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
Combat Pretzel
Jun 23, 2004

No, seriously... what kurds?!
An 8-core Skylake-X at 600 bux or less would be nice, but I don't see it happening. Too much hubris over at Intel.

Adbot
ADBOT LOVES YOU

Canned Sunshine
Nov 20, 2005

CAUTION: POST QUALITY UNDER CONSTRUCTION



Watermelon Daiquiri posted:

i still find it funny people are legitimately upset and disappointed that AMD 'only' managed to get within like 15% of Intel's latest offering of a decade long refinement process (thats in turn based on an already established design skeleton!) after only a few years of alpha and beta development on a complete brand spanking new design (unless zen is a refinement of the old k6s or whatever)

Yeah, I think some people on this forum built up an idea that Ryzen would instantly surpass all of Intel's offerings, and when it didn't, apparently the release became a greater disappointment than the Titanic.

eames
May 9, 2009

Combat Pretzel posted:

An 8-core Skylake-X at 600 bux or less would be nice, but I don't see it happening.

:same: with ECC support but...

Combat Pretzel posted:

Too much hubris over at Intel.

BIG HEADLINE
Jun 13, 2006

"Stand back, Ottawan ruffian, or face my lumens!"
Finally some new info on Sky-X: http://www.guru3d.com/news-story/specs-unreleased-high-end-core-i7-and-core-i9-7900-series-processors-hit-the-web.html

Fuckers aren't putting 44 lanes on anything shy of the 10C part. :(

Beautiful Ninja
Mar 26, 2009

Five time FCW Champion...of my heart.
Those clocks on the higher end parts are really high for such dense CPU's. Same boost speeds as a 7700k on the 8c/10c parts.

If real, it gives me hope that the rumored 6c/12t mainstream parts can hit clocks similar to what the 7700k is hitting right now.

Paul MaudDib
May 3, 2006

TEAM NVIDIA:
FORUM POLICE

Beautiful Ninja posted:

Those clocks on the higher end parts are really high for such dense CPU's. Same boost speeds as a 7700k on the 8c/10c parts.

If real, it gives me hope that the rumored 6c/12t mainstream parts can hit clocks similar to what the 7700k is hitting right now.

Skylake clocked a little lower than Kaby Lake overall, so I think a reliable 5 GHz is a little too much to expect. Unless Intel backported some of the tweaks from Kaby onto the Skylake-X die, or is straight-up running this on the improved Kaby Lake 14nm+ process, I guess. I'd rate that unlikely but not impossible.

The HEDT chips have always been clocked really conservatively, which probably increases yields for Intel. Haswell-E would reliably do 4.5 GHz with extra voltage, it just runs really hot (200W+). Golden samples could go to 4.7 GHz and weren't too incredibly uncommon. The numbers sound high for factory clocks, and especially compared to what Intel usually does on their HEDT lineup, but it could be done if they are willing to divert some of the top-binned low-leakage chips and hit them with some voltage. So I'd rate this one as "unlikely but not impossible" too. If so, AMD is gonna clown on them for their power consumption, but Intel will have the last laugh on performance.

If 6C parts with some extra voltage were reliably hitting 4.6 GHz, often hitting 4.7, and golden samples were hitting 4.8 then I'd say that was pretty great. I'd even be happy with lowering all those clocks by 100 MHz - that would be Haswell-E clocks with Skylake IPC, which would still be quite a ways ahead of Ryzen's single-thread performance.

Paul MaudDib fucked around with this message at 22:06 on May 12, 2017

Combat Pretzel
Jun 23, 2004

No, seriously... what kurds?!
What's with the cache? The 8MB on the 4C/8T and 8.25MB on the 6C/12T. And jesus, shifting the high lane count to the 10C one, that's mean.

BIG HEADLINE
Jun 13, 2006

"Stand back, Ottawan ruffian, or face my lumens!"

Combat Pretzel posted:

What's with the cache? The 8MB on the 4C/8T and 8.25MB on the 6C/12T. And jesus, shifting the high lane count to the 10C one, that's mean.

Especially since it'd stand to reason they'll charge $999 minimum for the 10C/20T CPU, and probably closer to $1299.

They should've made a 7820X and a 7850X, both 8C/16T, but give the 50X the 44 lanes.

Anime Schoolgirl
Nov 28, 2002

BIG HEADLINE posted:

Finally some new info on Sky-X: http://www.guru3d.com/news-story/specs-unreleased-high-end-core-i7-and-core-i9-7900-series-processors-hit-the-web.html

Fuckers aren't putting 44 lanes on anything shy of the 10C part. :(
"i9"

this naming scheme is knee-slapping

Yaoi Gagarin
Feb 20, 2014

Does anyone remember that article that showed that RAM frequency was important for frametime consistency on Intel CPUs from the last few years? I think it at least covered Sandy bridge through haswell, and basically demolished the old "ram frequency doesn't matter at all" mantra that was common for a long time. Was written about one or two years ago iirc

craig588
Nov 19, 2005

by Nyc_Tattoo
I think it was 2400 15 where the diminishing returns hit real hard. At least that probably was the reason I bought my memory rated for that. Setting it to 2666 14 doesn't feel any faster, but the XMP setting is good for it, so why not?

LRADIKAL
Jun 10, 2001

Fun Shoe

VostokProgram posted:

Does anyone remember that article that showed that RAM frequency was important for frametime consistency on Intel CPUs from the last few years? I think it at least covered Sandy bridge through haswell, and basically demolished the old "ram frequency doesn't matter at all" mantra that was common for a long time. Was written about one or two years ago iirc

http://www.eurogamer.net/articles/digitalfoundry-2016-is-it-finally-time-to-upgrade-your-core-i5-2500k

This? We need things like reddit bots to post poo poo that people always ask for! This is 50 percent of my posts for the last year!

Yaoi Gagarin
Feb 20, 2014

Jago posted:

http://www.eurogamer.net/articles/digitalfoundry-2016-is-it-finally-time-to-upgrade-your-core-i5-2500k

This? We need things like reddit bots to post poo poo that people always ask for! This is 50 percent of my posts for the last year!

That's probably the one I'm thinking of. I could have sworn somebody looked at 99th percentile frametimes across different memory speeds though. Maybe I'm conflating two different articles

Col.Kiwi
Dec 28, 2004
And the grave digger puts on the forceps...

VostokProgram posted:

Does anyone remember that article that showed that RAM frequency was important for frametime consistency on Intel CPUs from the last few years? I think it at least covered Sandy bridge through haswell, and basically demolished the old "ram frequency doesn't matter at all" mantra that was common for a long time. Was written about one or two years ago iirc
You might be thinking of this http://www.techspot.com/article/1171-ddr4-4000-mhz-performance/page3.html

Doesn't have 99% but has minimums

Sidesaddle Cavalry
Mar 15, 2013

Oh Boy Desert Map

Combat Pretzel posted:

What's with the cache? The 8MB on the 4C/8T and 8.25MB on the 6C/12T. And jesus, shifting the high lane count to the 10C one, that's mean.

1.375MB of L3 per core on those models above the 7740K but also 1MB(!) of L2 per core if i'm reading the rumors correctly

GRINDCORE MEGGIDO
Feb 28, 1985


I hope the 1MB l2 makes it to the non HEDT chips.

Anime Schoolgirl
Nov 28, 2002

it would be a very krizanich move to make a separate core for consumer products that have less l2 cache even though ever since sandy bridge the core structures outside the l3 cache have been identical

Combat Pretzel
Jun 23, 2004

No, seriously... what kurds?!

Sidesaddle Cavalry posted:

1.375MB of L3 per core on those models above the 7740K but also 1MB(!) of L2 per core if i'm reading the rumors correctly
If that's right, then I'm in.

eames
May 9, 2009

Place your Core i9 7920X price bets now! I'd say $2299.

suck my woke dick
Oct 10, 2012

:siren:I CANNOT EJACULATE WITHOUT SEEING NATIVE AMERICANS BRUTALISED!:siren:

Put this cum-loving slave on ignore immediately!

eames posted:

Place your Core i9 7920X price bets now! I'd say $2299.

There are many reasons to buy at this price/performance ratio instead of AMD or last-gen two socket Xeons, such as

Paul MaudDib
May 3, 2006

TEAM NVIDIA:
FORUM POLICE

blowfish posted:

There are many reasons to buy at this price/performance ratio instead of AMD or last-gen two socket Xeons, such as

the ability to clock past 4 GHz

PerrineClostermann
Dec 15, 2012

by FactsAreUseless

blowfish posted:

There are many reasons to buy at this price/performance ratio instead of AMD or last-gen two socket Xeons, such as

Muh frames

SlayVus
Jul 10, 2009
Grimey Drawer
I just don't see a use case for even prosumers for a 10/12 core $1500+ CPU. They could potentially buy dual xeons. I only see hardware enthusiasts buying something like this, I would rather purchase a Ryzen just cost wise. I'm not an AMD fan boy, but I would rather overspend on my video card where I might see an appreciable difference than my CPU which might perform less so.

GRINDCORE MEGGIDO
Feb 28, 1985


1MB L2 on cheaper chips might really help frametime consistency. That would own.

Sidesaddle Cavalry
Mar 15, 2013

Oh Boy Desert Map

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

GRINDCORE MEGGIDO posted:

1MB L2 on cheaper chips might really help frametime consistency. That would own.

It would, but a 4x jump in L2 cache on the same node is not going to come cheap.

It'll also be real interesting to see how the lack of TurboBoost on all the 4c/6c chips turns out--hope for some overclocking headroom, I guess, but that's probably gonna hurt those in search of muh frames.

dont be mean to me
May 2, 2007

I'm interplanetary, bitch
Let's go to Mars



2500K poo poo sample. 4.1. 1333 RAM (1CT at least).

I could switch to Ryzen (2nd gen, I'd like to think I'm not an idiot), and not miss the extra 'muh frames' I never had, and not have to deal with Intel's BS anymore.

Combat Pretzel
Jun 23, 2004

No, seriously... what kurds?!

DrDork posted:

It would, but a 4x jump in L2 cache on the same node is not going to come cheap.
It seems like it shifted. The eight core Skylake-X is missing around 9MB of L3 cache compared to an octocore Haswell-E/Broadwell-E. Whereas the L2 bump totals 6MB (8x768KB). Are L2 cache structures bigger than L3 ones?

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

Combat Pretzel posted:

Are L2 cache structures bigger than L3 ones?

One would assume so, since they're quite a bit faster, and faster memory almost always means less dense structures.

I wonder if the shift is because they're not really seeing a lot of use for massive shared L3 in 6+ core chips?

Combat Pretzel
Jun 23, 2004

No, seriously... what kurds?!
I'd figure the more cores the more coherency issues (and locking? Does that happen in a shared cache?), plus given that each core probably always does its own thing (apart from kernel calls), there's probably plenty of contention over who gets to keep what in the cache or not.

Palladium
May 8, 2012

Very Good
✔️✔️✔️✔️
Forget about reasonable Intel HEDT prices, as long as AMD only manages to fight Intel's product segmentation division than their engineering division.

Palladium fucked around with this message at 11:52 on May 14, 2017

Combat Pretzel
Jun 23, 2004

No, seriously... what kurds?!
I still expect it quite a bit above AMDs offering, but I'm still crossing my fingers the eight core ones drop to a more reasonable level. The lack of lanes is slightly annoying though. If they keep the existing price structure, the octocore would drop to the price range of the current high lane count six core HEDTs, which is around 600-700€ where I live, which would probably be acceptable.

EdEddnEddy
Apr 5, 2012



What in the hell, you only get the extra PCI-E Lanes from the 10C chip? I know SLI is loosing it's appeal but Jesus that seams scummy.

Then at the bottom end, WTF is with the 4C/4T i7? I guess it goes back that the i# really doesn't mean jack crap from when it was introduced to now. And finally we are getting the i9 that we all though would be what the HEDT Chips were going to be called when the first 6C chip made its appearance, and they went with the -E instead.

Man I continue to hope, and then loose pretty much any interested in building a new system. The cost vs performance increase just isn't there vs my old 3930K outside of the power savings and NVME. Outside of that I am not left wanting enough to consider $1000 for just the CPU alone. (Or maybe going AMD, but I am going to have to see what the refresh/HEDT chips offer first).

GRINDCORE MEGGIDO
Feb 28, 1985


Yay for dual channel memory on HEDT platform.
That's so lame.

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

GRINDCORE MEGGIDO posted:

Yay for dual channel memory on HEDT platform.
That's so lame.

Yeah, Skylake-X is really making me think that I made the right call by picking up a 5820k to replace a fried 2500k a year ago, rather than hoping Intel would put out anything reasonably attractive in the <$400 bracket in the near future.

Kazinsal
Dec 13, 2011



Intel continues to build hype and then disappoint everyone except the hardcore fanboys who will buy the 2% performance improvement every 18 months because it says Intel on it.

Ryzen 2 when?

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

Kazinsal posted:

Ryzen 2 when?

If they take a card from their GPU team, rebadges will be out in 6 months.

Kazinsal
Dec 13, 2011



DrDork posted:

If they take a card from their GPU team, rebadges will be out in 6 months.

The GPU rebadges usually at least have some BIOS fixes that make the cards a bit beefier in the long run so that would kinda be a good thing

PC LOAD LETTER
May 23, 2005
WTF?!

Combat Pretzel posted:

I'd figure the more cores the more coherency issues (and locking? Does that happen in a shared cache?), plus given that each core probably always does its own thing (apart from kernel calls), there's probably plenty of contention over who gets to keep what in the cache or not.
More cache would help with that issue though right?

I always thought the whole reason for big caches was to make up for inadequacies of main system RAM (lack of bandwidth and/or high latency). Maybe with higher speed DDR4 + XPoint DRAM + 6 channel memory (maybe improved hardware memory prefetcher too?) Intel thinks lots of cache isn't necessary for Skylake Xeons? edit: Keeps the die size + power usage down if they can shrink the caches a bunch I guess.

PC LOAD LETTER fucked around with this message at 00:59 on May 15, 2017

Adbot
ADBOT LOVES YOU

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

PC LOAD LETTER posted:

Maybe with higher speed DDR4 + XPoint DRAM + 6 channel memory (maybe improved hardware memory prefetcher too?) Intel thinks lots of cache isn't necessary for Skylake Xeons? edit: Keeps the die size + power usage down if they can shrink the caches a bunch I guess.

Overall cache hasn't really gone down by too much, though. A 5820k, for example, had 6x256k L2 + 15MB L3 = 16.5MB overall to back 6 cores. The new 7800X has 6x1MB L2 + 8.25MB L3 = 14.25MB to back 6 cores. Similar for the 5960k at 22MB total vs the 7820k at 19MB for 8 cores each. So it's more of a shift from L3 to L2 cache than a huge drop in cache. Considering that L2 undoubtedly takes up more die space than L3, the drop of 2.25MB/3MB might be the sacrifice needed to keep the total cache die sizes similar.

It also looks like The 6/8/10/12c's will be sticking with quad-channel 2666 memory, while the 4c chips are stuck with dual-channel. Considering X99 already defacto supports quad-channel upwards of 3200, it doesn't seem like much of an improvement.

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