Register a SA Forums Account here!
JOINING THE SA FORUMS WILL REMOVE THIS BIG AD, THE ANNOYING UNDERLINED ADS, AND STUPID INTERSTITIAL ADS!!!

You can: log in, read the tech support FAQ, or request your lost password. This dumb message (and those ads) will appear on every screen until you register! Get rid of this crap by registering your own SA Forums Account and joining roughly 150,000 Goons, for the one-time price of $9.95! We charge money because it costs us money per month for bills, and since we don't believe in showing ads to our users, we try to make the money back through forum registrations.
 
  • Post
  • Reply
eames
May 9, 2009



Comet Lake-H, Intelís response to Renoir:

https://www.hd-tecnologia.com/estas...ara-portatiles/

Adbot
ADBOT LOVES YOU

eames
May 9, 2009



gradenko_2000 posted:

so if Rocket Lake is expected to have Xe graphics, and that very early leak from Tiger Lake, which does have Xe graphics, suggests performance that's comparable to a Ryzen APU, doesn't that create a situation where any non-F Rocket Lake CPU is going to be able to drive graphics at a level similar to, say, an Athlon 3000G or a Ryzen 2200G?

That seems kind of interesting, I'm kinda thinking from the perspective of even office computers with a Rocket Lake Pentium having a decent iGPU.

Meanwhile Comet Lake on the desktop is still going to be using UHD, right?

Yep. Itís also interesting because the dedicated Intel GPU card should have 2x or 4x the performance of a single Xe iGPU due to the rumored chiplet design, though there will be differences in CUs and certainly frequency.

eames
May 9, 2009



Igors Lab found some Comet Lake leaks.

PL1 170W, PL2 224W on a i9-10900F with 65W advertised TDP and thatís not even the top end -K SKU.
Clearly thereís not a lot left in 14nm, though I almost canít believe that Intel (or the motherboard vendors) would ship these numbers as the default settings.

eames
May 9, 2009



gradenko_2000 posted:

I don't really have a perspective on how demanding those kinds of numbers are: if you were going to air-cool something like that, what would you be using?

The power consumption numbers are quite similar to the "special edition" 9900KS.

It can be air cooled with any of the better air coolers, particularly since the Die size is larger so you have more area to dissipate the heat from, but these are very high numbers for what was historically a relatively efficient (non-K, fixed multiplier) part of their lineup.

One can only assume that the top end -K model will have very little overclocking headroom or extreme power consumption (>300W), or both.
It does at least explain the new socket, though my understanding is that Rocket Lake will need a new chipset again?

eames
May 9, 2009



momomo_us appears to have found some Z490 prices starting at 200Ä, no idea who's supposed to buy these when Zen 3 is out in a few months.

https://twitter.com/momomo_us/statu...512721276760065

eames
May 9, 2009



Ihmemies posted:

I have a 8700K with Z370 chipset and Asus Maximus X Hero. Currently I have a F4-3600C15D-16GTZ kit. So 2x8GB CL15. I could get another of those F4-3600C15D-16GTZ kits for 170Ä new. They are samsung b-die.

Or I could get a 32GB kit like F4-3600C16Q-32GTZ for 335Ä (4x8GB).

How important it is to have a matched kit? I had two 3x2GB kits in my old i7 920 (6 sticks total) and it ran just fine on advertised specs. Would 2 separate 2x8GB CL15 kits run fine on 8700k? If I bumped the CL up to 16 or something. I could save money and hassle by just getting another 16GB kit.

My first hand experience is that the RAM will work but the XMP profiles won't. I have the exact same CPU/MB and bought a 2x8GB b-die first and an identical 2x8GB kit a few months later.

XMP didn't work and the generic "auto" settings were very loose/slow. It took a lot of tweaking and hours of stability testing to get good performance out of the two identical sets and if I had to do it all again, I'd definitely sell the existing kit and buy a full kit from the motherboard's QVL.

If leaving some performance on the table doesn't bother you, you might be fine. Technically 4x8GB at CL16 might outperform 2x8GB at CL15 in synthetic benchmarks because of interleaving and that particular motherboard works best with 4 DIMMs. On the other hand if one of the memory training algorithms won't train the secondary or tertiary timings correctly, you're in for a lot of frustration.
Due to the nature of memory training you can end up in situations where the system is stable 9 out of 10 coldboots and that last one is unstable. Then you get to choose from a few dozen parameters to fix the issue. Yeah, I'm never touching my memory settings again.

eames
May 9, 2009



NewFatMike posted:

Apple also make extremely performant ARM cores as well, and having locked down the hardware side, they have a distinct advantage to move to another CPU architecture.

Yes, and they also get to design their own integrated GPU. It would be limited to their own Metal API, but I'd expect the performance to be absolutely groundbreaking in its segment.

eames
May 9, 2009



Cygni posted:

New socket. For the last decade, Intel has given you 2 generations per socket and thats it. So you can probably expect to get Rocket Lake on this same socket next year, and thats it. To be fair, after next year, both AMD and Intel will be forced to new sockets for DDR5.

Yeah, there are rumors about a 7nm Meteor Lake floating around already, with a new socket again (LGA1700). I donít think the launch Z370 boards would be able to handle the new -K parts, I expect those power consumption numbers to be quite up there.

eames
May 9, 2009



Modern Macs already boot from an Apple designed ARM chip. One could make an argument that the Intel CPUs in Macs are already co-processors.

The co-processor approach is interesting to fantasize about, particularly with small x86-64 chips as co-processors that only spin up to natively run legacy code. This would retain backwards compatability while teaching users to avoid the old "inefficient" (primarily battery lifetime) architecture.
I imagine the current suppliers wouldn't be too pleased about this but there are others that wouldn't mind selling a few cheap, small, low power x86-64 cores as custom chips.

eames fucked around with this message at 08:58 on Apr 30, 2020

eames
May 9, 2009



I assume the Comet Lake NDA was lifted earlier today.

10900K apparently comes with a stock PL2 of 250W for 56 seconds (9900K was 128W for 28 seconds). I can understand the need for a new socket now, although the overclocking headroom is likely going to be smaller than it was.

The new 10 series also allows users to enable/disable Hyperthreading per core as a security overclocking feature.

obligatory Anandtech link:

https://www.anandtech.com/show/1575...et-lake-desktop

eames
May 9, 2009



The computerbase article has more detailed information, but you'll have to machine translate it unless you can read german:

https://www.computerbase.de/2020-04...aunch-lga-1200/

eames
May 9, 2009



Computerbase figured out that all -K SKUs and the 10400/-F seem to use the 10C Die with a soldered IHS, while everything below that seems to use the 6C Die. They updated their article.

They also note that the 10700K at 95W TDP-down has a 100 MHz lower baseclock than the 9900K, which would make it a less efficient CPU. That would be strange.

eames
May 9, 2009



Paul MaudDib posted:

MSI discusses Intel Comet Lake-S processors binning and overclockability

203W stock, 347W at 5.1 all-core.

I think I remember reading that the 10700 series will be 10C with disabled cores. It also looks like the 10900K/KF are getting the cream of the crop as far as binning, an "average" 10900K should do 5.1 all-core at about 1.25V.

Seems like WCCF made a mistake there or the slide is wrong, because it clearly states 203W stock/240W at 51x. The 347W number is total system power.
Iím sure just was a honest mistake, WCCF is not known for manufacturing sensational content... right?

Those numbers are high but better than I expected. I suspect they would still go up into the 300s with synthetic workloads, because thatís just Cinebench we are seeing here.

It was Computerbase that speculated that all -K SKUs use 10C die. That die seems to be the only one that gets soldered TIM, the lower Z-height and the thicker IHS to make up for it. It could make the six and eight core enthusiast models quite decent overclocking candidates due to the lower heat density.

Cygni posted:

Obviously wait for reviews and such, but I guess what I'm saying is I am real skeptical that you really need any of these $300+ boards even if you getting the 10 core.

Yeah, I totally agree. The manufacturers made mistakes with some Z370 boards using Ącosmetic heatsinksď, so VRMs and cooling became a big topic, even though most boards still did ok.
Z390 already fixed that for the most part and now Z490 seems to double down on it because by now the topic has become a huge (if not the deciding) factor in reviews and thus marketing. One can only hope that overall MB quality doesnít suffer from this development.

eames fucked around with this message at 06:26 on May 7, 2020

eames
May 9, 2009



The buildzoid video on selective hyperthreading is quite interesting.

The idea, as he explains it, is that it allows you to pick up extra overclocking headroom by selectively disabling hyperthreading on the "weak" cores of a CPU. 10C/16T at 5.2 GHz will usually be faster than 10C/20T at 5.1 GHz, so disabling HT on the right cores can be a overall performance gain.

eames
May 9, 2009



Paul MaudDib posted:

Canít you set a per core multiplier anyway, so you could just run that one core at 50x and the rest at 51x (lets say)? Why wouldnít that be better than turning off HT on the core?

I'm not sure if Z490 can do that. I believe the newer Intel HEDT platforms can do it and I know Coffee Lake on Z370 can't.

Selective HT might still be better because the effect of Intel's SMT implementation is negligible (sometimes even negative) for some workloads while higher frequency always improves performance. This seems to be especially true past a certain thread count. To illustrate the point, you'd probably rather run the average 2020 game engine on a 5 GHz 32C/32T CPU than a 4.8 GHz 32C/64T CPU.
Buildzoid's whole explanation is aimed at video game titles and benchmarks that stop scaling with threads at some point. I imagine Ahmdal's law also ties into that.

Your approach ó if technically possible on Z490 ó makes sense for workloads that show perfectly linear scaling. Those workloads tend to run best on HEDT platforms which already have that feature, so maybe the engineers had that in mind.

Adbot
ADBOT LOVES YOU

eames
May 9, 2009



Computerbase benchmarked Coffee Lake vs Ryzen with various memory speeds and two different GPUs.
The Intel chips still like fast, optimized RAM, particuarly when overclocked. Most sites only publish stock vs stock results, but the overclocked comparison shows a sizeable gap in both FPS and frametimes.

here's the link: https://www.computerbase.de/2020-05...-9-3900x-ram-oc

e: attachment is way too large but the post form doesn't give me an option to modify it.

Only registered members can see post attachments!

eames fucked around with this message at 09:27 on May 19, 2020

  • 1
  • 2
  • 3
  • 4
  • 5
  • Post
  • Reply