New consoles coming with low-clocked AMD x86. Can we now save moneys on our PC CPUs?

[DuckThor Evil]

No.
The Core i7 offers hyper threading, which, to put it in simple terms, is a way of increasing utilization of the 4 physical cores in highly threaded situations. (Or put another way, a means to reduce the impact of latency.) It is not equivalent to having 8 physical processors. At best it will be some small fraction faster than its non-hyperthreading siblings. Typically, as now, the differences will be negligeable.

What? Hyperthreading offers a huge boost in many applications, games haven't really been programmed to take advantage of it for obvious reasons, but it's not hard to find various tests on real applications where the boost is large.

 

[Exophase]

No, but if I am doing intensive CPU tasks, with high priority or real time set I have no luck doing anything else with it either.

I don't get your point.

 

[Entropy]

What? Hyperthreading offers a huge boost in many applications, games haven't really been programmed to take advantage of it for obvious reasons, but it's not hard to find various tests on real applications where the boost is large.

Up until quite lately, when averaged over a large number of applications, the overall performance dropped with hyperthreading enabled.

The situation has improved, but the occasional exception to modest performance increases, is just that. An exception. Intel, of course, endorses the use of such examples in benchmarking.

 

[DuckThor Evil]

Up until quite lately, when averaged over a large number of applications, the overall performance dropped with hyperthreading enabled.

The situation has improved, but the occasional exception to modest performance increases, is just that. An exception. Intel, of course, endorses the use of such examples in benchmarking.

It's not hard to believe that if these large number of applications have mostly been using only one or few threads that this has in fact been the case, but applications that have been using more threads have seen large boost. Next gen consoles having 8 threads at least makes the question valid imo.

There are few examples. In my opinion, if your original statement was true, there shouldn't be any test showing good performance increase.

http://www.techpowerup.com/reviews/Intel/Core_i5_3570K_and_i7_3770K_Comparison/5.html

 

[Exophase]

Up until quite lately, when averaged over a large number of applications, the overall performance dropped with hyperthreading enabled.

Can you show evidence of that having ever been the case, much less if restricted to HT post-Prescott (Nehalem and beyond, and Atom if you wish)?

 

[Davros]

but applications that have been using more threads have seen large boost.

And to prove that point your showing benchmarks that show a small boost, I'm confused !!!

 

[Exophase]

And to prove that point your showing benchmarks that show a small boost, I'm confused !!!

He showed benchmarks where the HT enabled (+ 2MB more L3) processor beat the non-HT enabled version clocked nearly 30% higher. You really consider 30% a small deal?

 

[Albuquerque]

Ignoring synthetic benchmarks, I can also personally attest to Skyrim getting a measurable boost from SMT (hyperthreading) when running at lower clockspeeds.

I can directly show a single core Intel Sandy Bridge processor at 1.5GHz picking up a ~55% increase in performance by enabling HT, and can show the same dual core SB picking up another ~35% increase by enabling HT. At higher clockspeeds (in this example, 3GHz), Skyrim's gains in HT are a bit lower. Enabling HT on a single core picks up around 33%, and enabling HT on a dual core picks up another 25%.

Granted, these were contrived examples, but they're easily replicated: http://forum.beyond3d.com/showpost.php?p=1636962&postcount=56

 

[Blazkowicz]

Even an Atom can end up matching or exceeding a Bobcat with that HT.
(with friendly applications such as media encoding)

 

[itsmydamnation]

Even an Atom can end up matching or exceeding a Bobcat with that HT.
(with friendly applications such as media encoding)

bobcat and atom have very similar amount of execution units per core, so it makes perfect sense that they both can get somewhere around the same throughput per core.

Personally I find HT on my nehalem I7 a whole lot of meh, I turn it off and clock my cores 300mhz faster.

 

[Ethatron]

... when averaged over a large number of applications, the overall performance dropped with hyperthreading enabled.

I read this as taking 500 benchmarks with/without HT and averaging the delta. It sounds plausible.

 

[sir doris]

Error

 

[Albuquerque]

I read this as taking 500 benchmarks with/without HT and averaging the delta. It sounds plausible.

It was true in the Netburst years; I would very much challenge this statement with any of the Core i* series processors.

 

[homerdog]

I have read that HT can cause problems with uneven frame delivery in some games. Sorry I can't remember the source but it seemed legit, although written in chinese I believe.

 

[Albuquerque]

I have read that HT can cause problems with uneven frame delivery in some games. Sorry I can't remember the source but it seemed legit, although written in chinese I believe.

I've been through a few of those tests myself using my 3930k as a testbed. The culprit in my case was core parking, which can be exacerbated by the neurotic behavior of Windows kernel default thread affinity behavior rotating through cores. The link to hyperthreading was tenuous at best...

 

[Pete]

homerdog, you're thinking of this thread (Albuquerque explores HT vs core parking on page 8).

 

[Albuquerque]

It wasn't my intention to say HT is completely devoid of fault, however I still firmly believe the link between microstutter and hyperthreading is "iffy" at best. The highest performing CPU's (an i7-3770k) demonstrates microstutter with hyperthreading enabled, but the CPU is never truly fully utilized. One of the lowest performing CPU's (an i3-2100) does NOT demonstrate microstutter with hyperthreading enabled, and the CPU stays at nearly 100% utilization the whole time.

<hypothesis>
The key construct here seems to be keeping the CPU busy. If it isn't busy, the cores get parked, and then have to "wake up" when the wonky Windows kernel scheduler rolls a perfectly well executing thread onto an unsuspecting parked core. HT amplifies Windows bad behavior because the CPU presents enough additional computing resource that, in the interim, other cores have enough idle time to get parked again.
</hypothesis>

Turning off core parking on my i7-3930k immediately solved the issue; someone else reported their i7-3770k responded in the same way with core parking disabled.

 

[Grall]

<hypothesis>HT amplifies Windows bad behavior because the CPU presents enough additional computing resource that, in the interim, other cores have enough idle time to get parked again.</hypothesis>

I totally buy that explanation. It's unbelievable that the windows scheduler is uselessly juggling tasks between cores the way it is, when it's well known to be detrimental to both performance and power useage (which has been becoming increasingly important for years, and in fact was important long before win7 even launched.)

 

[Blazkowicz]

I didn't know core parking existed on mainstream CPUs already, I thought it was cell phone stuff and future CPUs (Jaguar, Haswell, next-gen Atom)

 

[Albuquerque]

A tiny bit of Google suggests (I didn't dig deep) core parking came with the Nehalem architecture on the Intel side. AMD has supported it for at least as long as the last set of FX processors, however Microsoft purposefully sent out a patch to turn it off for most AMD processors for (ha, who could guess?) performance reasons.