark draws about 104W at the wall, whilst y-cruncher draws 151W, from 48W idle. Both are 100% loading the CPU, yet the delta goes from 56W to 103W - nearly double for the same CPU!

There's no such thing as a simple 'max power draw' on current CPUs for normal programs - one CPU can draw more than another in one application, but less in another, for a whole host of reasons.

[CAT-THE-FIFTH]

Its quite easy to see what has happened - Intel always were going to release these chips this year,but probably with less aggressive clockspeeds(hence the use of TIM and not solder under the IHS) and maybe a bit later in the year. In order to pre-empt Threadripper they have upped the voltage and clockspeed and brought the launch forward.

[CAT-THE-FIFTH]

Apparently the review guide says SKL-X needs a water cooler even at stock clockspeeds:

https://videocardz.com/70338/intel-c...erclockability

[aniilv]

haha amd got sat with that 4.7ghz

[kompukare]

Apparently the review guide says SKL-X needs a water cooler even at stock clockspeeds:

https://videocardz.com/70338/intel-c...erclockability

So, as many of us here predicted, most of the parts are essentially factory overclocked already. Reminds me of AMD's recent GPU launches, or the FX-9590. Now the telling thing will be whether Intel will now develop a reputation for running hot and be subject to furnace or 'needs-new-powerplant' jokes, or whether those comments are reserved for AMD.

[Ozaron]

Now the telling thing will be whether Intel will now develop a reputation for running hot and be subject to furnace or 'needs-new-powerplant' jokes, or whether those comments are reserved for AMD.

Sharing is caring!

But probably not, since fairly few reasonable people will go out and buy any of the higher end i9 chips. If they go anywhere it should be heavy duty privately owned workstations, and not servers, so not a huge market.

Also bashing AMD is everyone's favourite pastime.

[CAT-THE-FIFTH]

So, as many of us here predicted, most of the parts are essentially factory overclocked already. Reminds me of AMD's recent GPU launches, or the FX-9590. Now the telling thing will be whether Intel will now develop a reputation for running hot and be subject to furnace or 'needs-new-powerplant' jokes, or whether those comments are reserved for AMD.

Its even funnier that all of the crowd on other forums who were bleating on about how games were not threaded and single core performance is the only metric important to consider,and nobody needs 6 to 8 cores for gaming(since Ryzen was far cheaper than the Intel equivalent), have now gone from saying any more cores than what a Core i7 7700K has is useless,to now saying the Core i7 7900X is the bestest CPU for gaming.

Apparently also if you don't match SKL clockspeeds when overclocked to the max,single thread performance is "poor" yet many of them even a few months ago were even saying Haswell,etc had perfectly fine ST performance,which is weird since none of the recent major new releases are actually using one core.

I can appreciate some games run better on Intel but these are the kind of games either a Core i3 7350K/Core i5 7600K would be a better choice for,not a £1000 HEDT CPU with all the added costs.

Plus in the real world I am yet to see any large scale deployment of HEDT CPUs,actually have any of them being overclocked in any way.