Next Gen Core i5 Desktop Processor Confirmed to Feature HyperThreading

erek

erek

[H]F Junkie
Joined
Dec 19, 2005
Messages
10,897
Finally HTT Support!

"A 3DMark results database entry confirmed that the 10th generation Core i5 desktop processor will indeed feature HyperThreading (HTT). Based on the 14 nm "Comet Lake" silicon, the next-gen Core i5 processor will be 6-core/12-thread. Besides HTT, the processors will feature higher clock-speeds than their 9th generation counterparts. In the 3DMark validation, a Core i5-10600 processor is referenced, featuring 6 cores and 12 logical processors. The chip has a nominal clock-speed of 3.30 GHz in its name string (a 200 MHz increment over the i5-9600), although its Turbo Boost frequency hasn't been detected properly by SystemInfo."

https://www.techpowerup.com/262020/...processor-confirmed-to-feature-hyperthreading
 
So they will stop disabling a feature of questionable benefit, so they can do the old sales man standby, "Look what you give up by going down a step... double the cores."

I'm sure it will perform basically exactly the same as the previous refreshed refresh in everyday that matters and continue being a far worse value proposition compared to a 3600. Of course depending when these ship for real it may be more bad value proposition vs a 4600.
 
ChadD said:
So they will stop disabling a feature of questionable benefit
Click to expand...

Questionable benefit?

https://www.techspot.com/review/1730-intel-core-i9-9900k-core-i7-9700k/#:~:targetText=The Core i9-9900K is,enabled for 16 logical threads.&targetText=The Core i7-9700K packs,by a marginal 100 MHz.

tldr;
9700k has no hyperthreading, 9900k has hyperthreading, in productivity/content creation apps there is a 25-35% delta in performance. In some circumstances the 8700k which has two fewer cores is actually faster than the 9700k by virtue of hyperthreading.

This is echoed through most review sources, and personally is the key reason why I went 9900k.
 
Last edited:
Keljian said:
Questionable benefit?

https://www.techspot.com/review/1730-intel-core-i9-9900k-core-i7-9700k/#:~:targetText=The Core i9-9900K is,enabled for 16 logical threads.&targetText=The Core i7-9700K packs,by a marginal 100 MHz.

tldr;
9700k has no hyperthreading, 9900k has hyperthreading, in productivity/content creation apps there is a 25-35% delta in performance. This is echoed through most review sources, and personally is the key reason why I went 9900k.
Click to expand...

Well hyperthreading isn't the main thing to focus on there. 25% more cache and a 100 mhz.

25% more cache.... 15-20% bump in performance when cache is getting nailed (multi threading) yep makes sense.

Hyper threading is of questionable advantage in basically any situation. Don't get me wrong if its designed in why turn it off. Only reason Intel ever disabled it on i5s was to give the sales folks a bullet to fire at people considering them. Go i7 and get double the "cores".

The biggest bump to multi threaded performance (beyond actual real cores) has always been increased amounts or enhanced operation cache. It gives branch predictions the space they need to stretch out and do their thing.
 
ChadD said:
Well hyperthreading isn't the main thing to focus on there. 25% more cache and a 100 mhz.

25% more cache.... 15-20% bump in performance when cache is getting nailed (multi threading) yep makes sense.

Hyper threading is of questionable advantage in basically any situation.
Click to expand...

No. Please read this: https://arstechnica.com/features/2002/10/hyperthreading/

In intel's case, you get 30% more processor resources effectively, in AMD's case it is closer to 40% - provided your software is threaded sufficiently to be able to feed those extra front end units.
 
Keljian said:
No. Please read this: https://arstechnica.com/features/2002/10/hyperthreading/

In intel's case, you get 30% more processor resources effectively, in AMD's case it is closer to 40% - provided your software is threaded sufficiently to be able to feed those extra front end units.
Click to expand...

Well you can believe articles based on 17 year old marketing if you wish. SMT/HT can absolutely gain you that much on very specific things (on 17 year old CPUS)... over the years it has also become less and less helpful as cache size and core count have increased. One of the main advantages 17 years ago was with out massive amounts of L2/L3 cache branch prediction was rather basic. Today branch good branch prediction makes SMT less valuable. Both Intel and AMDs Zen2 use a almost identical branch predictor that does a far better job of keeping real cores feed the the basic stuff on earlier CPUs.

SMT is not a useless feature... and I'm not saying that. It is however of questionable value on a lower core part with a decent amount of cache and a modern branch predictor imo.

SMT increases pipe length and increases latency, and in the majority of the cases provides no real uplift in performance. (in many cases it degrades performance... its not hard to search youtube and find 1001 vidoes of people running games on their Intel and AMD machines with and without SMT on. It is a wash in most cases with SMT off often gaining low to high single digits. We have enough cores now, latency is BAD)

I imagine Intel is trying to appeal to OEMs / the clueless with a move to re enable HT on I5s. It looks better on a spec sheet and that is about it.

Again don't get me wrong I didn't say it was useless... just of questionable benifit to i5 users. (better to be engaging 6 real cores for the type of work people would be doing on a 6 core chip... lite stuff and gaming) Without a doubt 9 times out of 10 a i5 with HT with it disabled will perform better.

Don't take my word for it. lol
https://www.intel.ca/content/www/ca/en/architecture-and-technology/mds.html
According to Intel disabeling HT on a 9900 depending on the situation means NOTHING... a loss of at most high single digit % wise or an actual gain in performance.

HT is pure marketing. It made more of a difference perhaps when we had dual core chips.... or on the other end if we are talking about 16+ core systems doing 16+ core work where SMT CAN leads to major power efficiency gains, with the wrong workloads it can also do the exact opposite. In super computing clusters there are cases where they compile their work with SMT disabled to avoid having it degrade performance.
 
Last edited:
Alright, I am going to agree to disagree for now cause I can’t be bothered looking at more examples. I will say that added cache and 100mhz will not give you >10% gains at ~5ghz

Back in the Athlon 64 days there were two variants of the 3200+

One was 3ghz with 1megabyte of cache, the other was 3.2ghx with 512k of cache, yet they were both rated to the same performance. To put this into perspective, double the cache, equal to 200mhz or 7.5% worth of performance (in fact, a good deal of the time the 200mhz won out)
 
Last edited:
Keljian said:
Alright, I am going to agree to disagree for now cause I can’t be bothered looking at more examples. I will say that added cache and 100mhz will not give you >10% gains at ~5ghz
Click to expand...

It will absolutely. Not the frequency ya no doubt not a big deal. But going from 12mb to 16mb of L3 cache... yes for memory intensive multi threaded workloads ya that will easily be 20% more performance.

CPUs fill their L3 cache and purge the oldest data. In multi threaded workloads almost all modern chips purge data that could be used later cause they just don't have enough room. SMT increases the demand on this memory space... which is why it a lot of cases performance goes up if you turn SMT off. A modern branch predictors job is to decide what it should be holding on to in that cache. So a combination of a smarter BP and more Cache diminish the potential gains from SMT. I sort of doubt anyone ever does away with SMT at this point... but I would expect in another generation or 2 it will make very little sense to include SMT for any reason outside of marketing. Branch prediction and large cache spaces will render SMT a liability.

No doubt when Intel first made HT a marketing term back with the Pentium 4... it could gain you 30 and even 40% bumps in at least some real world things. Over the years though BP and cache increases have reduced those potential gains, and in more and more cases SMT is actually a performance detriment.

I wish I could find you some good recent bench marks of CPUs that are nearly identical accept their cache systems but with all the CPU skus these days, its not like the old days where Intel or AMD are selling low end chips with no cache or something. :) The most recent stuff you will find probably goes back to the core days when they sold 1 2 4 MB cache versions of basically the same chips, and ya back then you would see 20-40% bumps in performance between the SKUs. Nothing has changed.
 
Last edited:
ChadD said:
Again don't get me wrong I didn't say it was useless... just of questionable benifit to i5 users. (better to be engaging 6 real cores for the type of work people would be doing on a 6 core chip... lite stuff and gaming) Without a doubt 9 times out of 10 a i5 with HT with it disabled will perform better.

Don't take my word for it. lol
https://www.intel.ca/content/www/ca/en/architecture-and-technology/mds.html
According to Intel disabeling HT on a 9900 depending on the situation means NOTHING... a loss of at most high single digit % wise or an actual gain in performance.
Click to expand...


Urrr no- You're comparing synthetic benches vs real world workloads, that will never end well.

Please back up your statement re "without a doubt 9 times out of 10 an i5 with HT disabled will perform better" - in games? probably a wash, in productivity, I say based on the above results, HT theory, and reviews, I'd say definitely not. Worst case it will perform 0-5% better, best case it will perform 25-30% better.
 
You must log in or register to reply here.
Back
Top