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
ConanTheLibrarian
Aug 13, 2004


dis buch is late

Fallen Rib

Intel want to boost performance so they hire a guy who retired before the whole Spectre mitigations business

Adbot
ADBOT LOVES YOU

DrDork
Dec 29, 2003
commanding officer of the Army of Dorkness

ConanTheLibrarian posted:

Intel want to boost performance so they hire a guy who retired before the whole Spectre mitigations business

At least they're not trying to roll all the way back to NetBurst and hoping that 5nm or whatever is what lets them finally get to 10Ghz.

mobby_6kl
Aug 9, 2009

"You are the best poster... do not let anyone say otherwise."


Beef posted:

At first I'm excited about Intel doing another high performance CPU project. But then I realize we're only going to hear from it in 3 to 5 years, if at all
There's the graph processor Intel is working on, but I doubt it's something the general public is going to be programming.
They didn't necesserily just start the project, they hopefully had enough sense to have various ongoing projects.

Cygni
Nov 12, 2005

raring to post



So the 11900K is +300mhz (with TVB) over the 11700K, but thats it. Kinda back to the old days of the 2700K vs 2600K, or the X Ryzen SKUs.



https://videocardz.com/newz/intel-core-i9-11900k-i7-11700k-and-i5-11600k-rocket-lake-s-specifications-confirmed-by-msi

Nomyth
Mar 15, 2013

And if a Nyto get a attitude
Pop it like it's hot
Pop it like it's hot
Pop it like it's hot


Imagine the year 2016 in an alternate timeline where we got Rocket Lake instead of Kaby Lake

Twerk from Home
Jan 17, 2009

This avatar brought to you by the 'save our dead gay forums' foundation.


Cygni posted:

So the 11900K is +300mhz (with TVB) over the 11700K, but thats it. Kinda back to the old days of the 2700K vs 2600K, or the X Ryzen SKUs.



https://videocardz.com/newz/intel-core-i9-11900k-i7-11700k-and-i5-11600k-rocket-lake-s-specifications-confirmed-by-msi

If they hold pricing, the i7 and i5 should both be pretty competitive against AMD. Is Intel expected to actually hard launch, with CPUs easy to find at MSRP?

mobby_6kl
Aug 9, 2009

"You are the best poster... do not let anyone say otherwise."


Twerk from Home posted:

If they hold pricing, the i7 and i5 should both be pretty competitive against AMD. Is Intel expected to actually hard launch, with CPUs easy to find at MSRP?

Thankfully they're using 14nm++++ for these so they should be rolling out of the factories at record speeds. Right?

Cygni
Nov 12, 2005

raring to post



Twerk from Home posted:

If they hold pricing, the i7 and i5 should both be pretty competitive against AMD. Is Intel expected to actually hard launch, with CPUs easy to find at MSRP?

With their server stuff moving to 10nm this quarter, hopefully that frees 14nm production for Rocket Lake... but the Z590 board pricing has been as gougy as X570 was at launch (if not more so), so i doubt its gonna be much of a "value" over the AMD side.

Both platforms are effectively deadends but that doesn't really imply they are bad purchases. DDR5 is probably gonna launch around 4800 speeds with atrocious latencies based on the current info. The JDEC spec for DDR5-4800 is CL34-42, and the sticks are expected to be double the cost at launch. And with PCIe 5 likely having no consumer SSD or GPU impact in the moderate to long term, and a lot of the Z590 boards having TB4 plugs that will work with USB4 devices, a X570/Z590 board and fast DDR4 may be a good play while weathering the RAM prices and likely low availability of 10nm SuperFin for Intel and 5nm TSMC for Zen 4 on the next gen.

gradenko_2000
Oct 5, 2010



Lipstick Apathy

Cygni posted:

Both platforms are effectively deadends but that doesn't really imply they are bad purchases. DDR5 is probably gonna launch around 4800 speeds with atrocious latencies based on the current info. The JDEC spec for DDR5-4800 is CL34-42, and the sticks are expected to be double the cost at launch.

What kind of stats should we expect before DDR5 has good enough latency to be useful relative to what can be gotten out of good DDR4?

Paul MaudDib
May 2, 2006

"Tell me of your home world, Usul"


Nomyth posted:

Imagine the year 2016 in an alternate timeline where we got Rocket Lake instead of Kaby Lake

Comet Lake was a complete mistake in multiple respects, now people are going to get upset about losing two cores (even if the drop in MT performance is zero or small, while being vastly better in most other task due to ST bumps, it's bad optics) and it's basically only leapfrogging Zen3 by a little bit. It probably flat-out would have been better not to launch it and to just drop the price on 9900K a bit. And there's the mess with PCIe 4.0 boards as well.

if rocket lake had launched a year after the 9900K (so like, september/october 2019) in their usual launch window, the conversation would be drastically different. The narrative would be more like "yeah Zen3 almost caught up to it, and it's more efficient, but it's a year late and blah blah". Now Intel is "leapfrogging by a little bit while being vastly more inefficient".

I realize if it's not ready then it's not ready, but products do have "practical" launch windows dictated by market conditions (eg "voodoo 5 is delayed for a year and then nobody cares anymore") and Rocket Lake has pretty much blown its window of market relevance. Maybe it would have worked even as late as Spring 2020, and just drop 9900K/8700K prices in the meantime, but Rocket Lake absolutely had to come before Zen3, coming 9 months after Zen3 is just lol at this point.

Twerk from Home posted:

If they hold pricing, the i7 and i5 should both be pretty competitive against AMD. Is Intel expected to actually hard launch, with CPUs easy to find at MSRP?

mobby_6kl posted:

Thankfully they're using 14nm++++ for these so they should be rolling out of the factories at record speeds. Right?

It's probably going to be about as hard a launch as anything can be in Mr Freeze's Hellworld. Fortunately most of the other products have moved off 14nm at this point. Intel is ramping Ice Lake-SP this quarter and at the end of the quarter they're launching 10nm-based 8c laptop chips. That basically leaves cost-reduced 14nm server and laptop chips, and desktop products (of all price ranges) on 14nm, along with miscellaneous poo poo like chipsets.

Like, these days it seems practically a given that stuff is going to sell out on launch day, and right now it practically seems like a given that stuff is going to sell out for 3+ months, but I don't think it should be any worse than Ryzen 5000 (and hopefully better) given Intel has a lot of spare 14nm capacity nowadays. It really is not hard to go buy a 14nm chip right now unless you are fishing for the 10900K (which as the very highest bin is a little scarce vs the 10850K) and once the initial launch shortages clear it'll probably be fine.

Of course Zen3 is starting to stabilize too, and Intel is going to have their launch shock just at the moment when Zen3 is stabilizing...

Paul MaudDib fucked around with this message at 01:17 on Jan 28, 2021

Cygni
Nov 12, 2005

raring to post



gradenko_2000 posted:

What kind of stats should we expect before DDR5 has good enough latency to be useful relative to what can be gotten out of good DDR4?

I think youll get gamer spec kits with way lower CAS than the JDEC spec, just like now (JDEC spec for DDR4 3200 is CL22, lol), but its probably gonna be real pricey or take some time to show up. I'm basing that purely on past launches, no insider info on the DDR5 ramp.

But for example, I jumped on DDR4 early and the hottest kit I could find in stock was DDR4 2400 CL15 for some insane $250+ price, and the price equivalent DDR3 at the time was straight up faster. Thats mostly cause servers generally drive memory adoption, and generally care more about bandwidth than latency. Games, unfortunately, care about both. The ratio of what matters more depends on the game and the memory controller, but memory latency is basically the only saving grace of Intel's architectures vs AMD at the moment, if that tells you anything about its value. Comet Lake beats Zen3 in places where latency matters, and loses to Zen3 where it doesn't (yes this is a huge generalization, i know).

The hottest memory kit on the market for latency at the moment is a DDR4 4400 kit with CL16, giving a 7.2ns first word latency. To beat that with DDR5 4800, you are looking at CL18 which uh... aint gonna exist at first, at least. The bandwidth difference in that example is 35.2GB/s on the DDR4 vs 38.4GB/s on the DDR5, which isn't a huge diff that the latency benefits on the DDR4 will probably be enough to overcome in benchmarks.

It's not the end of the world, latency has basically plateaued since the DDR1 era, but I do think it helps DDR4s current value proposition.

Cygni fucked around with this message at 03:02 on Jan 28, 2021

BlankSystemDaemon
Mar 13, 2009

System Access Node Not Found



Yeah, it's an even bet that that Micron, Samsung, or SK Hynix (the only three manufacturers left, who sell globally) will bin memory ICs in such a way that they or their customers (Kingston, HP, IM, et cetera ad nauseum) will be able to get away with a lower strobe frequency.

The reason why strobe frequency was so important back in the DDR1 days is that if you went from 1 to 2, you effectively halved the commandrate of the memory, cutting your bandwidth in half. However, the good news is that while strobe frequency keeps climbing as memory gets faster, the difference between commandrates decreases asymptotically.

Palladium
May 8, 2012


According at my ancient retail pricelists, DDR1 was somewhere like 3x more expensive per MB than PC133 SDRAM in 2000 (also, drat were those 600-700MHz Durons cheap, the mobos were priced twiced as much)

Looking back, I don't think there was ever a time where being an early adopter for a new DRAM gen made any sense from a cost-benefit PoV, maaaaaybe perhaps an exception for DDR4 + X99.

ConanTheLibrarian
Aug 13, 2004


dis buch is late

Fallen Rib

Any recollection about how long it typically took for the new mem tech's prices to settle down to something reasonable? Since next year's CPUs will be the first to use DDR5, I've mentally pencilled in the follow gen as a decent time to buy in, but that's assumes a quick enough drop in RAM prices.

DrDork
Dec 29, 2003
commanding officer of the Army of Dorkness

Yeah, somewhere around a year is usually a good marker for when you can expect things to be "matured" enough for prices to have dropped and speeds to have increased enough to make sense buying in.

Beef
Jul 26, 2004


ConanTheLibrarian posted:

Any recollection about how long it typically took for the new mem tech's prices to settle down to something reasonable? Since next year's CPUs will be the first to use DDR5, I've mentally pencilled in the follow gen as a decent time to buy in, but that's assumes a quick enough drop in RAM prices.

Depends on https://en.wikipedia.org/wiki/List_of_earthquakes_in_Taiwan

SwissArmyDruid
Feb 14, 2014



Why the crap are NUC8s still loving $550 used in 2021?

NewFatMike
Jun 11, 2015



Because demand exceeds production? And there is a God, but He is cruel?

SwissArmyDruid
Feb 14, 2014



Either that, or people are still sitting on them, rather than flipping them.

Looks like I may have to go the AsRock Deskmini route instead.

And to think that what started me on all this was that I wanted was a loving Netflow collector that was more performant than a raspberry pi...

Cygni
Nov 12, 2005

raring to post



https://twitter.com/VideoCardz/status/1356249089541894147





Make sure your "14 core" is the 8-6 config and not the 6-8 config because both will exist and yes this is insane!!

CFox
Nov 9, 2005


Why on earth are there so many small cores? I figured they'd set an upper limit of 4 on the things since they'll mainly be used for background processes and the like.

Cygni
Nov 12, 2005

raring to post



CFox posted:

Why on earth are there so many small cores? I figured they'd set an upper limit of 4 on the things since they'll mainly be used for background processes and the like.

They are physically small on the die, so its not a huge deal to give enough of em to be meaningful. And with win10, there are a lot of background things to run. The vast majority of the die size (75%+) is dedicated to graphics, encoding/decoding, and the TB4 controller. The small cores are gonna be a very small portion of the die.

But because they are so small, i really am wondering why they are bothering offering these stepped SKUs that disable 2 at a time. The chance of a defect on specifically a pair of those tiny units has to be so small, how is it even worth harvesting? I could see a yes/no on the little cores being a worthwhile harvest, especially for low end desktop devices where the little cores wont matter as much, but this seems like... a lot.

SwissArmyDruid
Feb 14, 2014




Videocardz posted:

The test on Geekbench wasn’t focused on CPU power, but rather the integrated Xe Graphics. The CPU offers 96 Execution Units (GT2 graphics) clocked at 1.15 GHz. The iGPU scored 13446 points which is comparable to GeForce GTX 660 Ti desktop.

It took almost seven years, but we're finally almost approaching the point where integrated graphics will be equivalent to a bus-powered 750 Ti, Praise Moore. </sarcasm>

SwissArmyDruid fucked around with this message at 20:00 on Feb 1, 2021

Perplx
Jun 26, 2004


Best viewed on Orgasma Plasma

Lipstick Apathy

Intel showing some real innovation in sku creation, apple doesn't have the courage to ship 18 different mobile cpus in a year.

BobHoward
Feb 13, 2012

The only thing white people deserve is a bullet to their empty skull


Cygni posted:

But because they are so small, i really am wondering why they are bothering offering these stepped SKUs that disable 2 at a time. The chance of a defect on specifically a pair of those tiny units has to be so small, how is it even worth harvesting? I could see a yes/no on the little cores being a worthwhile harvest, especially for low end desktop devices where the little cores wont matter as much, but this seems like... a lot.

It's because intel's product planning team loving LOVES artificially segmenting things to ridiculous degrees, OP

Also why is there no 8 big + GT2 option, lol

p.s. sometimes it's not really harvesting. Time under test is money, and test can be a surprisingly large fraction of the manufacturing cost. So in this case, maybe Intel steps back to testing only 6 small cores as soon as they've filled the order sheet for 8-small-core chips, then shifts to testing 4 after filling the 6-core sheet, and so on.

Obviously that's hypothetical. Test programs can be optimized, e.g. I'd want to try to test cores in parallel if I could. I've never been directly involved in test and I've never worked for Intel, so I have no idea whether they could or did do that for Alder Lake. But I can tell you that test economics are absolutely a real thing and sometimes the reduced functionality version of a part could've been yielded as 100% functional, it just wasn't tested that way.

Still doesn't excuse the silly degree to which Intel plays the game. They make dozens of SKUs from every tapeout and routinely piss customers off. They got away with it for ages due to monopoly power, but ought to back off now that it's very clear ARM is a threat to overturn x86 hegemony. But, as is often the case, it's difficult to change the course of a dinosaur.

ConanTheLibrarian
Aug 13, 2004


dis buch is late

Fallen Rib

CFox posted:

Why on earth are there so many small cores? I figured they'd set an upper limit of 4 on the things since they'll mainly be used for background processes and the like.
I was sceptical about the big little thing for desktop, but I think it could work out pretty well if handled correctly. If you take games as an example, worker threads taking care of less computationally expensive stuff could be distributed across lots of little cores that use gently caress all power, leaving more of the CPU's thermal budget for the big cores running the main thread(s).

BobHoward posted:

Also why is there no 8 big + GT2 option, lol

Not enough die space is my guess.

Inept
Jul 8, 2003

3 olives has a payment plan with his phone company for his cell phone

What better way to convince people that efficiency cores are good and useful than to offer a stupid amount of configurations of them, including none at all.

BobHoward
Feb 13, 2012

The only thing white people deserve is a bullet to their empty skull


ConanTheLibrarian posted:

Not enough die space is my guess.

Since GT2 is available with 6 big cores, that only works as an explanation if the 8-big and 6-big parts are different tapeouts. Which would seem like an odd decision.

SwissArmyDruid
Feb 14, 2014



SwissArmyDruid posted:

Either that, or people are still sitting on them, rather than flipping them.

Looks like I may have to go the AsRock Deskmini route instead.

And to think that what started me on all this was that I wanted was a loving Netflow collector that was more performant than a raspberry pi...

I was wrong. On a whim, I decided to go see if Dell had anything on outlet.

WHY ARE 9TH GEN REFURB SFF BOXES GOING FOR CHEAPER THAN IDENTIAL 8TH GEN BOXES AND NUCS USED?

Was there actually something special with 8th Gen that I don't know about, or did everyone just shy the gently caress away from them because they're just the same silicon but with the voltage jacked?

SwissArmyDruid fucked around with this message at 23:14 on Feb 1, 2021

WhyteRyce
Dec 30, 2001



BobHoward posted:

It's because intel's product planning team loving LOVES artificially segmenting things to ridiculous degrees, OP

Still doesn't excuse the silly degree to which Intel plays the game. They make dozens of SKUs from every tapeout and routinely piss customers off. They got away with it for ages due to monopoly power, but ought to back off now that it's very clear ARM is a threat to overturn x86 hegemony. But, as is often the case, it's difficult to change the course of a dinosaur.

Define customers. Because while Intel loves artificially segmenting stuff to the nth degree, OEMs also love to laser target particular price points and start some kind of intricate dance of feature/price negotiation on both sides.

Agreed they go overboard to the nth degree but random OEM asking for oddball skus that make no sense that no one else asked for so they can hit some Walmart special is not unusual.

WhyteRyce fucked around with this message at 23:39 on Feb 1, 2021

ConanTheLibrarian
Aug 13, 2004


dis buch is late

Fallen Rib

BobHoward posted:

Since GT2 is available with 6 big cores, that only works as an explanation if the 8-big and 6-big parts are different tapeouts. Which would seem like an odd decision.

Good point. Rechecking the article, it says that's "a chart with all Alder Lake-S and P series configurations that has already been leaked". So there could yet be an 8-8-2 config to come!

canyoneer
Sep 13, 2005


I only have canyoneyes for you


-P is stored in the Alder Lake

mobby_6kl
Aug 9, 2009

"You are the best poster... do not let anyone say otherwise."


Aldar Lake? Why can't I still get a Tiger Lake replacement for my work laptop, Intel?

Cygni
Nov 12, 2005

raring to post



mobby_6kl posted:

Aldar Lake? Why can't I still get a Tiger Lake replacement for my work laptop, Intel?

I think they are fairly plentiful now, or are they still hard to find? I looked just now and Dell has a few shipping immediately and a bunch shipping in 1-2 weeks.

This 10nm rollout is obviously gonna be interesting considering the Ice Lake Xeons are just now showing up, and there is a ton of people waiting on Sapphire Rapids which should be ramping now.

Encrypted
Feb 25, 2016



lol that a company that's bad at modern efficient mobile cpu is going to make a cpu with different big.little cores and hand off the software side to a company that's bad at making efficient modern mobile os/software that will most likely require tight integration for the big.little to work well in the first place

things seem to be breaking at the seams already by the look of having 8+2 and 2+8 for ~*reasons*~


and they haven't even released anything yet!

SwissArmyDruid
Feb 14, 2014



Does that assume that they are pre-binning their chiplets before they mount them, or are they mounting everything and fusing off dysfunctional cores? I think the latter is more likely.

gradenko_2000
Oct 5, 2010



Lipstick Apathy

If AMD can claim the FX 8100 was an eight core CPU I'm sure there's not going to be any problems with people getting confused between 8+2 and 2+8 configurations

Right?

BlankSystemDaemon
Mar 13, 2009

System Access Node Not Found



Encrypted posted:

lol that a company that's bad at modern efficient mobile cpu is going to make a cpu with different big.little cores and hand off the software side to a company that's bad at making efficient modern mobile os/software that will most likely require tight integration for the big.little to work well in the first place

things seem to be breaking at the seams already by the look of having 8+2 and 2+8 for ~*reasons*~


and they haven't even released anything yet!
wait until it turns out that like ARM big.LITTLE, the differently sized cores implement different instruction sets

taqueso
Mar 8, 2004









however, a big core can translate and feed a little core

Adbot
ADBOT LOVES YOU

BlankSystemDaemon
Mar 13, 2009

System Access Node Not Found



little cores can have translations from big cores, as a treat

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