Intel is dropping hyperthreading from i7 chips!

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

TheGiant

Senior member
Jun 12, 2017
748
353
106
$450 seems ok to me? 8c/16t at 4.7GHz is very powerful for a mainstream cpu.

Well 5GHz 8C/16T CPU with CFL IPC and high perf DDR4 is a total monster for desktop

For me too like 400 EUR is perfectly reasonable price for that CPU. It is funny how relative prices are.Last year 6900K for 1000 vs 1800X for 500 it was ok but now 450 for 40-50% faster CPU is not ...

we need moar coarz :)
 

VirtualLarry

No Lifer
Aug 25, 2001
56,229
9,990
126
Is the only reason that they cut HT from the i7, so that they could add it in again (well, not really, they're all the same dies) in the i9 CPU?
 
  • Like
Reactions: french toast

TheELF

Diamond Member
Dec 22, 2012
3,967
720
126
A question.

It appears to me that the most popular interpretation of SMT, or HT in this case, is that there is a main thread and the 2nd one gets the unused resources. In other words the additional thread on the virtual core has less resources to operate.

I have always believed that there is no main thread but [2] threads vying equally for core resources. If HT gives, say a 30% increase, then, by my understanding, assuming equal loads, each thread will run at 65% full single thread performance. Am I correct?

If I'm correct, I think that I might prefer an 8C/8T model.
If one of the threads only uses less then half the resources of the core then there are another half of the resources left over for the other thread.
You could and you do have a lot of occasions where two threads run both at full speed on one core +HTT.
Proof of concept video.
https://www.youtube.com/watch?v=rJmIvYDPJiA

Even if your APP only uses x amount of cores and HTT wouldn't give you more performance for this one APP it would still allow you to run background crap or even some game while you do your work, all without influencing the speed of your main work.
 

TheELF

Diamond Member
Dec 22, 2012
3,967
720
126
Also there's no gain to be had if the application supports equal or less number of threads compared to the total amount of physical cores. No point of having 16 logical cores if the application support is 8 or less.
Sure there is,you can run two instances of that same app at once and still get at least a ~30-35% gain overall.

30% is usually the best case scenario for Intel/AMD's version of SMT because they want to add minimal transistor and die space.
There is no usual there is only apps that use a lot of IPC (so there is only a little bit left for the second thread) and apps that use much less IPC (so there is a lot more left for the second thread) .
 

french toast

Senior member
Feb 22, 2017
988
825
136
Is the only reason that they cut HT from the i7, so that they could add it in again (well, not really, they're all the same dies) in the i9 CPU?
Exactly! one of the reasons I believe is to enable higher clocks, or specifically higher all core turbo's.
But yea if they can get away with it then that gives them an opportunity to add it back into future SKUs..as a brand new feature!
This will benefit icelake I believe, as clocks on 10nm+ are going to be slightly lower than 14nm++..so they need extra selling points.
 

beginner99

Diamond Member
Jun 2, 2009
5,208
1,580
136
Is the only reason that they cut HT from the i7, so that they could add it in again (well, not really, they're all the same dies) in the i9 CPU?

Obviously. Else if i7 is 8/16, i5 would have to be 8/8 and i3 6/6 which obviously intel would never do. So they had to make a tier on top instead of pushing current skus down.
 

PotatoWithEarsOnSide

Senior member
Feb 23, 2017
664
701
106
i3 9300 4c4t
i5 9500 6c6t
i7 9700 8c8t
i9 9400 4c8t
i9 9600 6c12t
i9 9800 8c16t

If their naming structure was as simple as that, it'd be Intel's first naming structure that wasn't at least 95% retarded. Odd numbers, with i3/5/7, signifying core progression and lack of HT, and even numbers with i9 signifying core progression and HT enabled.

----

Intel's biggest problem is that they're clearly lacking innovation, which ultimately will cost them dearly if Ryzen 2 (7nm) proves another leap forward from AMD.

Edit: as an afterthought, they could have gone even moar moar moar with an i4, i6, and i8 as their HT enabled CPUs, but that's be too easy to understand.
 
Last edited:

NTMBK

Lifer
Nov 14, 2011
10,208
4,940
136
i3 9300 4c4t
i5 9500 6c6t
i7 9700 8c8t
i9 9400 4c8t
i9 9600 6c12t
i9 9800 8c16t

If their naming structure was as simple as that, it'd be Intel's first naming structure that wasn't at least 95% retarded. Odd numbers, with i3/5/7, signifying core progression and lack of HT, and even numbers with i9 signifying core progression and HT enabled.

----

Intel's biggest problem is that they're clearly lacking innovation, which ultimately will cost them dearly if Ryzen 2 (7nm) proves another leap forward from AMD.

Edit: as an afterthought, they could have gone even moar moar moar with an i4, i6, and i8 as their HT enabled CPUs, but that's be too easy to understand.

Well that makes no damn sense at all. An i9 should never be slower than an i7 (within the same power budget... not comparing e.g. desktop and laptop). You need to keep it making sense for average customers- i3 is worse than i5, which is worse than i7, which is worse than i9.
 

SPBHM

Diamond Member
Sep 12, 2012
5,056
409
126
Is that a bad move?
With core counts going to infinity the extra die space used for HT could be used for something else or simply save the space?

they are not saving anything other than power usage, it's just disabled, it's in there and I don't see that changing...


well, this must be good news for AMD, it means potentially the 2700x can still outperform the $350 i7 in some specific cases.
 
  • Like
Reactions: CHADBOGA

PotatoWithEarsOnSide

Senior member
Feb 23, 2017
664
701
106
Well that makes no damn sense at all. An i9 should never be slower than an i7 (within the same power budget... not comparing e.g. desktop and laptop). You need to keep it making sense for average customers- i3 is worse than i5, which is worse than i7, which is worse than i9.
If you base anything on absolutes then you'll never have an easy to understand coherent naming structure. Where do you put the 8/8 versus the 6/12 for instance? Which is better?
If I reordered my list, including the HT enabled CPUs, sorted by core count, its obvious that i9 simply means HT enabled, with everything else falling into an easily comprehensible order. The sole reason to even give an i3/5/7 moniker would be to differentiate core count, since you could live without it if your goal was to confuse consumers.
All that's happening is Intel would be organising their products to better reflect current product lines. It's surely less confusing than i3 by 2c4t, i5 being 4c4t, and i7 being a mix of 4c8t, 6c12t, and 8c8t, with i9 being the I've got a boner because I used a higher number to signify I'm the dog's bollocks, but with indication as to what exactly that means.
 

NTMBK

Lifer
Nov 14, 2011
10,208
4,940
136
If you base anything on absolutes then you'll never have an easy to understand coherent naming structure. Where do you put the 8/8 versus the 6/12 for instance? Which is better?

8 cores is better than 6 cores. Not hard.

If I reordered my list, including the HT enabled CPUs, sorted by core count, its obvious that i9 simply means HT enabled, with everything else falling into an easily comprehensible order. The sole reason to even give an i3/5/7 moniker would be to differentiate core count, since you could live without it if your goal was to confuse consumers.
All that's happening is Intel would be organising their products to better reflect current product lines. It's surely less confusing than i3 by 2c4t, i5 being 4c4t, and i7 being a mix of 4c8t, 6c12t, and 8c8t, with i9 being the I've got a boner because I used a higher number to signify I'm the dog's bollocks, but with indication as to what exactly that means.

i3 isn't 2C4T, and i5 isn't 4C4T. You're stuck in Sandy Bridge thinking. :) Coffee Lake (and AMD) have shook everything up. i3s have 4C8T, i5s 6C6T, i7s 8C8T, i9s 8C16T. And yes, i9 is only marginally faster- it's the Intel ePeen Edition. But that's been true of the top SKU for years and years, you pay a big premium for the right to say you have The Best.
 
  • Like
Reactions: ozzy702

PotatoWithEarsOnSide

Senior member
Feb 23, 2017
664
701
106
8 cores is better than 6?
You're sure, despite many comments in this threads suggesting it ain't even remotely clearcut.
The problem I have is that it's a bit fluffy to be placing a HT disabled CPU in a higher classification than a HT enabled CPU. The simplest differentiation is on whether HT itself is enabled. Since Intel want to go with an i9 moniker, for no reason other than penis envy, then let the HT be just that. The i9 9400 being just that little bit better than it's i3 9300 counterpart. You get to keep your "more cores is better" structure, and you can boast about having an i9 if you're a total bellend if you so wish. Meanwhile, the guy with the i7 9700 laughs on at the teenagers bragging about shit they don't know how to use properly.
 

epsilon84

Golden Member
Aug 29, 2010
1,142
927
136
8 cores is better than 6 cores. Not hard.



i3 isn't 2C4T, and i5 isn't 4C4T. You're stuck in Sandy Bridge thinking. :) Coffee Lake (and AMD) have shook everything up. i3s have 4C8T, i5s 6C6T, i7s 8C8T, i9s 8C16T. And yes, i9 is only marginally faster- it's the Intel ePeen Edition. But that's been true of the top SKU for years and years, you pay a big premium for the right to say you have The Best.

i3 are 4C/8T? Pretty sure they are 4C/4T only. Intel won't do 4/8 for '9th gen' chips either because it would be too close to a 6/6 i5.

So basically the new stack would be
i3 4/4
i5 6/6
i7 8/8
i9 8/16

As you said, you have to pay for the best. It's always been this way, especially with Intel, and especially with a flagship SKU aka i9

I still think $450 would be reasonable for a 8/16 9900K, especially if clocked at 4.7GHz. That's 17.5% higher than the 2700X ACT, plus CFL has higher IPC so you're looking at a CPU that could be potentially 25 - 30% faster than a $320 2700X.
 

NTMBK

Lifer
Nov 14, 2011
10,208
4,940
136
8 cores is better than 6?
You're sure, despite many comments in this threads suggesting it ain't even remotely clearcut.

Yeah, I'm sure. There's more resources available.

The problem I have is that it's a bit fluffy to be placing a HT disabled CPU in a higher classification than a HT enabled CPU. The simplest differentiation is on whether HT itself is enabled. Since Intel want to go with an i9 moniker, for no reason other than penis envy, then let the HT be just that. The i9 9400 being just that little bit better than it's i3 9300 counterpart. You get to keep your "more cores is better" structure, and you can boast about having an i9 if you're a total bellend if you so wish. Meanwhile, the guy with the i7 9700 laughs on at the teenagers bragging about shit they don't know how to use properly.

Intel has been putting a HT disabled CPU in a higher classification than a HT enabled one since Sandy Bridge. i5 4C4T vs i3 2C4T.

Most consumers don't give a damn about whether HT is on or not, or how much cache is available. They barely even care about clock speeds. They just want to know "is this one faster than the other one"? An having an i9 slower than an i5 makes no sense in that context.
 
  • Like
Reactions: french toast
Aug 11, 2008
10,451
642
126
Nice, the author even managed to cast intel in a bad light for increasing cores "limited benefit for the mainstream user". If the pricing is right, I dont see a big problem. Instead of 4 or 6 cores with hyperthreading, just move up to six or eight cores without hyperthreading. I would pick the extra cores over hyperthreading anyway. And for those that want the max threads there is still 8 cores with hyperthreading.
 

ryzenmaster

Member
Mar 19, 2017
40
89
61
On the plus side since processor vulnerabilities are now a thing, an i7 without HT is actually slightly less broken than one with. I'm talking about TLBleed in specific; a side-channel attack which has been demonstrated to work against HT enabled Intel chips. TBH it may not be the kind of attack most should be worried about anyway, but it exists nonetheless.
 

rbk123

Senior member
Aug 22, 2006
743
345
136
Is the only reason that they cut HT from the i7, so that they could add it in again (well, not really, they're all the same dies) in the i9 CPU?
Maybe they aren't getting the yields on this new chip that they need and so are splitting the bad eggs. Maybe they can't get the clock up to 4.7+ on many of the chips with HT, so the good ones get HT and bad ones don't.

I doubt this is a strategic direction decision, but more a pragmatic result of having to make lemonade with lemons. Time will tell.
 

Cerb

Elite Member
Aug 26, 2000
17,484
33
86
A question.

It appears to me that the most popular interpretation of SMT, or HT in this case, is that there is a main thread and the 2nd one gets the unused resources. In other words the additional thread on the virtual core has less resources to operate.
No. The back end of the CPU doesn't even see that there are different threads. Both threads have equal priority. The result is often like that statement, but that's not what the CPU is doing.

I have always believed that there is no main thread but [2] threads vying equally for core resources. If HT gives, say a 30% increase, then, by my understanding, assuming equal loads, each thread will run at 65% full single thread performance. Am I correct?
It can be like that, but usually it's not. Usually, the performance hit is fairly small, and both the OS and many multithreaded programs don't try to overload the CPU. So, you may see that with video encoding, but are unlikely to see that with gaming. The +30% is a typical expectation. Usually, it's a bit less, sometimes it's more. Usually, the benefit of it is that processes that can come in and take up a lot of CPU time, but don't use much of the CPU's execution resources, can run alongside processes that do use a decent amount of the CPU's resources, with very little performance impact. Try to search for reviews comparing CPUs with background applications running, to find examples.

For one example, memory management, such as allocating and freeing, defragmenting (AKA compacting), and garbage collection, use a lot of total CPU time, but have absurdly low IPC, just by their nature, and tend to not be too bad in terms of cache usage. With SMT, that code can do its thing, and hardly affect anything else running. Another would be antivirus, which often wakes up several threads, or in MS' case, spawns a bunch of separate processes, each making a lot of calls to the OS, that end up using up a bunch of CPU time, to end up doing very little. In many of these cases, you don't actually have multiple programs you are, or even can, test performance against, to say that each thread is X% as fast (in most cases, the larger concern is whether they remain "fast enough" that you don't notice a slowdown under moderate loads). But, it is possible to benchmark multithreaded programs in realistic environments, or repeatable amalgams of them, and show more consistent results with SMT than without.

Now, as far as getting rid of HT, if they can add more actual cores, instead, economically (the power use alone may be an issue that favors spread out non-SMT cores, once 8C and up becomes mainstream), that'd be perfectly fine, even if they left SMT only in workstation and server chips. However much I may roll my eyes at Intel's market segmentation efforts (niche sharing among SKUs will not be tolerated!), SMT vs. non-SMT is a perfectly good way to split product lines, and more actual cores is always better for performance.
 
  • Like
Reactions: Space Tyrant

MrTeal

Diamond Member
Dec 7, 2003
3,554
1,658
136
I really don't see this as a huge deal or a terrible move. All the same chips are there; you get the full die with HT in the i9-9900k, the i7-9700k with all cores enabled but HT off, and lower core counts filling in the i5 and i3 lineup.

It would be different if they disabled HT entirely on the mainstream platform for the 8 core chips to push people to HEDT, but that's not what's happening here. They're just rejigging the iNumber system.It's functionally no different than the 2500k being the "gamer" chip and the 2600k being the halo chip for people who needed the threads or just wanted the best.
 

Roger Wilco

Diamond Member
Mar 20, 2017
3,857
5,688
136
I would take two physical cores over four virtual cores anytime. The 8c/8t chip should provide a nice performance boost over the 8700K and stay in the same price bracket.
 

samboy

Senior member
Aug 17, 2002
217
77
101
My biggest problem with the I7-9700k is that they have gimped the cache as well. Only 12MB for 8 cores instead of the usual 2MB/core (which would be 16MB). Scaling the cores (6-8) and leaving the cache the same doesn't make a whole lot of sense to me.

I guess Intel is just segmenting the market and getting people to step up to the I9 model with the full 16MB and SMT and spend an extra $100.
However, this price is starting to get into the 16 core territory from AMD?
I would have thought that Intel would have wanted to do more to keep AMD at bay until they sort out 10nm!
 
  • Like
Reactions: Cerb

PingSpike

Lifer
Feb 25, 2004
21,729
559
126
My biggest problem with the I7-9700k is that they have gimped the cache as well. Only 12MB for 8 cores instead of the usual 2MB/core (which would be 16MB). Scaling the cores (6-8) and leaving the cache the same doesn't make a whole lot of sense to me.

Yes, I noticed this as well. It seems like a bigger deal to make a stink over.

Of course, this isn't really anything new. Intel typically has paired hyperthreading with increased cache together as long as I've been paying attention. It makes it hard to distinguish whether the increased performance from the step up is coming from hyperthreading or not when you read benchmarks.