Should Intel Provide Different SKUs Based On Hyperthreading Alone? (Discussion Thread)

[TechyInAZ]

Dividing SKUs based on hyperthreading is no new thing for Intel who have been doing this for over 15 years starting with the Pentium 4; now that hyperthreading/SMT is a perfectly reliable technology, what do you think about Intel's strategy for making different SKUs of processors? Do you believe that all Intel CPUs should be hyperthreaded or not? Lets discuss...

(FYI, yes this discussion thread was made in part because of the rumored i7-9700K.)

 

[velocityg4]

I'd like a system where iX denotes number of cores and hyperthreading. Even numbers being number of cores and odd number to signify number core (number -1) and hyperthreading. I hate that an i7 can mean dual core through octo core.




For Example

- i6 Hexa Core
- i7 Hexa Core + Hyperthreading
- i18 Octodeca Core
- i19 Octodeca Core + Hyperthreading

While I do like the K, U and Y suffixes which denote other characteristics and the first number denoting generation. I'd also like the remaining number to denote base clock in Ghz.

In this system and i7-7820X would become an i9-7360X. 8 Cores, Hyperthreading, 7th Gen, 3.6 Ghz.

As for Hyperthreading being standard. I'm curious if there is even any physical difference between an i5 and i7. Besides a small difference in Cache. It seems as though an i5-8400 for example is the exact same chip as an i7-8700 with different factory clock rates and hyperthreading turned on or off.

If there is physically do difference between hyperthreading and non-hyperthreading. I see no reason from a consumer perspective why hyperthreading shouldn't be included. Since there would be no cost increase. Financially keeping them separate is good for Intel. It means higher profit margins on a CPU with little cost difference. Besides 3MB Cache. Because no one would pay that much extra for just a little more Cache.

 

[James Mason]

I would also like velocityg4's idea to be implemented. You could know 98% what a CPU is from name alone without having to rely on memory.

The reason Intel doesn't is because i7 has name recognition, so they can use it to trick people into thinking they're getting something good. Like this i7-3667u tablet-laptop-detachable thing I have at work that's only a 2core 4 thread at 2ghz.

 

[Barty1884]

I also like velocityg4's idea.

I think at this point, there[strike] is[/strike] was an understanding of Intel's product line. I don't understand why they went the routes they have recently (and are rumored to be doing).

i3 = 2+HT
i5 = 4
i7 = 4+HT

I'm surprised they didn't take the opportunity presented (ie 3/5/7 being copied) to either supplement the product stack, or launch a new product stack entirely.

Obviously it's irrelevant now.... but if it were up to me:

i3 = All low end stuff (replacing Pentiums/Celerons etc). Various cores w or w/o HT. There can be some ambiguity in this space.
i5 = 4 core
i6 = 6 core
i7 = 4 + HT
i8 = 8 core
i9 = 6/8 + HT

In sheer numbers, it's no more product lines than they've ever had really.
Recognition of the existing 3/5/7 lines would be helpful and, of course, higher numbers = better - so the 6/8/9 chips would sell themselves on that basis.

While the i7 dropping from the top of the (consumer) stack to middle of the road would be a sort of bait & switch, it's not like that hasn't/doesn't occur already - Moreso in the mobile space though..

As JM mentioned, the i7-3667U is a 2/4 chip.
My work laptop uses an i7-6650U which is 2/4 also 2.2 to 3.4GHz.
Now there are differences between it & the i5/i3 equivalents but, at a high level:
i5-6360U = 2/4 @ 2.0 - 3.1GHz
i3-6100U = 2/4 @ 2.3GHz.

Now the i7 I have is a "U" ("ultra" low power?), so bound by the intended TDP.
BUT, I would argue that if you're looking at a low power machine, and you're comfortable with 2/4, then you could buy an i3.

If you wanted the i7 (4/8) then a 35-45W TDP is cost of entry IMO.


With the rumored i7-9700K. I don't like it.
To me, an i7 has Hyper-Threading..... that's just what an i7 has.

 

[IInuyasha74]

I would say yes they all should be, as HyperThreading improves performance for a minimal increase in power consumption. The gains in efficiency are too huge to ignore. And all CPUs have the hardware for HyperThreading, so there isn't any change in cost to have this.

Really, unless there is a sizable number of cores that have defects in the HyperThreading section of the die, I think they all should have it enabled. If too many do have problems with defects, then non-HyperThreaded SKUs should be limited to Celeron only.

The product line-up should be this way I think.

Celeron - non-HyperThreaded Core processors or Atom SKUs.
Pentium - dual-core + HT + TB
Core i3 - quad-core + HT + TB
Core i5 - hexa-core + HT (No turbo boost)
Core i7 - hexa-core and up with HT and turbo boost

The Core i5 units I think are the only ones that shouldn't have turbo boost also. Just because there needs to be something to differentiate them from the Core i7 models, and the easiest way to do that is with reduced clock speeds.

 

[genz]

3 things as to why I'd have to disagree with the above.

1. All of this is informative to us, but not the user. For example, a Nehalem based quad core is inferior to a Ivy Bridge by a LONG stretch, but an Ivy Bridge CPU is clock for clock not far off Haswell or Broadwell, so using this naming, a i7-(1)350K vs i7-3320K looks like a similar battle to an i7-2350K vs an i7-4320K.

2. Same can be said for an i13-5100 vs an i3-2350. The latter would be vastly better for the average user.

Which brings me to 3. Any naming strategy that makes too much sense leads customers to think that understanding Cpus is intuitive. It's not. You HAVE to read up on it to know what you're talking about. A is not always better than B. Don't make it too easy or the user feels stupid asking for explanations. That's what Intel did, and that's what AMD did before Intel, because arch differences don't have any hard and fast rules.

As to OPs question, SMT should be standard on all above-Pentium systems as an energy saving feature, not a performance feature. Doing two sums in a single clock uses a fraction of the power and should be seen as such. Intel need to vastly revisit their product range for Mobile as they have obviously been taking money under the table for breaking naming conventions they set themselves in order to sell product to a misled user (MacBook i9 and X99/299 naming issues, Core m invention and dozens of other examples)? i3/i5/i7/i9 should be retained as a naming strategy for 4/6/8/8+ cores respectively, with dual cores taking pentium naming, without SMT for minimum possible prices. Burn off anything else that could be an issue to maximise yields, just make it boot windows lol.

This should also apply to mobile, and mobile names should match their equivalently performing desktop part e.g i5-7600U is actually much slower than 7600K due to TDP being much lower, and thus can't do max clock most of the time, even heat allowing. As long as the naming is consistent from there I don't mind.

I also think it should be law to put 3 independently verified benchmarks on every box.

But pointing the finger at Intel here belies the fact the whole industry does this. It's called rebranding. From 720p SD to 'HD Ready' to 'LED Backlit' (which means LCD but they used 1 LED white light at the back, not LED screen). From 4K (4096) to UHD(3840). AMD's whole range of R7/9. Nvidia's 8 series, twice! (GT8X00 series and GTX 8XX mobile)

 

[Barty1884]

@genz, I'm not quite following what you're saying there.

Why would the battles in point (1) appear similar? I don't think anybody is suggesting these naming schemes would remove the need for independent verification, just that it would make it a little easier at a glance.

For an average consumer, 3000 > 700..... 4000 > 3000 etc. No matter how small/large the improvement may have been between generations, they're accurate enough at a glance -- comparing the same product stack anyway.

In point (2), why would an i3-2350 be vastly better than an i3-5xxx?

 

[genz]

@genz, I'm not quite following what you're saying there.

Why would the battles in point (1) appear similar? I don't think anybody is suggesting these naming schemes would remove the need for independent verification, just that it would make it a little easier at a glance.

For an average consumer, 3000 > 700..... 4000 > 3000 etc. No matter how small/large the improvement may have been between generations, they're accurate enough at a glance -- comparing the same product stack anyway.

In point (2), why would an i3-2350 be vastly better than an i3-5xxx?

You're thinking from your point of view. This is about a user that doesn't know generational differences even exist. We're all way ahead of the average laptop buyer.

There are multiple products in the stack (different generations as well as completely different cores e.g Atom/CoreM/Xeon/Phi) which would be integrated via lettering (e.g i3-2140A to indicate 2nd gen dual Atom@1.4). I hate fragmentation. On the point of clarity, it may be simpler to just separate the numbers then. i3-2A140 if that's what it takes for people to understand the first digit hasn't any correlation with the next few.

Thinking on it now, there are other reasons why gen is important (features like QS and iGPU) but actually it's mostly irrelevant for desktop.... and the complete opposite for laptop. If I put time into this I'd probably move that number to last and start requiring some indication of average power draw as that is really the master of performance in mobile. Above Atom becomes i3-A14 v2 15W with a little twist from Xeon. State that the TDP must match the BIOS setting in marketing materials and we'll weed out all the poorly designed premium laptops with crappy cooling and BIOS throttling to make up for it. When you take this idea, the i5 1350 vs i5 3320 becomes i5-35 v1 vs i5-32 v3. Laptops releasing either would print i5-32 v1 40W for example to indicate cooling TDP avaliable.

ll. Not i3, i13 following GPs idea of i number being an indicator of cores. I was stating that a 12 core 1Ghz chip would be much slower than a 3.5 dual core to most users, even with HT.

 

[TerryLaze]

https://www.tomshardware.com/reviews/cpu-hierarchy,4312.html
Intel's 6/12 i7-8700K is already matching the best AMD has to offer even in applications so it makes perfect sense for intel not to make a SKU too powerful because they don't want to end up like AT&T.

 

[theonerm2]

Celeron Should be 2 cores with hyperthreading
Pentium should be Quad Core
Core i3 Should be Quad Core with hyperthreading or 6 cores
Core i5 Should be 8 Cores
Core i7 Should be 8 Cores with hyperthreading
Core i9 Should be 10+ Cores with hyperthreading.

That's where Intel should be with recent processors in my personal opinion.

 

[Barty1884]

I think asking for 4+HT in the i3 space is asking a lot, considering 4+HT was "high end" only a gen/two ago.

In a perfect world though, I agree with you.

 

[theonerm2]

I think asking for 4+HT in the i3 space is asking a lot, considering 4+HT was "high end" only a gen/two ago.

In a perfect world though, I agree with you.

I think Intel should have made 6+HT the new high end for mainstream CPU's back in 2013 with the 4770K. Then I think they should have bumped it up to 8+HT with Coffee Lake and that's for the 8700K. The i3 has generally been half an i7 minus the boost and lower clocks. I think if the i7 is 8+HT it's logical to expect 4+HT out of the i3. The 4+HT CPU was high end for too long. A lot longer than it should have been. With AMD you can get a 4+SMT or a 6+SMT CPU for i3 money. The i3 8300 is $150 and Ryzen 5 1600 is $150 that's a 4 thread CPU vs a 12 thread CPU for the same price. Even the Ryzen 5 2600 is being sold for i3 level money. If AMD can offer 4+SMT to 6+SMT for i3 level money why can't intel offer 4+HT for i3 prices? That makes AMD the price to performance leader right now I think.

 

[Barty1884]

Not disputing the value proposition of Ryzen, as I totally agree.

Ultimately, when Intel brought the 4770K, AMD were just starting to beat the dead horse in 'Dozer cores.
If AMD had remained competitive, Intel would've had to have done something. They didn't, so Intel didn't have to up the ante.

FWIW, not all cores are created equal, so the IPC gains of an i3 vs an R5 still keep it viable for certain use cases.....but it's hard to argue with the value of Ryzen.

 

[genz]

I think Intel should have made 6+HT the new high end for mainstream CPU's back in 2013 with the 4770K. Then I think they should have bumped it up to 8+HT with Coffee Lake and that's for the 8700K. The i3 has generally been half an i7 minus the boost and lower clocks. I think if the i7 is 8+HT it's logical to expect 4+HT out of the i3. The 4+HT CPU was high end for too long. A lot longer than it should have been. With AMD you can get a 4+SMT or a 6+SMT CPU for i3 money. The i3 8300 is $150 and Ryzen 5 1600 is $150 that's a 4 thread CPU vs a 12 thread CPU for the same price. Even the Ryzen 5 2600 is being sold for i3 level money. If AMD can offer 4+SMT to 6+SMT for i3 level money why can't intel offer 4+HT for i3 prices? That makes AMD the price to performance leader right now I think.

A bit of an arbitrary time to pick for 6 cores. Intel would be 4+ main releases after the last core lift. Imo Intel should have retained their older fabs or hired out space to produce older core refreshes longer and brought up cores there on 2011, so there was a point. Ultimately the only reason they are bringing up cores now is that they cannot get 10nm done or they may well have ignored AMD's release.

The main reason Intel didn't go past 4 is the same reason AMD did. AMD's yield for multicore doesn't go down with Zen, as they fab each core then put it on the infinity fabric base like lego bricks. Intel make the whole core at once, meaning even on it's a top spec CPUs there are disabled cores expected, hence every core doubling doubles failure rate for a whole CPU class. Big Xeon can have 6+ cores dead on the die. This is also why HT isn't on top Coffee Lake.

Late update: https://www.semiaccurate.com/2018/09/07/intel-cant-supply-14nm-xeons-hpe-directly-recommends-amd-epyc/

This is why Intel can't really up core count like AMD. Because of thier yield problems even on MM-C 14nm they are already at peak output without switching over consumer fabs, which would mean having supply problems at the lower end.