Real cores vs HT in future engines

[KKRT]

Hi,
I've got a question about future PC gaming and engines. I know that every new engine is wrote with massive parallelization in mind, like Frostbite or CryEngine 3.
So what will be better for those engines in term of 'future proof' processor - i5, which is 2 cores+HT that give us 4 threads, or quad core/old version of i5 without HT/x4?

We all know that for now in most application i5 is faster than quad core, but will it change if engine will not virtually but psyhically use all 4 cores? HyperThreading as we know just splits one core for 2 virtual ones and manage core ressources for those 2 threads, in applications that dont use all cpu cycles its quite efficient, but if application really needs 4 real threads at one moment it should stall, right?

And what will be better Phenom X6 [6 real cores] vs i7 [4 cores, 8 threads] in those engines?
Do greater cache memory really matters in those game engines calculations?

 

[ShaidarHaran]

You're making comparisons across CPU architectures that can't properly be made. In general, physical cores are better than virtual cores, but it depends on the architecture. Right now Intel's multi-threading implementation is vastly superior to AMD's.

 

[KKRT]

You're making comparisons across CPU architectures that can't properly be made.

Ok, so just compare i5 with HT and without HT. Simple 4 cores vs 2 cores with 4 threads.
I just want to know if hyperthreading can be a bottleneck in future games, thats all ;] I know thats a question to limited audience like Repi, but i think its quite interesting.

 

[fellix]

Right now Intel's multi-threading implementation is vastly superior to AMD's.

Care to elaborate on any AMD SMT tech in existence?

 

[Silent_Buddha]

So what will be better for those engines in term of 'future proof' processor - i5, which is 2 cores+HT that give us 4 threads, or quad core/old version of i5 without HT/x4?

Short answer...

Lynnsfield with 4 physical cores and no HT (i5-750) will be better than Clarksdale with 2 cores and HT in almost all multithreaded situations.

Assuming equal CPU speed (both turbo and normal). Faster speed (for Clarksdale) will compensate somewhat for the natural advantage of 4 physical cores depending on how heavy the threading load is.

That said, if you are on a budget, the Clarksdale chips aren't bad at all.

Regards,
SB

 

[Davros]

I saw some benchmarks recently (cant remember the link)
and they panned out sort of like this

imagine
2 cores = 50
4 cores = 100
2 cores + 2 ht cores = 75

 

[sunscar]

^^^^

And I'd say for the i5 comparison it should be pretty easy to figure. We have two actual cores divided up to act like four, or we have four actual cores. Which sounds better? It's a gross over-simplification, but for this particular scenario, it's kinda fitting.

 

[Albuquerque]

I saw some benchmarks recently (cant remember the link)
and they panned out sort of like this

imagine
2 cores = 50
4 cores = 100
2 cores + 2 ht cores = 75

I think it really depends on the type of work you're doing. Just like anything though, right? You're probably about right across all apps evAr.

Also consider power usage; it's very likely that a two-core processor with HT will be more power efficient (ops per sec per watt) than a true four core would be. Remember that Intel's new stated strategy is that CPU performance must increase super-linearly with power consumption.

 

[Davros]

It was a gaming benchmark iirc

 

[FrameBuffer]

I would imagine that with the court mandated Intel Compiler update AMD solutions would gain in performance (iirc suggested was about 10-15% on avg, 20% occasional) where as Intel would stay the same.

 

[ShaidarHaran]

Care to elaborate on any AMD SMT tech in existence?

I didn't say SMT And CMT is right around the corner.

 

[ShaidarHaran]

I would imagine that with the court mandated Intel Compiler update AMD solutions would gain in performance (iirc suggested was about 10-15% on avg, 20% occasional) where as Intel would stay the same.

Wait, what? Last I heard AMD & Intel settled their recent lawsuit out of court so how can there be a court mandate for Intel to update their compiler to enable greater performance on AMD processors? I know it's just an ID check that blocks optimizations for all but GENUINE_INTEL procs and therefore quite lame, but where did they agree to stop doing that?

 

[3dcgi]

Wait, what? Last I heard AMD & Intel settled their recent lawsuit out of court so how can there be a court mandate for Intel to update their compiler to enable greater performance on AMD processors? I know it's just an ID check that blocks optimizations for all but GENUINE_INTEL procs and therefore quite lame, but where did they agree to stop doing that?

I remember it being part of the agreement though I don't have a link.

 

[Silent_Buddha]

Wait, what? Last I heard AMD & Intel settled their recent lawsuit out of court so how can there be a court mandate for Intel to update their compiler to enable greater performance on AMD processors? I know it's just an ID check that blocks optimizations for all but GENUINE_INTEL procs and therefore quite lame, but where did they agree to stop doing that?

Heh, it also used to block optimizations on older Intel CPU's in order to make newer ones that much more attractive. This was back in the P4 days however, when Intel was desperate to make the P4 look better clock for clock compared to the P3.

Not sure what their compiler does currently.

Regards,
SB

 

[dkanter]

2-way SMT is worth about 10-25% performance boost, so it won't make up for 2 extra cores.

The other tricky part is that a newer Nehalem CPU can hit far higher frequencies thanks to turbo boost.

Overall, you're probably better off with Nehalem rather than a Core2.

David

 

[Silent_Buddha]

His question was a layman's way of asking whether Lynnfield was better than Clarkdale (without knowing he was talking about Lynnfield and Clarkdale).

Thus turbo boost is going to be equal on both. And it comes down to whether i5-750 (only quad core without HT) is better than the other i5 chips. And at that point it'll come down to clockspeed and how heavily the threaded loads are.

In that case an i5-750 at 2.66 GHz versus an i5-660/661 at 3.33 GHz isn't so cut and dry. In heavily threaded apps that put a large load on each core the 750 will still be better probably. But in other instances (gaming for example) the 660/661 at 3.33 GHz would probably be faster.

Used those two as they are closest in price at Newegg. Doesn't change significantly if we go with the i5-650 at 3.2 GHz which is 10 USD cheaper.

Regards,
SB

 

[eastmen]

Gaming isn't as cut and dry either. More games are using multiple cores. Dragon age really took advantage of a third core and with dx 11 coming many games should use even more cores.

 

[hoom]

I wish Intel would do a 32nm quad/triple core i5/7 series at 3.something Ghz.
Big L3, fast power state switching, lots of turbo-boost, no HT.
It'd be a single/light-threading monster.

Clarkdale looks very good in terms of single-thread performance but i5 670 costs about the same as an i7 860 & i5 660 is about same as i5 750

 

[trinibwoy]

They can't do that cause it would cannibalize sales of their more expensive parts which will be slower in most workloads. The big L3 and high clocks of Gulftown are the only things saving it from being outgunned in lightly threaded apps.