HoodAll unlocked, all soldered

That'll never happen from Intel. AMD would have to beat them to within an inch of their life for that to happen. Though, even I don't think they need to all be soldered. Looking back at the golden C2D days, not even all those chips were soldered. They got away with it on I think E5xxx on down, and nobody complained. I think they can do the same here. Probably any i series chip should be soldered, because that's when higher power draw (and higher temps) starts to be a factor. Pentiums and such can take paste easily.

All unlocked? They've spent years shaving down OC capability by making it impossible to raise BCLK in an amount any more significant than farting on a brick wall, and limiting, and later eliminating, what you can do with non-K chips. If it were possible to actually move BCLK like you could the old FSB, I'd be happy with a regular i3... and I wouldn't care much about unlocked multipliers. They were always there in the extreme edition CPUs, but I always got by on non extreme CPUs. I've even had a few of those unlocked Black Edition AMD chips, and the unlocked multiplier didn't matter to me much then, either.

lynx299700k might be best bang for buck, as a gamer i don't need HT, and 8 cores soldered and 4.6 out of box, very possible it could OC to 5.3 ghz on a nocuta nh-d15 even. 8 cores at that speed... = gaming domination for many years to come. 8 cores 16 thread prob will only be able to do 4.9ghz or 5 if lucky. i don't see the point of going past 8c 8t if your just a gamer.

It should be but I think the 9900k could be binned by default, hence the higher turbo, so 5.3 will be hard(er) to achieve on the i7 IMO.
As with most Intel chips, the halo products are usually the best, not in terms of VFM though.

lynx29i don't see the point of going past 8c 8t if your just a gamer.

It depends on how long you want/need to keep the system. A lot of folks need to keep a system for 4+ years and need to buy the most raw performance (i.e. hardware) they can afford today so they get a good experience as long as possible.

In that scenario, with core counts increasing from both companies, developers should be taking note and making their software (including games) more parallel/core/thread aware.

So in e.g. 2 years an 8/16 or 16/32 CPU may perform better than it does today (including games) and better than an 8/8.

If you can afford new, top of the line hardware and the possible hassle of reinstalling your OS and software (or can pay someone to do it) every 6 to 12 months, then it doesn't really matter.

As usual, those with $$$$/€€€€ have a much easier time.

Vya DomusThe usefulness of HT decreases with the number of cores to the point where it might even hurt multithreaded performance. This isn't exactly news but it's largely unknown to most people.

www.pugetsystems.com/labs/hpc/Hyper-Threading-may-be-Killing-your-Parallel-Performance-578/

Single core performance can be hurt too by hyperthreading.

That is in one application. Cores is better than SMT but I'm sure there are applications in which it always is a help. Linux seems better at handling threads than Windows.

EDIT: He mentions in the comments how cinebench benefit from SMT (it's a bench program though so not really interesting), but in the article he mentions how Zenmax Opticstudio scaled fine with HT on.

RejZoRAnd i3 with just 4 physical cores, no HT. No more 2c/4t nonsense. Ryzen 3 cut that crap from the beginning.

They have their place at the low end, and they absolutely kill it in any price/performance metric you care to mention. The G4560 is one of the best chips ever made tbh, and one could only dream of what it would be if it were unlocked...

R0H1TIt should be but I think the 9900k could be binned by default, hence the higher turbo, so 5.3 will be hard(er) to achieve on the i7 IMO.

All products are sorted in bins at the factory. i7-9700K and i9-9900K are obviously different bins, so technically all products are "binned".
Some people mean something different with "binning", like the sub-binning some of the AiB vendors do for GPUs, where they re-test the GPUs and determine which ones are better.

In average, I believe any overclock x will be a little harder to achieve on i7-9700K vs. i9-9900K, there might be a little difference if you're trying to set a record. But sometimes you are bottlenecked by other factors leading to a similar maximum overclock. Sill, you have to remember that overclocking is becoming more and more just symbolic with such high clocks out of the box. Now you need to bump voltage and have extreme cooling just to get a few hundre MHz extra, it's not like old Sandy Bridge any more, where easily get a good overclock before even touching the voltage. If you're overclocking for the experience of overclocking, then go ahead, that's the only reason to do it at this point.

FrickThat is in one application. Cores is better than SMT but I'm sure there are applications in which it always is a help. Linux seems better at handling threads than Windows.

The Linux kernel is surely much better at scheduling, and even allows a lot of finetuning for various workloads, but that applies to specific server workloads.

SMT scales poorly with many applications at once, part of it is the OS kernel's fault, and part of it is that any synchronous task will suffer from the added latency.

FrickThat is in one application. Cores is better than SMT but I'm sure there are applications in which it always is a help. Linux seems better at handling threads than Windows.

Of course there is software that see an uplift in performance when SMT is enabled , they aren't wasting die space for nothing. But for most mainstream use it's sort of useless unless we talking about dual cores where you really need to squeeze the last drop of performance. SMT in high core count CPUs comes into play where you run a lot of threads simultaneously and you have a lot of scheduling overhead associated with them, for pure FPU/ALU number crunching it's pretty useless most of the time.

I think there's a mindset in this thread that misses the actual application of hyper-threading. Yes, it is true that a physical core is better than SMT, but what about real-world scenarios? In typical PC use you almost never have a use for 8 threads, 6 threads, or even 4 threads; but in the scenarios where additional threads aid productivity there is almost a limitless benefit from additional threads (e.g., encoding, compression). In that sense a 6C/12T will always be better than an 8C/8T clock for clock because there is no highly multi-threaded scenario where 12T does not have an advantage over 8T and yet no scenario where 8 logical cores has an advantage over 6 logical cores.

BlueberriesI think there's a mindset in this thread that misses the actual application of hyper-threading. Yes, it is true that a physical core is better than SMT, but what about real-world scenarios? In typical PC use you almost never have a use for 8 threads, 6 threads, or even 4 threads; but in the scenarios where additional threads aid productivity there is almost a limitless benefit from additional threads (e.g., encoding, compression). In that sense a 6C/12T will always be better than an 8C/8T clock for clock because there is no highly multi-threaded scenario where 12T does not have an advantage over 8T and yet no scenario where 8 logical cores has an advantage over 6 logical cores.

No, 6C/12T will always be worse in general workloads than 8C/8T. SMT doesn't give extra cores in any way, it just utilizes idle cycles for other threads. The total performance of any SMT threads of a single core can never exceed the maximum theoretical performance of that core. In HT it's two threads, in IBM Power it's 4 or 8, but these threads share the performance of one core. So A 6C/12T CPU scales towards a fully saturated 6C CPU, never beyond that.

SMT is really challenging for the OS kernel scheduler. The scheduler measures the wait time of each thread to load balance the cores, matching more and less intensive threads together so the total throughput is as high as possible. This means threads are shuffled around a lot, creating a lot of latency for single threads, but may give slightly higher total throughput. SMT only scales well you have one large workload of similar threads, but scales terribly when you have a mix of high, medium and low loads, especially if they keep changing constantly.

cucker tarlsonSilly point considering how much x99 and 5820k cost.

Only reason for gamers to choose HT CPUs was the lack of CPUs with enough physical cores, HT was always a half assed solution. It was much smarter to go with 4790k/5775c and upgrade to 9700k now than go with x99, pay a sick premium for something that neither was the best solution back in 2014-16 nor is it now.


You are making a very stupid mistake in your thinking. Just cause games can utilize HT does not mean there's equvalency between physical cores and HT, e.g. 8c/8t and 6c/12t. There isn't. Physical cores will always be better and HT is known to take a hit on games that rely heavily on single threaded perfromance

www.purepc.pl/procesory/test_procesora_intel_core_i7_8700k_premiera_coffee_lake?page=0,42

Someone here did a comparison about month ago. The conclusion was that HT was not to be dismissed, on two of the test he ran there was no difference with HT on vs a same core count chip with it off and the other, I don't remember the other graph but it also could be considered an outlier. The rest HT matched about 80% of an actual core, it pretty much boils down to pricing whether HT is worth it or not, but downplaying it outright as you are is just silly.

TotallySomeone here did a comparison about month ago. The conclusion was that HT was not to be dismissed, on two of the test he ran there was no difference with HT on vs a same core count chip with it off and the other, I don't remember the other graph but it also could be considered an outlier. The rest HT matched about 80% of an actual core, it pretty much boils down to pricing whether HT is worth it or not, but downplaying it outright as you are is just silly.

I'm downplaying it just the right amount. It's a must have on four cores, it's doing much less when you have six cores, and it's better to simply have 8 physical cores with no HT than both of the former. I'm only speaking of gaming though, in productivity/workstation use 8700K will probably outperform 9700K.

Remove it so you can add back a few years later as a new feature / marketing argument.

ExV6kRemove it so you can add back a few years later as a new feature / marketing argument.

That will happen when Intel got their 10nm process right and got the thermal headroom as they're known for penny pinching

A feature is a "feature" only when it does something for you. While Ivy Bridge certainly suffered from the change from solder, since Haswell, I haven't found a reason to care. I really don't care if we can "delid and CPU will drop 10C" ... if it doesn't deliver anything for all that work. Have not a CPU OC limited by temps since Haswell so really don't care if if core temps are 75C or 65C under stress testing because Im just fine with 75C. With voltage creeping to to peaks of 1.40+ (1.5V peaks under AVX) under stress testing, I'm not going to raise voltage any higher so whether its 75C, 65C or whatever, I just don't care. If it need solder to do what I need, then I hope that they use it or I'll skip it ... if theyt use whatever, to get what I need, I'm fine with that.

John NaylorA feature is a "feature" only when it does something for you..

That could applied to the craze about more cores. Few of us need them...or have software that utilizes it.

Yet.

As I have said before, if we don't have these new techs, no one will use them. If we have them, there is more chance of them being used.

It's a bit chicken and egg, but I would rather have them and not need them than need them and not have them; assuming I can get them within my budget. ;)

StrayKATFew of us need them...or have software that utilizes it.

That's a misconception, even your browser is multithreaded, the issue is it doesn't scale linearly. When someone goes from a quad to an 8 core and doesn't see 2x the performance in everything they mistakenly conclude that those extra cores are useless.

Agreed. I actually wish some reviewers (of everything) would add a multi-tasking section.

Do some performance analysis of e.g. copying a load of files in the background while gaming/watching video.

Running some video encoding while gaming/web browsing etc. etc.

i.e. real world multi-tasking scenarios. Given the media content explosion, I would think these kinds of activities are becoming more common.

Likewise including streaming + gaming.

For professional creators too, things like gaming + blender rendering, video encoding etc. and indeed several of these at once.

Being able to do everything on one rig versus needing several could be a huge cost saving. Maybe even things like having a VM running in the background using a dedicated video card for some work task while you game on another.

/rambling thoughts :)