Intel Core i9-9900K soldered heatspreader 'confirmed'

[HEXUS]

Leaked presentation slides highlight new Solder Thermal Interface Material (STIM).

Read more.

[Biscuit]

"NEW" Soldered Thermal Interface Material...

NEW?....

NEW?

[watercooled]

I was just about to post exactly the same thing!

[w1ntergr33n]

Lol. Have to love how they are trying to sell this as 'new'.

Marketing 101: remove feature from your product, then re-add later as 'brand new' (and priced accordingly of course...)

They must think we are idiots.

[Ozaron]

They must think we are idiots.

Well, according to the sales figures..

[lumireleon]

AFTER 9900K next? 10,000k. (NOTE: Hyper Threading is VERY expensive technology it requires duplicating certain areas of the processor by upto 5% extra silicon resulting in 30% extra cost)

[philehidiot]

AFTER 9900K next? 10,000k. (NOTE: Hyper Threading is VERY expensive technology it requires duplicating certain areas of the processor by upto 5% extra silicon resulting in 30% extra cost)

Didn't they include hyperthreading in some chips and just not enable it? Or were those chips where there were flaws with the hyperthreading areas and so they were disabled and sold and non hyperthreaded?

[peterb]

Didn't they include hyperthreading in some chips and just not enable it? Or were those chips where there were flaws with the hyperthreading areas and so they were disabled and sold and non hyperthreaded?

Most likely the latter case.

[Corky34]

AFTER 9900K next? 10,000k. (NOTE: Hyper Threading is VERY expensive technology it requires duplicating certain areas of the processor by upto 5% extra silicon resulting in 30% extra cost)

Cheaper than duplicating an entire core and all the associated gubbins that go along with it though.

[watercooled]

I'm assuming sarcasm from lumireleon.

WRT the binning process, as far as I know it's not really possible/likely for SMT to be broken in isolation. Either way, even if it were a tiny percentage of the die on which a defect would cause a problem, it's highly improbable you'd have enough to justify setting aside a specific bin for that.

In short, disabling SMT is a choice for product segmentation and/or potentially power limitations. SMT isn't a block feature on a die - it's an integral design feature of some sections, and a small number of blocks are duplicated in the core. It's not like disabling cores or cache for binning where a big chunk can be disabled for binning.

I remember Semiaccurate reporting on this, I'll edit the post if I find the article.

Edit: Found it: https://semiaccurate.com/2012/12/13/...n-crystalwell/

The other two are just technically indefensible, they are solely there to milk the buyer. Why? HT is an integral part of the core pipeline, it is not a unit that can be turned off or on without preventing the CPU from working. If there is an error in the unit, the whole core is scrap. It has to be turned off artificially, basically by being told to not work programatically, the circuitry must continue to function for the core to at all. Intel uses this fake methodology to make a buyer think they are getting more when they buy a ‘premium’ product, but it is just the same exact silicon with one less blown fuse.

[philehidiot]

I remember Semiaccurate reporting on this, I'll edit the post if I find the article.

Edit: Found it: https://semiaccurate.com/2012/12/13/...n-crystalwell/

Cheers, looks like good reading.

[Seeven]

I am stil rocking an 4790k paired with 32 gigs of ram and a 1070.
every game max out, 144 fps smooth, exept the more demanding ones, that kills my gpu XD

[Corky34]

I remember Semiaccurate reporting on this, I'll edit the post if I find the article.

Edit: Found it: https://semiaccurate.com/2012/12/13/...n-crystalwell/

The other two are just technically indefensible, they are solely there to milk the buyer. Why? HT is an integral part of the core pipeline, it is not a unit that can be turned off or on without preventing the CPU from working. If there is an error in the unit, the whole core is scrap. It has to be turned off artificially, basically by being told to not work programatically, the circuitry must continue to function for the core to at all. Intel uses this fake methodology to make a buyer think they are getting more when they buy a ‘premium’ product, but it is just the same exact silicon with one less blown fuse.

While sort of true, in that there's not much hardware that goes into making HT work and a chip without it is mostly the same as one with it, other than what's probably little more than a switch in the firmware, i don't think it's done to fool buyers into thinking a CPU with SMT is a ‘premium’ product, it's done because some workloads actually perform worse with SMT enabled.

Personally i wouldn't enable SMT on a CPU as i don't use heavily threaded workloads and even if i did I'd much prefer X amount of real cores over logical ones, SMT is sort of a halfway house and comes with disadvantages, i get why it's a thing (working on another thread while the main one's idle increases efficiency) but being idle isn't necessarily a bad thing.

/Saturday morning ramble.

[watercooled]

I think Charlie's point is that you're made to pay a premium for something which is already on the silicon, and only artificially disabled as a method of product segmentation.

I know what you're saying though, SMT doesn't always help in some applications and I'm not too sure of the state of things now, but it would often degrade gaming performance. Didn't stop people paying more for the 'premium' i7 though when an i5 would be almost identical otherwise - marketing FTW? It was even more worthy of a facepalm when people would get an unlocked i7, disable SMT, and use it predominantly for games. An unlocked i5 with a few notches on the clocks, nothing major, would yield effectively the same CPU...

WRT allowing cores to idle - it depends on the workload, contemporary CPU cores still use a fair amount of power when idle but still powered, getting the work off of some physical cores allows you to power/clock gate (depending on how the die is divided up into power/clock islands of course, and it tends to be more aggressive on mobile CPUs). I remember reading a justification of SMT in some low power-targeted scenarios as you could feed multiple threads into one core without waking other physical cores - in some workloads it's an effective and more power efficient way of doing things, but needs some awareness in the scheduler. Cores also take a while to enter/exit those deeper sleep states.