[WCCF] CPU-Z shows AMD Carrizo APU A10-8890K

[Linus_torvalds]

yeah indeed. maybe i was kindoff wrong stating some stuff, or should atleast explain why it is so.. but mostly, i cant stand seeing people here spread misinformation so much. sometimes i ignore it, sometimes i try to fix their POV, mostly i just fail, because i never have enough time all together to write up a good post on it

Now you are genius right? You were stating that it's 2x performance boost! LOL

[dalekphalm]

Now you are genius right? You were stating that it's 2x performance boost! LOL

Actually @LukaP admitted that he didn't know everything about the topic (HT vs APU Modules). But that doesn't take away from the fact that you were also saying incorrect information. So yeah, you don't really have any foundation to stand on while "insulting" him with this post. Let's please keep it civil!

 

Even if this article is total BS and a fake, as per several posters earlier in the thread, we'll just have to wait and see what AMD actually comes out with... Though a 6-Core APU with R7 level graphics would be pretty decent indeed. In theory, it would be great as an HTPC or a very low budget Gaming PC build.

[LukaP]

Now you are genius right? You were stating that it's 2x performance boost! LOL

it is. if you have a thread that utilises memory a lot (rendering) then you get almost perfect 2x scaling. and that is in real world scenarios. in theory, having a thread that for every exectuted set, takes the same amount of time to fetch from ram, then you get a 2x boost. 

 

now stop trolling or being stubborn or whatever the heck you are trying to do...

[Linus_torvalds]

Actually @LukaP admitted that he didn't know everything about the topic (HT vs APU Modules). But that doesn't take away from the fact that you were also saying incorrect information. So yeah, you don't really have any foundation to stand on while "insulting" him with this post. Let's please keep it civil!

 

Even if this article is total BS and a fake, as per several posters earlier in the thread, we'll just have to wait and see what AMD actually comes out with... Though a 6-Core APU with R7 level graphics would be pretty decent indeed. In theory, it would be great as an HTPC or a very low budget Gaming PC build.

He admitted but he is still saying that it's 2x boost. fuck logic!

 

actually if programmers will make software that supports AMD HSA technology it would be great. There are 5 programs on Linux that support HSA. In that programs A10 7850k is much faster then i5 4670k because of 4 CPU core and 8 GPU cores

[Faa]

I said you do not know HT well.

firstly it's not even theoritical 2x boost. 

secondly second task can be executed exectly when first task is executed. they work independently from each other on one core. That's main thing about HT. It makes core work simultaniously on 2 tasks. no one is waiting. 

You described what AMD did in their buldozer modules. each module has 2 cores, but when one executes task second is waiting. they have one source. It's like 2 line way and bridge,  wich only one car can pass through on 

It's true what he said. See below why.

 

Let's take the Bulldozer module and look at it:

It contains 2 integer cores and one floating point core.

No. They have one Flexfpu which can split up in 2x 128 bit FP's to process 2 threads (FP calcs) at the same time. "Integer cores", you pulled that from their diagram an ALU cluster isn't a core at all. They only have 2 ALU's per "cluster" so 4 per module, Haswell has 4 alu's per core. A core is a Single CPU on its own with the memory controller taken out and some things that can be shared like L3 etc.

 

Some people take this (like yourself), as meaning it's "fake" dual core. However, that's not entirely correct. The technology is different enough that you can't compare it directly with an Intel "Core".

Like I said the ALU clusters aren't cores, both alu clusters share the front/back-end. A true dual core has 2 front-ends, not 1. FPU's aren't a requirement for x86, they came later out for x86 so they don't make a cpu a fake x core.

 

Now, why are there half the amount of FP cores? 

 

If it only had 4 FPU's, it wouldn't be able to process 8 threads simulanteously. You clearly see 8 blocks being rendered in Cinebench at the same time with a 8350 or 6 with a 6300 or 4 with a 4300. 

 

Because many computer applications don't use them! Secondly, Graphics Cores are MUCH MUCH MUCH MUCH (did I say much?) better at performing FP calculations. This is why you have things like OpenCL and CUDA to take advantage of GPU Compute.

 

Yeah, you might as well run an OS complety of the GPU >.< Gaming is nothing more besides FPU dependent, rendering as well.

 

Now as @LukaP has mentioned, AMD took this route on purpose. They could have VERY EASILY went the HT route (Obviously they'd have to call it something else, but technologically they could easily do it). But AMD is looking at the long game, which unfortunately means in the interim, their CPU "Modules" lag behind quite a bit.

 

They've done this with CMT or in other words with bulldozer. CMT or SMT can be in co with CMP or a single core. See below what HT is and you'd get a better picture of CMT

 

Also, HT does NOT do calculations simultaneously. That would be a factual lie. HT works on the principle that pretty much any calculation a CPU Core does won't fully utilize that core. It also works on the principle that once a Core has done an instruction, it is often waiting for *something* else to complete. Be that accessing the RAM, getting a result from another Core, accessing a different piece of system Hardware, etc. During these little tiny gaps, which by themselves are very small - but extremely frequent - the CPU can leverage all those combined little bits and equal a large performance boost by using the power as the Core is idle waiting for another action.

 

Complety wrong, complety. Lets say you have a single core with no HT - If you have two threads; one of it doing integer instr and the other one doing fp instr -> the first instr would be pushed in the ALU which will take a cycle time and the 2nd instr will be in the FPU takes another cycle so in total 2 cycles. So the execution resources weren't processing at the same time. What HT would do is (you need a 2nd thread for it or byebye) allowing both instructions being processed at the same time in a single cycle which means simulateneously multithreading. So you have a 100% gain so @LukaP was right about this. You can do this as well with a plain dual core (no HT) so push the integer thread to core#0 and the 2nd floating thread to #core1 that would process at the same time.

AMD tried to find an alternative to HT with duplicating their resources (CMT) but it failed so badly.

Another word for HT is SMT

 

Hyperthreading does NOT automagically make a single core able to perform two tasks out of thin air. That's not how it works. You obviously could use some additional reading on the subject, so I suggest that you check out the wiki on HT:

 

Cpu's can't execute a single thread on more than 1 core/thread. For a single core CPU with HT you need 2 threads to take advantage of SMT, or 4 threads for 2 cores & SMT or 8 threads for 4 cores & SMT. No SMT; For a true dual core you need 2, for a true quadcore you need 4, for a true octacore you need 8 and so on. SMT is made to make the cpu perform more tasks at the same time.

 

With this in mind, this is a very very simplified explanation, and there are many technical bits missing from it.

Everything in your post was complety wrong.

[DildorTheDecent]

Got so much salt my heart has failed. 

But seriously mini hex core sounds cool. 

[Linus_torvalds]

Complety wrong, complety. Lets say you have a single core with no HT - If you have two threads; one of it doing integer instr and the other one doing fp instr -> the first instr would be pushed in the ALU which will take a cycle time and the 2nd instr will be in the FPU takes another cycle so in total 2 cycles. So the execution resources weren't processing at the same time. What HT would do is (you need a 2nd thread for it or byebye) allowing both instructions being processed at the same time in a single cycle which means simulateneously multithreading. So you have a 100% gain so @LukaP was right about this. You can do this as well with a plain dual core (no HT) so push the integer thread to core#0 and the 2nd floating thread to #core1 that would process at the same time.

hey i said that HT works simultaniously and they said to me that i was wrong! so i was right about it!

 

i said this

secondly second task can be executed exectly when first task is executed. they work independently from each other on one core. That's main thing about HT. It makes core work simultaniously on 2 tasks. no one is waiting. 

  LukaP

said this

i know its not a 2x boost... its a theoretical 2x boost if correct code is written. and the second task, when its being executed is executed at the same speed, its just that its only executed when the first task is waiting...

[patrickjp93]

No one bothered to think, did they? When you increase core counts, clock rates do not, I repeat: DO NOT jump by .7 GHz. They go down, or at best remain the same with a higher TDP.

 

If anything this would have been like the FX 9590 being a binned 8350 but with APUs and a slight reworking to add cores.

 

This just smelled wrong from the stats on down.

[Faa]

hey i said that HT works simultaniously and they said to me that i was wrong! so i was right about it!

 

i said this

  LukaP

said this

Yeah you're right in that, there wouldn't be that much of a point bringing a technology like HT out if it doesn't offer simult. multithreading >.< If HT cant allow you to process 2 threads at the same time then there's no way an i7 can outperform an i5.

[LukaP]

Yeah you're right in that, there wouldn't be that much of a point bringing a technology like HT out if it doesn't offer simult. multithreading >.< If HT cant allow you to process 2 threads at the same time then there's no way an i7 can outperform an i5.

its not really simult. there is no way one ALU can do 2 things at once. but every thread needs to wait every once in a while for stuff that is not cached but is in ram. and thats when the secondary thread comes in. otherwise, all youre saying is correct

[Faa]

its not really simult. there is no way one ALU can do 2 things at once. but every thread needs to wait every once in a while for stuff that is not cached but is in ram. and thats when the secondary thread comes in. otherwise, all youre saying is correct

Haswell has 4 ALU's per core meaning they can process 4 integer instructions a cycle at the same time, they don't have one ALU. AMD can do 2 instructions per cycle per core or 4 per module. Forget FPU's, if they only have 1 ALU per core there wouldn't be any point bringing HT out other than for taskmanager sexappeal. All thats being done is doubling the instruction pipeline, lapic and registers to achieve SMT.

 

Also storing it in RAM is the last part of your process, we are talking about the steps before it. Fetch > decode > execute > store.