As much as I wanted a Haswell rig, now I'm thinking twice about it.

[SPBHM]

That is only a 14% bump from IB to Haswell. And yes, that's about the best increase they got.

Ivy Bridge 1 year old, and is the same 22nm process...
so I think a gain (over something as good as ivy) as high as 14% without using the new instructions or the new IGP is quite good.

 

[jaqie]

way too much for me to read here. I caught something on page one I thought I would chime in on...

Someone mentioned AMD's current AM3+ cpus as being 8 core and intel's desktop cpus as being 4 core... this is not entirely accurate. Ever since the 486, all cpus have had their own mathco, which IMO (note that there, IMO!) ever since then is required to consider it a full processor. The highest actual cpu count in amd's arsenal for desktop boards is the now defunct phenom II thuban x6. To call AMD's newer AM3+ cpus 8 core IMO is like calling the hyperthreaded 4 core cpus 8 core cpus. its more like 4 cores and 4 half-cores, IMO. for both.

 

[Shmee]

I am wondering how well Haswell will OC...

 

[frozentundra123456]

way too much for me to read here. I caught something on page one I thought I would chime in on...

Someone mentioned AMD's current AM3+ cpus as being 8 core and intel's desktop cpus as being 4 core... this is not entirely accurate. Ever since the 486, all cpus have had their own mathco, which IMO (note that there, IMO!) ever since then is required to consider it a full processor. The highest actual cpu count in amd's arsenal for desktop boards is the now defunct phenom II thuban x6. To call AMD's newer AM3+ cpus 8 core IMO is like calling the hyperthreaded 4 core cpus 8 core cpus. its more like 4 cores and 4 half-cores, IMO. for both.

That is what amd calls them. Really though does it matter: the performance, power usage, and price are what really matter. They do advertise it as an 8 core cpu though.

 

[Atreidin]

way too much for me to read here. I caught something on page one I thought I would chime in on...

Someone mentioned AMD's current AM3+ cpus as being 8 core and intel's desktop cpus as being 4 core... this is not entirely accurate. Ever since the 486, all cpus have had their own mathco, which IMO (note that there, IMO!) ever since then is required to consider it a full processor. The highest actual cpu count in amd's arsenal for desktop boards is the now defunct phenom II thuban x6. To call AMD's newer AM3+ cpus 8 core IMO is like calling the hyperthreaded 4 core cpus 8 core cpus. its more like 4 cores and 4 half-cores, IMO. for both.

Everyone has their own arbitrary definition of what a "core" is, and some people have stronger responses to their definition being contradicted than if somebody slapped their mother. Let's not go there again.

 

[UaVaj]

way too much for me to read here. I caught something on page one I thought I would chime in on...

Someone mentioned AMD's current AM3+ cpus as being 8 core and intel's desktop cpus as being 4 core... this is not entirely accurate. Ever since the 486, all cpus have had their own mathco, which IMO (note that there, IMO!) ever since then is required to consider it a full processor. The highest actual cpu count in amd's arsenal for desktop boards is the now defunct phenom II thuban x6. To call AMD's newer AM3+ cpus 8 core IMO is like calling the hyperthreaded 4 core cpus 8 core cpus. its more like 4 cores and 4 half-cores, IMO. for both.

the way AMD advertised it. sounds like 8 physical core. thanks for the clarification.

 

[Idontcare]

the way AMD advertised it. sounds like 8 physical core. thanks for the clarification.

It is 8 physical cores. But no two cores are the same.

You wouldn't expect an AMD 386 core from 1999 to be comparable to a core in Phenom II in 2010, would you?

They both have cores, but the capability of the cores are radically different. You wouldn't use the 386 core to define today's core, nor would you use today's core to define the 386 core.

"a 386 cpu did not have a full core, it was more like 1/4 or 1/3 of a full core"

^ no one would make that kind of comparison. Both products had full-fledged cores. Just the cores had different capabilities.

So why would you expect/assume an AMD core today would be comparable to what Intel fields as their core? Cores are not comparable.

Does my FX8350 have 8cores? Yes, absolutely.

Are those cores the cores in my 3770k? No, not at all.
Are they the same as the cores in my Athlon K7? No.
The same as the cores in my Northwood P2.4C? Nope.
How about the cores in my P3-800MHz laptop? No again, not the same.

However, the metrics that are comparable across all those CPU products are IPC (single and multi-threaded, how much work does the core get done per clock cycle), clockspeed (determines performance when IPC is factored), price, and power-consumption.

Those four factors alone determine the ROI you get for your CPU purchase.

Core count is about as relevant as transistor count. Neato fact to know but doesn't mean jack unless I know the other four things I bolded above.

 

[inf64]

I' m still sticking with my opinion that although FX does indeed has 8 integer cores, correct way of marketing it would have been "8 thread processor". The penalty of sharing certain units(like FP,decoder,I cache) is making it lose some performance which in turn hurt AMD's brand image when some reviews were pointing this out. AMD knew this would happen and they should have preemptively marketed it differently. Sure, SR core may finally provide that full FP performance of 2 (whatever core) in on FlexFP unit, but 2 years have passed now and they cannot market SR based APUs differently when they launch. Good news is that 6 "core" Kaveri may finally perform like a "real" 6 core chip in almost all workloads,which is different from what BD or even PD does today(they do excel in certain workloads but fall short on others).

 

[Homeles]

Decode yes, execution no. So please refrain from using IPC when its not the case.

How about you refrain from speaking on the subject? You're saying ridiculous stuff like this:

Its simply 30% faster decode on the frontend.

Had you actually read about Steamroller, you'd know that the decode width is being doubled. Since you are clearly uninformed on the changes that are coming with Steamroller and also with AMD's claims regarding it, you have no business discussing it.

 

[axium]

I am wondering how well Haswell will OC...

I think one of the intel engineers said we're in for a very nice surprise when it comes to OC.

After seeing the Haswell performance preview, I think I'll hang on to the ivy for a little while longer.

 

[VirtualLarry]

I think one of the intel engineers said we're in for a very nice surprise when it comes to OC.

After seeing the Haswell performance preview, I think I'll hang on to the ivy for a little while longer.

But is it a good or bad surprise? I heard something mentioned somewhere that the ring bus/L3 cache on Haswell was decoupled from core clock. So maybe the cores will OC to 5-5.5, but the L3 cache and some of the uncore will remain at a lower clock speed, which will put it at the situation that AMD chips are in.

 

[Homeles]

The "surprise" has been taken to mean the return of BCLK overclocking.

 

[Shmee]

Interesting, so possibly the k versions are no longer needed?

 

[MisterMac]

Wasn't there more leaks of SB by now?


How come it's leak season - and there's like no leaks.


There's very few stating modest lower end scale improvements comparing to certain intel spokes people (Well lower end to my expectations anyway, intel gotta deliver with all those resources - no room for error).


Like i said - stupid tree huggin hippies are ruining my desktop experience.
Mobile -_-

 

[ShintaiDK]

Wasn't there more leaks of SB by now?


How come it's leak season - and there's like no leaks.


There's very few stating modest lower end scale improvements comparing to certain intel spokes people (Well lower end to my expectations anyway, intel gotta deliver with all those resources - no room for error).


Like i said - stupid tree huggin hippies are ruining my desktop experience.
Mobile -_-

There was a huge bust in Taiwan with selling ES samples. That might have put a lid on the jar.

 

[WhoBeDaPlaya]

But is it a good or bad surprise? I heard something mentioned somewhere that the ring bus/L3 cache on Haswell was decoupled from core clock. So maybe the cores will OC to 5-5.5, but the L3 cache and some of the uncore will remain at a lower clock speed, which will put it at the situation that AMD chips are in.

As long as it'll hit 5GHz somewhat easily, I'll get Haswell. My magic number is a 6GHz Nehalem equivalent

 

[jaqie]

. . . . Core count is about as relevant as transistor count. Neato fact to know but doesn't mean jack unless I know the other four things I bolded above.

I actually disagree with this single point you made - the rest is quite accurate IMO.

Core count and core capability matters simply because of the way programs are made - this is why people talk about per-core execution speed... it matters in today's programs. Otherwise what you said would be completely true, and I truly look forward to the day when that is true, but until then, core count and individual core capabilities does matter simply because of the inadequacy (IMO) of how programs have been and are made... things are getting better but we aren't there yet.