Intel Releasing 10 core 20 thread i9-10900KF for $499 very soon... 5.2 Ghz boost

[ratirt]

Intel implemented TSX in Haswell (and it was broken), it is implemented in a fixed way in Skylake - and used in one of the KASLR-breaking vulnerabilities which again prompts for disabling it. AMD has (proposed) ASF extension as counterpart that have not actually been implemented in anything so far.

It is not about prefering Intel's extension but using what CPU manufacturer has implemented in actual products (= can be used). TSX exists and works in an actual product, ASF does not.

AS always you have missed the conversation we been having here before you showed up. It is well known that Intel been the one developers for software turn into and used features they have offered. That's why software has been developed with Intel's TSX not AMD's ASF due to the fact Intel's been a leader for some time and has some sort of advantage. You repeating what I wrote dude
And yes, developers preferred Intel's feature. Even though it didn't work and now it is disabled. Just like Metroid mentioned. ASF works but has not been used. Hopefully this will change soon and I hope developers will give it a shot.

 

[londiste]

There is no CPU with ASF.

Edit:
ISA Extensions are hardware features. Extensions can be used (software written to utilize them) when there is hardware that implements the extension.
I mean, I suppose theoretically you could write software to support ASF but that would be an academic exercise.

Edit2:
Not everyone thinks in terms of Intel vs AMD. Software is developed to use whatever can be used to squeeze more performance out of it. If an extension turns out to be useful and efficient enough, it (or its counterpart) will be implemented by other vendors as well, eventually. Generally these counterparts are close enough the the original that they can be used by the same software with only minor additions.

 

[ratirt]

There is no CPU with ASF.

Edit:
ISA Extensions are hardware features. Extensions can be used (software written to utilize them) when there is hardware that implements the extension.
I mean, I suppose theoretically you could write software to support ASF but that would be an academic exercise.

What's the point of implementing it when no one want's to use it when they've got TSX and have chosen to go with the last one?
I hope this will change due to vulnerabilities with TSX and maybe developers will turn into this. It works. I didn't say it's in the processors. Besides it's an alternative not actual TSX. Maybe from a developer side it's TSX or ASF?

 

[londiste]

What's the point of implementing it when no one want's to use it when they've got TSX and have chosen to go with the last one?

This is not one or the other. Two extensions can be used by the same software depending on which one is available.

I hope this will change due to vulnerabilities with TSX and maybe developers will turn into this. It works. I didn't say it's in the processors. Besides it's an alternative not actual TSX. Maybe from a developer side it's TSX or ASF?

Both TSX and ASF define CPU ISA instructions, commands you give to a CPU. If there is no CPU to respond to these commands why write software for it?

 

[Vya Domus]

TSX is incredibly sparsely used.

 

[ratirt]

This is not one or the other. Two extensions can be used by the same software depending on which one is available.Both TSX and ASF define CPU ISA instructions, commands you give to a CPU. If there is no CPU to respond to these commands why write software for it?

Is that so?
Do You remember 3Dnow!!! feature by AMD? It is not being useed on any of the new CPUs. Even bulldozer didn't have it.
Can you tell why that's the case? 3DNow!!! was really good. Even better than SSE back in the days. Yet developers chose SSE instead and the development of the SSE instructions expanded to SSE2 SSE3 and so on. Even though the 3DNow!! was sucked in by Intel to its SSE later on but that's beside the point. It was better.

 

[ppn]

...4 cores are 125mm^2, 6 cores are 150mm^2, 8 cores are 175mm^2. Add two cores and do nothing else is a 200mm^2 chip, comparable to Zen/Zen+. This is not extremely large. Get rid of iGPU and it's back down to 175mm^2 or less. They might take a hit on the margin but it would still be very-very far from losing money.

IGPU is around 50mm^2, so the 10 core Comet GFX-less is as big as the 6 Core + IGPU, 6 more cores gets 9mm wider
2-core group is 3mm wide, the die size can only grow in width and the heatpsreader is the limit. 16 core on 14++,++, looks very real.

 

[londiste]

Is that so?
Do You remember 3Dnow!!! feature by AMD? It is not being useed on any of the new CPUs. Even bulldozer didn't have it.
Can you tell why that's the case? 3DNow!!! was really good. Even better than SSE back in the days. Yet developers chose SSE instead and the development of the SSE instructions expanded to SSE2 SSE3 and so on. Even though the 3DNow!! was sucked in by Intel to its SSE later on but that's beside the point. It was better.

3DNow! did have nice traction with developers. Not all extensions succeed whether they get software written for them or not. If my memory serves right, K6-2 that brought in 3DNow! went against Pentium II that came earlier and was probably better at the time.

Didn't 3DNow! share resources with something else while SSE didn't?

Edit:
Registers, with MMX/x87. SSE brought in new registers for its use.

It was a time with faster changes in processors and SIMD extensions were the new thing - MMX in 1997, 3DNow! in 1998, SSE in 1999. 3DNow! was initially met with some success but was later simply overshadowed by SSE. Something to keep in mind is that new extensions are not going to get used immediately. It usually takes year or two unless they are something fundamental. By the time 3DNow! got wide enough usage in software, SSE was already out with benefits over it. 3DNow! and Enhanced 3DNow! were definitely a during Thunderbird era.

 

[dirtyferret]

I found the video to prove my point at 4ghz it clearly shows intel still lead in some games. So frequency is not the only reason amd is losing in games.

Hardware unboxed, they only have a youtube channel. Very very interesting review. One of the best reviews i have seen so far.

Hardware Unboxed is techspot, Steve in the video is the head review guy for techspot. He just makes a video while doing his web site review.

 

[white phantom]

I'm sold...Who else is getting a 10 core 20 thread i9-10900KF for $499 very soon!!?.....

@Knoxx29

Yeah when it arrives for 400 UK pounds like rumoured I'd think about it, atleast I know I'll still have 400 quid laying around because I highly doubt its that price

 

[StrayKAT]

That's not a great plan considering that the 3900x keeps up with the 7960x, at least in Linux. The difference in price between those two chips is so large it makes literally zero sense to get the 7960x and why get something lesser when you know the 3900x will outperform it?

I admit my ignorance on the pricing. I wouldn't do that either.

 

[HTC]

AMD succeeded in keeping the clock speeds and even increasing them a bit - if we count 1-2 cores at 4.5-4.6 at 1.5V(!) an increase - because 14/12nm did not clock high either. The sudden downturn in efficiency is still around 4.1-4.3GHz.

That only happened precisely because of the "chiplet approach". It's why Intel is having so many problems with clock speeds: since their CPUs are monolithic, they can't have high enough clocks to have them "be an upgrade" VS current offerings.

Much more information regarding chiplets here: http://www.eecg.toronto.edu/~enright/Kannan_MICRO48.pdf (page two, "Background and Motivation").

 

[TheGuruStud]

"very soon TM (c)" They might as well TM "Sold At A Loss" too

 

[TheoneandonlyMrK]

Intel Core i9-10900KF - 10C/20T @ 5.2GHz for $499 on 14nm+++

Oh, would you look at that: Intel's next-gen 10th-gen CPU roadmap leaks.

www.tweaktown.com

ryzen 3900x already ancient history if the 5.2 is all 10 cores no downclocking...

Hype train leaving the station already.

All ten cores on 14nm+++ @ 5.2Ghz, reallyyy.
The article doesn't say that ao why are you.

Plus imagine the Tdp and heat flux off your unicorn version.

A core , perhaps 2 at up to 5.2ghz Is my version of made up non-sense to add to this Pr Bs.

As for yet another 14nm based socket Wtaf intel , you reaLly really are taking the piss.
Note they added pins at least to totally f#ck over any attempt to prove they are lying arseholes when it comes to socket swaps.

 

[Aquinus]

That only happened precisely because of the "chiplet approach". It's why Intel is having so many problems with clock speeds: since their CPUs are monolithic, they can't have high enough clocks to have them "be an upgrade" VS current offerings.

Much more information regarding chiplets here: http://www.eecg.toronto.edu/~enright/Kannan_MICRO48.pdf (page two, "Background and Motivation").

I'm glad someone said this because this is exactly why AMD is killing it with the 3900x. Intel has multi-die CPUs in their Xeon lineup for the *really* big CPUs with The Good Glue™, but they really have no solution to this problem in the mainstream market. Intel really needs to be taking the same approach to CPU design because you can only make dies so big. I think AMD hit a sweet spot. We'll find out when we start seeing 8c/16t chiplets on EPYC chips because a competing AMD chip with this goodness in the server market at the right price would be a gut punch for Intel and it definitely has been earned.

 

[Space Lynx]

I'm glad someone said this because this is exactly why AMD is killing it with the 3900x. Intel has multi-die CPUs in their Xeon lineup for the *really* big CPUs with The Good Glue™, but they really have no solution to this problem in the mainstream market. Intel really needs to be taking the same approach to CPU design because you can only make dies so big. I think AMD hit a sweet spot. We'll find out when we start seeing 8c/16t chiplets on EPYC chips because a competing AMD chip with this goodness in the server market at the right price would be a gut punch for Intel and it definitely has been earned.

killing it in regards to non-gaming applications. new benches are showing not only is FPS slower in 3900x, but latency is actually worse than predicted as well because of the chiplet design. gamersnexus was talking about it some. all I do is game, so eh.

 

[Aquinus]

killing it in regards to non-gaming applications. new benches are showing not only is FPS slower in 3900x, but latency is actually worse than predicted as well because of the chiplet design. gamersnexus was talking about it some. all I do is game, so eh.

The server market is where the real money is at though. There is definitely a price to pay, but if you need 12c/24t, there is a good bet that such a tradeoff is acceptable. If you need this CPU, there is also a really good bet that gaming isn't the only thing you're using it for.

 

[Space Lynx]

The server market is where the real money is at though. There is definitely a price to pay, but if you need 12c/24t, there is a good bet that such a tradeoff is acceptable. If you need this CPU, there is also a really good bet that gaming isn't the only thing you're using it for.

yeah its great in that regard, I agree in full.

 

[FireFox]

I'm sold...Who else is getting a 10 core 20 thread i9-10900KF for $499 very soon!!?.....

@Knoxx29

I assume it will run hot but that is not a problem for me so i guess i am getting one just for the sake of having it

Edit: I am curious how much they will charge for it here in Germany/Europe

 

[JustNiz]

I guess "Very Soon" here has a different meaning to what most people understand.

 

[Space Lynx]

I guess "Very Soon" here has a different meaning to what most people understand.

That's what the article said, so I don't know, if you are in this industry you already know to take that source with a grain of salt. /shrug

 

[R-T-B]

Still monolithic

I love this part, frankly.

I'll take a small monolithic core for my needs any day... it's simply faster than an identical mcm, but I am a dying breed. I still don't know what to do with 8 cores. MCM is the way to go if you want more really.

TSX is incredibly sparsely used.

Not in serverland. It's like god's gift to database performance in terms of cpu overhead.

 

[Aquinus]

Not in serverland. It's like god's gift to database performance in terms of cpu overhead.

Do you have an example of a database that uses TSX? I mainly work with PostgreSQL and as far as I know, it doesn't take advantage of TSX at all. The only one I'm aware of is MS SQL Server 2016+ from what I've read online, but beyond that, I can't find anything else that uses it.

 

[Vya Domus]

Not in serverland. It's like god's gift to database performance in terms of cpu overhead.

I've looked at many database benchmarks comparing Epyc and Xeons and I never witnessed something that I could describe as god's gift on Intel's side. Of course I have to mention that I have no idea if any of those even use TSX, I doubt they do, but I would like if you'd give me an example.

Also : https://www.zdnet.com/article/intels-cascade-lake-cpus-impacted-by-new-zombieload-v2-attack/

Certainly not a stellar feature. Ironically the only piece of software that I know which uses TSX is the RPCS3 emulator and it has nothing to do with databases.

 

[Darmok N Jalad]

I shall pay for mine with hen’s teeth.