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
Volguus
Mar 3, 2009

priznat posted:

I do like that no one cares even in the Intel thread to mention the discontinuation of Itanium that was announced recently.

Rest in piss, Itanium. Owned by amd64. (A rare win)

Wasn't itanium only kept for the last decade because HP wanted them to? Itanium died the second AMD announced amd64 back in 2000, it was just on life support since then.

Adbot
ADBOT LOVES YOU

priznat
Jul 7, 2009

Let's get drunk and kiss each other all night.

Volguus posted:

Wasn't itanium only kept for the last decade because HP wanted them to? Itanium died the second AMD announced amd64 back in 2000, it was just on life support since then.

Pretty much, I was actually surprised it was still around.

BobHoward
Feb 13, 2012

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

priznat posted:

Pretty much, I was actually surprised it was still around.

HP got some of their captive PA-RISC customers to port to Itanium since that was the way forward for increased performance without porting to other operating systems. Some of those remained captive enough to keep buying Itanium equipment even as it became ever more clear that Itanium was never going to live up to its original promises.

Basically, porting costs can drive certain types of organizations to stick with weird machine architectures long after everyone else has moved on.

necrobobsledder
Mar 21, 2005
Lay down your soul to the gods rock 'n roll
Nap Ghost
That's just sunk cost and a lot of organizations really don't have the agility to re-write anything perhaps ever during the lifetime of the company. What kind of programmer is totally cool with working on a dead-end chip? Not one with good career prospects after it goes EOL. That's how you get government programmers that spend most of their days in meetings rather than in front of a text editor.

mewse
May 2, 2006

priznat posted:

Rest in piss, Itanium.

:boom:

in a well actually
Jan 26, 2011

dude, you gotta end it on the rhyme

phosdex
Dec 16, 2005

Would we all be using some iteration of Itanium if not for amd64?

fishmech
Jul 16, 2006

by VideoGames
Salad Prong

phosdex posted:

Would we all be using some iteration of Itanium if not for amd64?

No, Intel would just have invented their own 64 bit superset of x86 if AMD had never figured out amd64.

in a well actually
Jan 26, 2011

dude, you gotta end it on the rhyme

phosdex posted:

Would we all be using some iteration of Itanium if not for amd64?

On the consumer side, x86 with PAE.

Server? Depends if SPARC or somebody else could be competitive with Itanium despite Intel's fab advantage.

BobHoward
Feb 13, 2012

The only thing white people deserve is a bullet to their empty skull
Intel actually did invent their own 64-bit x86 that never saw public release, as a backup plan and/or the x86 faction of the company not being completely on board with Itanium supremacy.

Support for this was present in some P4 chips, but disabled in everything sold to the public by blowing fuse bits. When AMD64 gained enough momentum that Intel had to respond (and abandon its previous public position that 64-bit was only going to happen in Itanium), at first they tried to pressure Microsoft to support their version of 64-bit x86 over AMD64. Microsoft famously rebuffed them and basically said "if you wanted input on 64-bit x86 you should have spoken up when AMD published specs, we already ported windows and we're not doing it again, too late". This forced Intel to revise the hidden P4 64-bit mode to support AMD64 instead.

priznat
Jul 7, 2009

Let's get drunk and kiss each other all night.

This graph cracks me up whenever I see it.

I want to search linkedin for the head of marketing for intel enterprise in that era and see what they are doing these days.

AARP LARPer
Feb 19, 2005

THE DARK SIDE OF SCIENCE BREEDS A WEAPON OF WAR

Buglord

priznat posted:

This graph cracks me up whenever I see it.

I want to search linkedin for the head of marketing for intel enterprise in that era and see what they are doing these days.

eames
May 9, 2009

I want this but I'm pretty sure it'll cost a fortune. :smith:

Only registered members can see post attachments!

priznat
Jul 7, 2009

Let's get drunk and kiss each other all night.

eames posted:

I want this but I'm pretty sure it'll cost a fortune. :smith:



Ooh yeah.

All those PCIe lanes, need to go all-nvme raid :getin:

So many sockets now too.

Gucci Loafers
May 20, 2006

Ask yourself, do you really want to talk to pair of really nice gaudy shoes?


At the very least, I feel like that Intel Integrated WLAN is a nice addition.

Saukkis
May 16, 2003

Unless I'm on the inside curve pointing straight at oncoming traffic the high beams stay on and I laugh at your puny protest flashes.
I am Most Important Man. Most Important Man in the World.
In what kind of situations do those extra PCIe lanes provide noticeable benefit?

crazypenguin
Mar 9, 2005
nothing witty here, move along

Saukkis posted:

In what kind of situations do those extra PCIe lanes provide noticeable benefit?

Most people only ever plug one graphics card and nowadays an NVMe SSD in, so reasonable question.

But: more than 1 x16 graphics card (the bandwidth is actually useful for deep learning, if not gaming). More NVMe SSDs without being bottlenecked. Thunderbolt cards. FPGAs. Raid controllers. More high-speed USB ports (less of an issue now that we're getting those on the newest chipsets...)

The CPU-chipset bottleneck is DMI 3.0 (or x4 PCIe 3.0), or 32 Gbps of bandwidth. With single USB ports offering 10 Gbps, and an SSD that can eat the whole 32 Gbps, we're actually surprisingly close to this bottleneck being a problem even for normal consumers. (Not yet a big one, mind, but still... it's supposed to be a total non-issue.)

Here's hoping it won't be too long until PCIe 4.0 is actually adopted (and we get DMI 4.0). And I also kinda hope Intel will follow AMD's lead and add another 4 CPU lanes to their consumer CPUs (for the SSD). It's a nice design.

Lowen SoDium
Jun 5, 2003

Highen Fiber
Clapping Larry
DMI 3.0 didn't come as soon as PCIe 3.0 was available. No telling if 4.0 will be the same or not.

lllllllllllllllllll
Feb 28, 2010

Now the scene's lighting is perfect!
Right now Intel CPUs suck for their lack of performance gains and AMD CPUs suck for not catching up to Intel (and because of their mainboards).

Palladium
May 8, 2012

Very Good
✔️✔️✔️✔️

lllllllllllllllllll posted:

Right now Intel CPUs suck for their lack of performance gains and AMD CPUs suck for not catching up to Intel (and because of their mainboards).

I feel everything is just meh. More threads and PCIE lanes that I don't need or care about.

champagne posting
Apr 5, 2006

YOU ARE A BRAIN
IN A BUNKER

You can just not buy them, and at the same time you all have to live with the fact that the big-huge performance jumps of yesteryear are goners.

redeyes
Sep 14, 2002

by Fluffdaddy
I have the problem of never having enough PCI-E lanes. I wish the drat 16x slots didn't change to 8x when you actually use both of them. More threads is always better for me too.

eames
May 9, 2009

New leaked Skylake-X benchmarks :salt:

https://d1rktuf34l9h2g.cloudfront.net/2/25/2513876a_7900X.jpeg

https://d1rktuf34l9h2g.cloudfront.net/c/cb/cb0313a2_6950X.jpeg


copypaste:
i7-7900X 10C (@4.0 Ghz)= 1386.94Mpix/s
i7-6950X 10C (@4.50 Ghz)= 897.28Mpix/s
i7-6950X 10C (@3.50 Ghz)= 746.64Mpix/s

https://www.techpowerup.com/233667/new-details-on-intels-upcoming-10-core-skylake-x-i9-7900x-surface

eames fucked around with this message at 23:19 on May 25, 2017

repiv
Aug 13, 2009

SiSoft uses AVX512 where available, so that would fit with the rumours of Skylake-X bringing AVX512 to the HEDT chips. I don't see how else they'd get such a massive IPC bump over Broadwell.

Combat Pretzel
Jun 23, 2004

No, seriously... what kurds?!
Will only be of use for media creation, and not even on the short term, since developers need to adopt it. Games will take even longer, given the consumer CPUs will not feature it for a while.

Paul MaudDib
May 3, 2006

TEAM NVIDIA:
FORUM POLICE

eames posted:

New leaked Skylake-X benchmarks :salt:

https://d1rktuf34l9h2g.cloudfront.net/2/25/2513876a_7900X.jpeg

https://d1rktuf34l9h2g.cloudfront.net/c/cb/cb0313a2_6950X.jpeg

copypaste:
i7-7900X 10C (@4.0 Ghz)= 1386.94Mpix/s
i7-6950X 10C (@4.50 Ghz)= 897.28Mpix/s
i7-6950X 10C (@3.50 Ghz)= 746.64Mpix/s

https://www.techpowerup.com/233667/new-details-on-intels-upcoming-10-core-skylake-x-i9-7900x-surface

4.0 GHz all-core turbo with 4.5 GHz single-core turbo!? :rice:

Looks like Skylake-X will OC well then. I figure you should easily be able to get every core to at least that fast with stock voltage or a minor increase. In fact with those clocks I wouldn't be surprised to see Kaby Lake-like clockrates overall - 4.8 GHz with liberal application of extra voltage and up to 5 GHz on a golden sample.

(Are they doing Skylake-X on the revised Kaby Lake node, or first-gen Skylake? :salt: and all but if that's serious it looks to me like they might just be going to Kaby-level performance. After all, Kaby was just a tweak, and apart from Kaby's process improvements and improved iGPU there really wasn't a whole lot of tweaking.)

Not thrilled about 175W TDP but I guess it probably goes along with having the AVX512.

I'm not a candidate for the 10C but if the 6C are reasonable I'd be real tempted to upgrade. Could be pretty close to Ryzen 7 multi-thread performance (within 5% or so), plus vastly better single-thread performance in gaming. Kaby with more cores and quad-channel memory is fine by me.

Combat Pretzel posted:

Will only be of use for media creation, and not even on the short term, since developers need to adopt it. Games will take even longer, given the consumer CPUs will not feature it for a while.

Pretty sure x264 and x265 already use AVX2, AVX512 shouldn't be too much of a problem.

GRINDCORE MEGGIDO
Feb 28, 1985


And more L2. That could own.

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

GRINDCORE MEGGIDO posted:

And more L2 L1. That could own.

GRINDCORE MEGGIDO
Feb 28, 1985


Oh, my bad. I wonder how that'll affect things.

DrDork
Dec 29, 2003
commanding officer of the Army of Dorkness
Unsure, but it can't be bad!

Paul MaudDib
May 3, 2006

TEAM NVIDIA:
FORUM POLICE
Is there any significance to the fact that the IMC is running at main clock speed on the 6950X and at 1/2 clock speed on the 7900X?

Palladium
May 8, 2012

Very Good
✔️✔️✔️✔️

GRINDCORE MEGGIDO posted:

Oh, my bad. I wonder how that'll affect things.

Cache sizing is a balance between latency and hit rate. The closer to the core the latency becomes increasingly more important.

priznat
Jul 7, 2009

Let's get drunk and kiss each other all night.
I had a better version of this but lost it, but it was the same general idea comparing the latencies between the different data accesses and is interesting:

https://gist.github.com/hellerbarde/2843375#file-latency_humanized-markdown

(for anyone who hasn't seen it, I'm sure several here have seen it or similar ones)

well why not
Feb 10, 2009




Out of curiosity, which current CPU gives the most cores per dollar? I do a lot of music production and it's a really parallelised load loves multicores.

Dodoman
Feb 26, 2009



A moment of laxity
A lifetime of regret
Lipstick Apathy
Ryzen

eames
May 9, 2009

I'm kind of losing track of all the architectures and codenames.

How will the entry level "HEDT" Kaby Lake X CPUs compare to "Desktop" Coffee Lake?
Aren't those supposed to launch at roughly the core count (4-6C) at the same time?

Paul MaudDib
May 3, 2006

TEAM NVIDIA:
FORUM POLICE

eames posted:

I'm kind of losing track of all the architectures and codenames.

How will the entry level "HEDT" Kaby Lake X CPUs compare to "Desktop" Coffee Lake?
Aren't those supposed to launch at roughly the core count (4-6C) at the same time?

Coffee Lake will supposedly introduce 6C to the consumer lineup, but it's limited to the standard consumer featureset (16 lanes, dual-channel RAM, etc). Also, 6-core non-hyperthreaded for some reason. :wtf:

SKL-X is still the HEDT lineup. Quad-channel ram, between 6 and (reportedly) up to 12 cores, 44 lanes (only 28 on the 8-core and below), etc. My speculation is that it may actually end up being a "HEDT Kaby Lake" and end up clocking well, since Kaby Lake's main improvement was a slightly revised process and some iGPU features that won't be on the HEDT chips. Intel might plausibly run SKL-X on the new process.

Kaby Lake-X is a 4C Small Kaby Lake on the X299 platform and retains most of the disadvantages of the consumer lineup (new socket doesn't put features there that aren't on the silicon), i.e. 2-channel memory and 16 PCIe lanes.

Basically if everything that is reported is true - there is now a crossover in the middle. You can get your 4C/6C on either the consumer platform or the HEDT platform. The HEDT platform is going to have better features and allow you a future upgrade path, but be more expensive.

Paul MaudDib fucked around with this message at 18:02 on May 26, 2017

fishmech
Jul 16, 2006

by VideoGames
Salad Prong
What'd be really nice is 6 cores/12 threads or more in a laptop CPU. I've had 4C/8T in that since 2010, it'd be nice to finally move forward.

eames
May 9, 2009

Paul MaudDib posted:

Coffee Lake will supposedly introduce 6C to the consumer lineup, but it's limited to the standard consumer featureset (16 lanes, dual-channel RAM, etc). Also, 6-core non-hyperthreaded for some reason. :wtf:

SKL-X is still the HEDT lineup. Quad-channel ram, between 6 and (reportedly) up to 12 cores, 44 lanes (only 28 on the 8-core and below), etc. My speculation is that it may actually end up being a "HEDT Kaby Lake" and end up clocking well, since Kaby Lake's main improvement was a slightly revised process and some iGPU features that won't be on the HEDT chips.

Kaby Lake-X is a 4C Small Kaby Lake on the X299 platform and retains most of the disadvantages of the consumer lineup (new socket doesn't put features there that aren't on the silicon), i.e. 2-channel memory and 16 PCIe lanes.

Basically if everything that is reported is true - there is now a crossover in the middle. You can get your 4C/6C on either the consumer platform or the HEDT platform. The HEDT platform is going to have better features and allow you a future upgrade path, but be more expensive.

Ok thanks, that clears things up.
KBL singlecore performance paired with 6C/12T and a halfway decent upgrade path to 10 cores/quadchannel sounds nice.
I hope that X299 idle power consumption won't be off the charts and perhaps for some comparable, unlocked and only slightly more expensive Skylake-W CPUs for ECC but who am I kidding...

Adbot
ADBOT LOVES YOU

Paul MaudDib
May 3, 2006

TEAM NVIDIA:
FORUM POLICE

eames posted:

Ok thanks, that clears things up.
KBL singlecore performance paired with 6C/12T and a halfway decent upgrade path to 10 cores/quadchannel sounds nice.
I hope that X299 idle power consumption won't be off the charts and perhaps for some comparable, unlocked and only slightly more expensive Skylake-W CPUs for ECC but who am I kidding...

(one thing - I'm unsure if all of these together - i.e. 6C consumer + 4C HEDT + SKL-X - are collectively referred to as Coffee Lake or whether it's specifically the 6C consumer processor that's Coffee Lake. But regardless, all of these should be coming together at once and people are definitely using it as a collective term for this launch regardless.)

The thing with the upgrade path is... if you aren't in the market for a $1500 processor right now, you're probably not going to be in 2 years or whatever either. Unless AMD can price Ryzen ridiculously aggressively ($750-100 for a 16C threadripper) I just don't see Intel's exponential high-end pricing changing. And I also don't think Threadripper will be particularly good for gaming (it's going to behave essentially like a multi-socket processor).

AMD has shot their wad for first-gen consumer Ryzen and don't really have any more counter-moves until they can do a revision in a year or so. If Intel can get the multithreaded performance of 6C SKL-X real close to 8C Ryzen performance then the dominant single-thread performance will make a solid argument all by itself.

Idle power is one area Ryzen does beat Broadwell-E, I think it's reasonable to hope some improvements there given the jump to Skylake uarch and hopefully the revised Kabylake process too.

For gaming and regular power-user stuff, I think the 6C consumer chips and the updated 6C/8C processors are the ones to watch here. You'll need to overclock heavily but I'm hoping for good IPC gains and clock rates. I think the 10C will be price cut ($900-1400 depending on model?) but will remain way too expensive for the gain most people would get. The 12C is obviously going to go above the previous 6950X pricing, probably like $1700. The 4C HEDT chips are a stupid idea, almost everyone would definitely go 6C on consumer/HEDT or 4C on consumer, rather than pay up for an X299 mobo and put a dinky 4C chip in it.

I have mixed feelings about the 6C6T idea. I think the logic is a lot the same as it was for i5s a couple years ago. A lot of games don't scale well on hyperthreading and you get more real cores (but Intel segments the HT into a higher segment). I've actually talked about it in the past as "an i5 for the HEDT platform" but it's weird as a consumer i7. Gaming isn't usually memory-bottlenecked but I'd still prefer to see it on the HEDT platform where you get quad-channel.

And I also think there should be a corresponding 6C12T i7 on the consumer platform as well, just the i5 is weird and I don't understand it. It can't be yields, the number of chips you get with fully-functional cores except for the HT thread unit is pretty much zero as far as I know.

Interestingly, I guess that does kinda set the stage for future consumer-lineup performance increases down the line (Intel doesn't do price cuts). If the full chip is now a 6-core... maybe we see 4C8T in the i5 lineup in the generation after Coffee Lake, with the i7s becoming 6C6T and 6C12T. So now you'd have all the Pentiums/i3s with 2 cores, the i5s with 4 cores, and the i7s with 6 cores, with both a hyperthreaded unlocked and non-hyperthreaded unlocked SKU in (at least) the i5s and i7s?

Paul MaudDib fucked around with this message at 19:01 on May 26, 2017

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