Intel is dropping hyperthreading from i7 chips!

Page 3 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
May 11, 2008
19,300
1,129
126
That a core has 100% load does not mean there is never a momentary stall in the pipeline because of memory requests or instructions needing result from previous instructions. The execution units are idle at those moments and that is when HT or SMT is convenient.
 

jpiniero

Lifer
Oct 1, 2010
14,509
5,159
136
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.

I think Intel is scaling back mainstream HEDT, perhaps because Threadripper has killed the profitability, especially given how tiny of a market it is.
 

maddie

Diamond Member
Jul 18, 2010
4,722
4,627
136
That a core has 100% load does not mean there is never a momentary stall in the pipeline because of memory requests or instructions needing result from previous instructions. The execution units are idle at those moments and that is when HT or SMT is convenient.
This statement appears to have internal contradictions. By definition, 100% load means no units are idle.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
This statement appears to have internal contradictions. By definition, 100% load means no units are idle.

It depends on what it means by 100% load. If its true 100% load, then maddie is right. But in the Windows definition the CPU could be mostly idling while task manager shows 100% utilization. Then you upgrade say your drive, then BAM your system performs better.

You can be 100% loaded like with LinX.
 

PeterScott

Platinum Member
Jul 7, 2017
2,605
1,540
136
The bolded part is going to be wrong in the majority of applications that take advantage of many threads. Hyperthreading gives 20-30% gain in such applications, something going from 6 to 8 core will achieve, as increasing physical cores don't scale 100% either. Hyperthreading is also Intel's version of SMT, and it also plays a part in hiding memory gap, which doesn't happen with extra cores.

This then becomes a "gaming" focused chip. Hyperthreading also increases power consumed by 25-30%. That allows them to keep the same TDP rating as the 6 core with HT part without changing uarch or process while likely being a benefit to games, as games that support less than the maximum logical thread count suffers from having Hyperthreading enabled.

No, 8/6= 1.33.

This makes the 8C/8T part 33% faster than 6c/6t on parallel applications.

6c/12t will only 33% or better faster on one or two niche applications that happen to suit HT very well.

The other 99.999% of Software will be faster on a 8C/8T part.
 

LightningZ71

Golden Member
Mar 10, 2017
1,627
1,898
136
You need the larger Caches with HT to better enable HT to shine. Its a significant increase to load on the memory bus, and having a larger L2 helps mitigate that. It seems logical that the 8/8 i7 would have a smaller cache than the 8/16 i9

I also think that the increasing evidence that intel's HT implementation (and to some extent, AMDs SMT implementation) has side channel and timing vulnerabilities, that having high core count chips without SMT will eventually be the way forward, especially in situations with high exposure. I'd rather have the ability to choose for myself in the BIOS/UEFI instead of having it restricted by the vendor, but, with Intel's product segmentation, they don't really have that choice.
 
  • Like
Reactions: Cerb

Ratman6161

Senior member
Mar 21, 2008
616
75
91
I wouldn't call it crazy, just artificial market segmentation. I'm not a fan of it, but the net result on performance won't be that great either way compared to 6/12. Some apps will prefer the virtual cores, others will prefer the extra 2 physical cores.

Saying this is a 'gamers only' chip is ridiculous, as it should perform pretty similarly to a 8700K or 2700X in most situations.

Yes, on the artificial market segmentation. If you have an 8C16T CPU, how do you get people to pay more for it? 1. Sell a crippled model with 8C8T so that you can 2. Charge more for one that isn't crippled.

To those who say "8 its better than 6", well that is true. But 8/8 Vs. 6/12 is more complex and really depends on what you do with it. On the other hand most consumers won't understand the difference between cores and threads anyway and those who do understand it enough to know if they need it or not will also know if they benefit from spending the $450. What I don't seen mentioned is that I would bet the more expensive CPU will also force you to buy a more expensive platform to run it on too.

We have always been at war with East Asia.
 

PeterScott

Platinum Member
Jul 7, 2017
2,605
1,540
136


I find Arstechnica has really gone a lot more click-bait in recent years.

As far as this being a Crazy move, would it have been better if all the 8C parts were i9, and the rest of the lineup stayed as it is today?

So the i7-9700K would just be a rebranded i7-8700K. Hurrah, i7 still has HT. 2 less real cores, but has HT.

I really don't get why people get so wrapped up in naming i7, or i9 is nothing to get upset about. Call it a Rose, call it Shirley, what difference does the name make, there is going to be an 8c/16t model for those that want it.
 
  • Like
Reactions: epsilon84
May 11, 2008
19,300
1,129
126
This statement appears to have internal contradictions. By definition, 100% load means no units are idle.

You know what i mean. I wrote it perfectly fine.
If a cpu was running at 100% according to the task manager there are still moments that there are execution units idle inside a core. And that is when HT or SMT is useful. If that is not the case, you can fully explain it in detail. I am all eyes.
 

XavierMace

Diamond Member
Apr 20, 2013
4,307
450
126
I really don't get why people get so wrapped up in naming i7, or i9 is nothing to get upset about. Call it a Rose, call it Shirley, what difference does the name make, there is going to be an 8c/16t model for those that want it.

It's the same reason people got mad at BMW when they re-did they're naming scheme. It has been the same for decades so people knew what new model they were searching for without any research.
 

PeterScott

Platinum Member
Jul 7, 2017
2,605
1,540
136
It's the same reason people got mad at BMW when they re-did they're naming scheme. It has been the same for decades so people knew what new model they were searching for without any research.

The vast majority (99%+) of people can't decode Intel CPU naming without using Google to take them to the Intel ARK, and for that matter the same applies to AMD CPUs. The Naming schemes are really meaningless gibberish.

How many people do you think could have made any kind of reasonable assessment of an i7-7660U without the ARK decoder? Where in the name does i7-8700K indicate that the CPU gained 2 more cores over the previous generation? Nowhere because it's meaningless gibberish, that you either have to read about in a review, or Google.

So I really don't think your argument passes the sniff test.
 

elpokor

Junior Member
May 22, 2017
23
9
51
You need the larger Caches with HT to better enable HT to shine. Its a significant increase to load on the memory bus, and having a larger L2 helps mitigate that. It seems logical that the 8/8 i7 would have a smaller cache than the 8/16 i9

https://ark.intel.com/es/products/135935/Intel-Core-i5-8259U-Processor-6M-Cache-up-to-3_80-GHz

yeah... I thought that before too. Then they released 4c/8t 8th gen mobile i5s with 6MB cache. It's all artificial market segmentation, just like with the PCIe lanes on X299.
 
Oct 19, 2006
194
1
81
Looking at Anandtech's review of the i7-8700K/ i5-8400, the 4/8 core 7700K is almost always faster than the 6/6 i5-8400. Granted this is probably because of the 7700K's extra cache and clock speed, not hyperthreading. Although if Intel doesn't keep the clock speeds of this new i7 decently higher than the old 8700K, I think we could see some benchmark upsets.
 

XavierMace

Diamond Member
Apr 20, 2013
4,307
450
126
The vast majority (99%+) of people can't decode Intel CPU naming without using Google to take them to the Intel ARK, and for that matter the same applies to AMD CPUs. The Naming schemes are really meaningless gibberish.

How many people do you think could have made any kind of reasonable assessment of an i7-7660U without the ARK decoder? Where in the name does i7-8700K indicate that the CPU gained 2 more cores over the previous generation? Nowhere because it's meaningless gibberish, that you either have to read about in a review, or Google.

So I really don't think your argument passes the sniff test.

There's one massive logical fallacy you're making and that's that the vast majority makes reasonable assessments. The vast majority of people don't know/care that it gained two more cores. The vast majority of people don't even know how many threads they currently have or their clock speed. You're generally doing pretty good if they can tell you "I think I have an Intel core something. Maybe a 5? ". Not once in 15 years of building PC's have I had a normal user come up to me and tell me they think they need more cores. They just want to know if they're buying the newer version of what they have and if it's faster. They know an i5 is better than an i3 and an i7 is better than both because the number is bigger so it's obviously faster.

It's was the same with the BMW's. 1 series was your small entry level, 3 series was your midsize entry, 5 series was your upscale mid size, and 7 series was your luxury boats. A 330 was better than a 328 and and 335 was better than both. The vast majority of their buyers were unaware of the fact that the 135, 335, and 535 all used the same engine and the 135i was actually the quickest of the 3. It doesn't matter if it made sense from a purely technical perspective, it let the vast majority of their buyers know what model they were looking for.

There's also the fact that for an average user (in other words the vast majority), they're better off with a 5Ghz 4/4 than a 4Ghz 8/16.
 

PeterScott

Platinum Member
Jul 7, 2017
2,605
1,540
136
There's one massive logical fallacy you're making and that's that the vast majority makes reasonable assessments. The vast majority of people don't know/care that it gained two more cores. The vast majority of people don't even know how many threads they currently have or their clock speed. You're generally doing pretty good if they can tell you "I think I have an Intel core something. Maybe a 5? ". Not once in 15 years of building PC's have I had a normal user come up to me and tell me they think they need more cores. They just want to know if they're buying the newer version of what they have and if it's faster. They know an i5 is better than an i3 and an i7 is better than both because the number is bigger so it's obviously faster.

It's was the same with the BMW's. 1 series was your small entry level, 3 series was your midsize entry, 5 series was your upscale mid size, and 7 series was your luxury boats. A 330 was better than a 328 and and 335 was better than both. The vast majority of their buyers were unaware of the fact that the 135, 335, and 535 all used the same engine and the 135i was actually the quickest of the 3. It doesn't matter if it made sense from a purely technical perspective, it let the vast majority of their buyers know what model they were looking for.

There's also the fact that for an average user (in other words the vast majority), they're better off with a 5Ghz 4/4 than a 4Ghz 8/16.

You aren't really making any kind of case that an i7 8C/8T somehow alters things for the oblivious. An i7 will still be better than an i5...
 

Ken g6

Programming Moderator, Elite Member
Moderator
Dec 11, 1999
16,218
3,796
75
I wonder if they're doing this to fight SpectreRSB, the new variant that exploits hyper-threading.

Otherwise, I don't see why they wouldn't sell some variant of these chips with HT, say for $10 less than the next step up?
 

PeterScott

Platinum Member
Jul 7, 2017
2,605
1,540
136
Looking at Anandtech's review of the i7-8700K/ i5-8400, the 4/8 core 7700K is almost always faster than the 6/6 i5-8400. Granted this is probably because of the 7700K's extra cache and clock speed, not hyperthreading. Although if Intel doesn't keep the clock speeds of this new i7 decently higher than the old 8700K, I think we could see some benchmark upsets.

It's the clock speed. There is a big delta between i5-8400 and 7700K.

I don't expect much of a clock bump for 8C CFL, maybe 100MHz, but at least it shouldn't be behind, and overall at even clock speed, I expect the overall win will favor 8c/8t, over 6c/12t, there may be a couple of benchmark where 6c/12t squeaks a lead, but overall 8C CFL should beat it.
 

jpiniero

Lifer
Oct 1, 2010
14,509
5,159
136
Otherwise, I don't see why they wouldn't sell some variant of these chips with HT, say for $10 less than the next step up?

As I mentioned, I think it's possible that the 65W and 35W i7 models (as well as mobile parts) are still 6/12.
 

PeterScott

Platinum Member
Jul 7, 2017
2,605
1,540
136
I wonder if they're doing this to fight SpectreRSB, the new variant that exploits hyper-threading.

Otherwise, I don't see why they wouldn't sell some variant of these chips with HT, say for $10 less than the next step up?

I don't see how that helps at all. Hyperthreading isn't going away. They are still going to sell the 8C/16T hyperthreading models. So that is no kind of solution.

It's a naming shift nothing more. Attaching a premium i9 tag to the 8C/16T HT part is all this is about.
 
Aug 11, 2008
10,451
642
126
Yea, I agree it is a clickbait headline. Hyperthreading is still available on the "i7", i.e. top of the line chip, they just call it i9 now!!!

In any case, thinking more about this, nobody is going to miss a hyperthreaded quad when hex cores are around 200.00. I do see a problem with six cores though. To get the multithreaded performance of the current i7 8700 (non-K, around 330.00) one would have to move up to the 8 core which will probably be more expensive. It will also force the competition to the ryzen 2600 to be the 8 core, which could also put them at a price disadvantage. Really what they should have done to eliminate overlap was to enable hyperthreading on *all* lines, and differentiate the top of the line 8 core with a standard (locked, lower clockspeed) version and a balls to the wall k version. This might decrease yields though, I would assume some chips have defective hyperthreading resources. Otherwise they would have had nothing to lose. But perhaps they will continue to sell the 8700 and 8700k as well.
 

XavierMace

Diamond Member
Apr 20, 2013
4,307
450
126
You aren't really making any kind of case that an i7 8C/8T somehow alters things for the oblivious. An i7 will still be better than an i5...

You'll note that at no time have I said I agree with all the complaints about the model numbers. But, that statement depends on the clock speeds. If Intel sacrifices clock speed for these extra cores you could legitimately end up with a newer processor of the same model range being slower than the old one for the average user. Take a look at the server side Xeon E5-2650 v2, v3, v4. They've been sacrificing clock speed for cores, which on servers is generally a decent decision. Single thread wise though, the v4 is actually the slowest of those 3 generations.
 

Asterox

Golden Member
May 15, 2012
1,026
1,775
136
Here is what the article mentions:



So $350 for 8C/8T and $450 for 8C/16T.

That doesn't seem too bad.

Yes not to bad vs cheaper R7 2700X, "nobody cares for SMT+ it will be sold without CPU cooler".
 

kjboughton

Senior member
Dec 19, 2007
330
118
116
I wonder if they're doing this to fight SpectreRSB, the new variant that exploits hyper-threading.

Otherwise, I don't see why they wouldn't sell some variant of these chips with HT, say for $10 less than the next step up?

This dont make no sense.

We know there are no major architectural differences across consumer market segments (just MORE cores or MORE cache or MORE speed or MORE access to built-in functionality). It's been this way since the start of Core.

Furthermore, a solution to a security issue is not optional. If this were the solution on i7 it would also be the solution for i9 (see previous paragraph).

Thanks for playing.
 

PeterScott

Platinum Member
Jul 7, 2017
2,605
1,540
136
Yea, I agree it is a clickbait headline. Hyperthreading is still available on the "i7", i.e. top of the line chip, they just call it i9 now!!!

In any case, thinking more about this, nobody is going to miss a hyperthreaded quad when hex cores are around 200.00. I do see a problem with six cores though. To get the multithreaded performance of the current i7 8700 (non-K, around 330.00) one would have to move up to the 8 core which will probably be more expensive. It will also force the competition to the ryzen 2600 to be the 8 core, which could also put them at a price disadvantage. Really what they should have done to eliminate overlap was to enable hyperthreading on *all* lines, and differentiate the top of the line 8 core with a standard (locked, lower clockspeed) version and a balls to the wall k version. This might decrease yields though, I would assume some chips have defective hyperthreading resources. Otherwise they would have had nothing to lose. But perhaps they will continue to sell the 8700 and 8700k as well.


Good point about 8700 non K. I hadn't really thought about how they cover that one. I guess we will have to wait and see now it all shakes out.

I also agree that enabling HT everywhere, might have been a good move, but Intel is so used to charging a premium for HT, that it is nearly hard wired at this point.

Plus maybe that is what the lineup gets for next years 10000 series, xLake on 14nm +++++. HT for everyone! Need something in reserve when 10nm just keeps failing.
 

epsilon84

Golden Member
Aug 29, 2010
1,142
927
136
Yes not to bad vs cheaper R7 2700X, "nobody cares for SMT+ it will be sold without CPU cooler".
In favour of 9700K: Higher boost clocks, higher IPC, ability to overclock to 5GHz+

Will that be enough to compensate for the lack of HT/SMT? In some cases yes, in some no.

Overall they will be quite evenly matched IMO

Fwiw the Prism cooler that comes with the 2700X is quite noisy especially under full load. It looks pretty though! It's more of a bonus selling point for the 2700X IMO