12-core Intel Core i9 7920X will get a 400 MHz slower base-clock

I'm just saying that the base clock matters, as it provides a baseline for the overall frequency that a chip can run at.

Frankly, I think it was unwise for Intel to promise up to 18 cores for the Core i9. Based on the thermals and power draw of the 7900X, it's clear that they will have to cut back significantly on the clock to keep within limits, which will lead to minor gains for the higher models. I wouldn't be surprised to see a 2.2 GHz base clock for the 18-core CPU.

 

I can't even fathom the logic here...like really "base clock means absolutely nothing" and "pretty much every single 7900X turbos in all cores at 4GHz minimum" :3eyes:

Leaving user-overclocking out of the picture here- which many users of this class of desktop won't be doing anyway- when you all refer to Base clocks vs Turbo clocks, please keep in mind that Turbo clocks ONLY IMPACT up to 2 cores or 4 threads total. NOT all 10c/20t on 7900X or 12c/24t on 7920X.

Per Hilbert:
"basically with Turbo mode 3.0 less threads will clock higher. E.g. for a low threaded application a processor could boost with 2 cores to say 4.5 GHz. Whereas Turbo mode 2.0 kick in, it would do say 4 threads at 4.2 GHz"
http://www.guru3d.com/articles_pages/intel_core_i9_7900x_processor_review,2.html

Even the Intel chart in the article has an * noting the "dual core" impact of Turbo 2.0 and 3.0

And based on what we have seen from a power consumption/thermal envelope situation from the 7900X...I highly doubt any sensible user will want to manually overclock a 7920X on all cores to any serious degree beyond the 2.9GHz base clock (which is apparently so irrelevant) without a serious custom water cooling solution.

 

Puttimg HBM on an add in card, daughterboard or something would defeat the purpose of having it be close to the processor and it would introduce long wiring traces that would consume energy. It would basically be like adding more DIMMs.

You can get single socket CPUs(Epyc) that support 2TB of DDR4, so there is no real need to have add in cards with HBM.

TSV based 3DS DDR4 offers 2x the capacity of normal 2D DDR4 as well as potentially better performance but its expensive.

https://blogs.synopsys.com/committe...-stacked-dimms-using-through-silicon-vias-tsv

Intel/Micron XPoint PCM offers a massive pool of byte addressable nonvolatile memory that can be used to dramatically improve system performance. They offer a hypervisor like system called Memory Drive that allows a PCI-e Optane SSD to show up as part of the DRAM pool to the system.

https://www.pcper.com/news/Storage/Intel-Officially-Kicks-Optane-Launch-SSD-DC-P4800X

Those are currently available and offer reductions in power consumption with huge increases in capacity and performance.

 

It is a baseline, but CPU always runs higher than baseclock even under full load(unless temp throttling occurs)

As for 12-18core.
I highly expect to see those CPUs soldered which will reduce temps by 25c~ which will allow the higher end products to come with higher factory clock speed.

No, turbo will affect ALL cores.

Turbo will come into affect even under full load if thermals are non-issue.

Less cores under load means higher boost clock for those cores. More cores in use lowers overall turbo bins but they still do go above base clock.

Since the introduction of turbo boost 2.0, pretty much all motherboards have added option to allow max turbo bin of 1 core to apply to ALL cores for performance increase.

That is why you will see every CPU boosting on all cores to the single core turbo bin.

FYI the ALL core turbo speed for the 7900x is 4ghz.

The 7920x will likely have a spec of 3.8ghz all core turbo, but of course with high quality motherboards the all core turbo speed will be even higher.

Here's a link saying basically the same thing.
http://www.pcgamer.com/dont-freak-out-over-the-core-i9-7920xs-relatively-slow-29ghz-clockspeed/

 

As I said, a lower base means lower overall frequency potential. The base clock always matters.

In what sense is the 2.9 GHz base clock for the 12-core 7920X a higher factory clock speed?

Also, I don't see anything that would indicate that Intel would go back to using solder for any of the Core i9 CPUs. If anything, the low base clock of the 7920X only reinforces the likelihood that they will be using paste. On what basis are you making these assumptions?

 

Base clock doesn't matter as much as you make it out to be.

Example past generations, CPU A has base clock of 3.5 but CPU B has base clock of 3.0.

CPU A only has a 200mhz turbo advantage, so in the end the clock speeds aren't as big in real world.

Not quite sure what you are asking here.

10 core 7900x is fully unlocked core.

The 12core will be a cut down 18core(full chip) which i expect will be derivatives of Xeon lineup.

Of course other 'pro' overclockers have made assumptions that they may be soldered.

 

Regarding the 12-18 core CPUs, you said they will be soldered and have higher factory clock speeds. And yet, this very thread is about the 12-core 7920X having a very low base clock speed.
Do you really not see the discrepancy here?

 

No you're just not reading as how i meant it to be.

I am not saying 12-18 core will have a higher base clock than lower end parts.

I only meant by having a soldered CPU would allow them to have higher base clock than if they had used paste instead.

Rumors are the high end parts will be soldered.

An 18core using paste will definitely have to be clocked less than if it was soldered.
Of course time will tell. We won't know until reviews.

 

Why would you expect such nonsense?

You really think Intel would do such a thing, and prove to everyone that they are correct about not soldering causes heat issues? After intel has claimed time and time again there is no difference?

I'm not saying it's impossible they will do that (and get a ****storm backlash if they do it) but that, and much of what you have posted in this forum, makes me wonder what drugs intel is providing you.

https://www.google.com/search?q=int...ZvVAhUC_4MKHShVDMUQ_AUICygC&biw=2560&bih=1310

Again, not sure what you're smoking that makes "we hope they will solder" turn into "rumors are the high end parts will be soldered"

 

I'm relaying information on what I've seen mentioned by deb8eur and other people.

Get off your AMD high horse, I am not defending Intel, so you can keep taking your own AMD drugs

 

That's so obvious that it doesn't even need to be said!

Your contention is that the 12-18 core models will use solder, and yet there is a pretty significant drop in the base clock speed of the 12-core model compared to the 10-core. This suggests strongly that it will not use solder (if it does, then it would seem to have little effect, as the base clock will still be very low). I've never heard of any rumors that the higher-core i9s will use solder, and the 7920X makes such rumors very unlikely. An internet search also reveals nothing so this seems more like wishful thinking on the part of Intel apologists.

 

Has Intel actually claimed there is no difference? I don't even recall them talking about the solder/TIM issue at all. When I was googling it a while back, the only thing I could find from Intel itself was a journal they published back in 2009ish, that talked about Microcracking in solder due to heat density on smaller nodes - which I took as the reason for the switch.

Either way, multiple people delidding x299 chips are claiming a much smaller reduction in temperatures than previous CPUs - only about ~10c drop. They attribute the difference to a smaller gap between the IHS/core due to a thinner glue application.

Should also point out that there are plenty of videos/reports of people delidding soldered CPU's and seeing similar 5-10c drops. So an argument can be made that the TIM application on x299 chips is similar to solder in terms of performance.

 

See above.
Stuff I've seen is just from youtube videos, not some article.

10 core is already hotter than hog balls.

If they were all using paste than the frequency of each would go as follows(starting at 12core).
12c = 2.9 14c = 2.5 16c = 2.1 18c = 1.7

Even if you did a conservative realistic 200mhz drop for each 2 additional cores that leaves the 18c at 2.3ghz..

They won't do that to themselves as that would be shooting themselves in the foot.

But then again even with a low base clock turbo speeds will still be mid to upper 3ghz. Like i said before baseclock doesn't always mean much, usually lowered to keep the same TDP.

That's likely why there is still no ETA on when those high end parts will be released.

Using glue and paste is much faster and cheaper for assembly process...
That is like the only reason to do it.

On these big dies I don't see heat stress cracking these dies in the solder process.

Also deb8eur got close 20c drop on his delided CPU.

Silicon lottery says delid will yield 10-20c.

Those people that do not use a quality metal based TIM will not see a huge drop.
Not enough surface area to properly conduct the heat with regular TIM.

 

I understand that some are speculating about it, but to my knowledge, there's not a single article or leak about it so I think it should be taken with a huge grain of salt - basically, assume that it's false until proven otherwise.

That's pretty much what I expect the 18c to be, based on the base clocks of the 7900X and 7920X. As you said, clock speeds typically decrease as more cores are added. I'd expect a 2.3 GHz base and perhaps a 2.8 GHz boost (boosting 18 cores is a lot harder than 10 so I expect the boost clock to drop off as well).

The thing is, I don't think they can really go faster than that without blowing up the power consumption (even with solder). It'll still be a very good multithreading chip, although probably not as good as Threadripper.

 

Base clocks doesn't matter?? LOL.... the joke of the century..... I'm not even arguing against such non-sense.



The solder-vs-tim accounts for probably 70% of the reduced clocks but there's other factor here u are forgetting.
More cores = bigger die = lower common denominator. The individual clock of a chip (delta from architecture+fab avg) depends on the amount of "minor defects" on the die. It's just math, a bigger die has a bigger chance to get those minor defects. Finding a "perfect" one capable of 4Ghz on lower V. becomes increasingly difficult.
AMD approach allows them to bin much smaller dies of only 2 cores (iirc) so they can make a "perfect" TR by "gluing together" only the good parts.
I'm just stating the obvious here but seems this was lost at some point in the discussion.