AMD FX 8370 vs Intel I7 5960x [GTX 970 SLI 4K benchmarks]

[Dabombinable]

No. Each Vishera core has 2 integer ALUs and 1 half of a shared FPU. Each module has 4 integer ALUs and 1 FPU.

1 Module. 2 ALU.

http://i.imgur.com/NgBdlEE.jpg

^Bulldozer

 

http://pc.watch.impress.co.jp/img/pcw/docs/556/374/21.jpg

^Piledriver

 

http://img694.imageshack.us/img694/1382/img00380211.jpg

^Steamroller

[Majestic]

Do you know how to undo the blatant bias for core 1 and core 2 in DX11?

 

DX12

[patrickjp93]

1 Module. 2 ALU.

http://i.imgur.com/NgBdlEE.jpg

Scheduler is the fetch/decode combined stages, not the actual bit-crunching unit.

[Dabombinable]

Scheduler is the fetch/decode combined stages, not the actual bit-crunching unit.

I edited my post, funnilly enough I can only find architecture diagrams for Vishera's predecessors and successor.

 

Edit: And don't deflect, it's still 2 ALU per module.

[patrickjp93]

Do you know how to undo the blatant bias for core 1 and core 2 in DX11? Because the extreme load put on those two, compared to other cores, no matter what, game it is. It always emphasis's Core 1 and Core 2, with a lower emphasis on Core 3,4,5,6,7,8,9,10 etc...

They way the API is written, is not giving a uniform load on all cores, thus some cores are just partially active (read: light load) or idling.

The uniform load can be achieved by better design. 1 core can handle the DX 11 draw calls. The rest can be getting info from the net about NPC and player positions and animation states, calculating the positions and views, and marshaling the information to be sent to the GPU. It all comes down to design. Seriously go look at my blog for game template code. You can create a dynamically balancing code structure really easy to keep all cores working as hard as possible moment to moment.

[Dabombinable]

Yeah, but thats not the point. This shows that the fx series is not the bottleneck people make it out to be.

Only at 4K where you are GPU bound-which people have known for a long time. Even a Pentium will do well at 4K.

[patrickjp93]

I edited my post, funnilly enough I can only find architecture diagrams for Vishera's predecessors and successor.

Edit: And don't deflect, it's still 2 ALU per module.

2 ALUs per core with a shared FPU per module. I know because Richland APUs built on steamroller have those specs exactly.

[Tomsen]

2 ALUs per core with a shared FPU per module. I know because Richland APUs built on steamroller have those specs exactly.

You are both right.

It have 2 ALU pipelines per "core", and have 2 ALU clusters per module.

 

EDIT: It is simply a game of using the right terminologies.

[Dabombinable]

No. Each Vishera core has 2 integer ALUs and 1 half of a shared FPU. Each module has 4 integer ALUs and 1 FPU.

Ok...then why did you say that^, more specifically each module having 4 ALU and 1 FPU (And again, CMT based CPU can not be described as having cores as if they were standard CPU).

 

2 ALUs per core with a shared FPU per module. I know because Richland APUs built on steamroller have those specs exactly.

 

Edit: (after your edit)

 

2 ALUs per core with a shared FPU per module. I know because Richland APUs built on steamroller have those specs exactly.

That is still 2x ALU and 1 FPU.

[patrickjp93]

Ok...then why did you say that^, more specifically each module having 4 ALU and 1 FPU (And again, CMT based CPU can not be described as having cores as if they were standard CPU).

 

 

Edit: (after your edit)

 

That is still 2x ALU and 1 FPU.

Review my last reply with provided core diagram. There are 2 integer pipelines per core, or 2 integer crunchers per core in the diagram.

[Prysin]

What is not in the benchmarks we are discussing on the OP?

 

DX12

 

 

Oh, thank you, you just said it!!!! DX12 is not there....

 

The uniform load can be achieved by better design. 1 core can handle the DX 11 draw calls. The rest can be getting info from the net about NPC and player positions and animation states, calculating the positions and views, and marshaling the information to be sent to the GPU. It all comes down to design. Seriously go look at my blog for game template code. You can create a dynamically balancing code structure really easy to keep all cores working as hard as possible moment to moment.

Well, perhaps you can. But aslong as this is not done in practice, benchmarks like the one in the OP, while not very logical, is by all probability true. Until DX12 becomes mainstream. In which case, i think perhaps FX might redeem itself in the sub i5 area (raw multitasking grunt will beat out the i3 and older i5s when all cores are properly loaded). But, time will show. Perhaps 2 years down the line, we will reccomend Vishera CPUs for lower end budget systems wanting to play modern games.

 

PFFFFT or NOT.

 

#jumpedship

#FXsucksandiknowit

#i7masterrace

[Hoppa]

I wonder if disabling hyper threading on the intel cpu would give it better results, since games don't fully utilize more than 4 or 6 threads.

And having 8 large cores as opposed to 16 smaller ones may eliminate any potential cpu bottlenecks.

[patrickjp93]

What is not in the benchmarks we are discussing on the OP?

 

 

 

Oh, thank you, you just said it!!!! DX12 is not there....

 

Well, perhaps you can. But aslong as this is not done in practice, benchmarks like the one in the OP, while not very logical, is by all probability true. Until DX12 becomes mainstream. In which case, i think perhaps FX might redeem itself in the sub i5 area (raw multitasking grunt will beat out the i3 and older i5s when all cores are properly loaded). But, time will show. Perhaps 2 years down the line, we will reccomend Vishera CPUs for lower end budget systems wanting to play modern games.

 

PFFFFT or NOT.

 

#jumpedship

#FXsucksandiknowit

#i7masterrace

I had been replying on a cell phone, so here's the link to the template code. It's not hard to build. The toughest part is the DX 11 kernel itself. Setting up the balancing system is stupidly easy for any college junior who's taken an HPC class.

[patrickjp93]

I wonder if disabling hyper threading on the intel cpu would give it better results, since games don't fully utilize more than 4 or 6 threads.

And having 8 large cores as opposed to 16 smaller ones may eliminate any potential cpu bottlenecks.

The only time hyperthreading gets in the way is if you have more threads than physical cores and those threads use the same resources. You fill up the physical cores before the logical ones get involved. I forget if it's 0-based or 1-based indexing, but if you have a quad core I7, you don't get HT in the way until you have 5 threads, and threads 0/4 or 1/5 want to use the same instructions on the same resources.

[Prysin]

I had been replying on a cell phone, so here's the link to the template code. It's not hard to build. The toughest part is the DX 11 kernel itself. Setting up the balancing system is stupidly easy for any college junior who's taken an HPC class.

stupidly easy you say, perhaps. But how complex is it to build onto this system?

How many hours did it take you to build this system?

Was this something YOU built, or something that was halfway done already that you just pieced together and completed?

 

From what i can gather from your signature, you're a student. And no disrespect, but i would think the people who actually make games, do have a bit better understanding of how to optimize and build around code then you might have. After all, its what they do for a living, and DX11 has been out long enough for the game devs to find out how to do so.

So again, no disrespect, but if it was this easy/didnt have any massive caveats, why has it not been done on a large scale?

 

My guess, is the time it takes over not bothering. As simple as that.

Them optimizing their game to hardware wont give them any money at all, it would just cost them money and more time to make the game.

 

That and bug-fixing. How prone is such a system to bugs?

How easy is it to untangle these bugs without setting off a chain reaction?

 

I am by no means any coder or game maker, but i do understand enough to see that there got to be a caveat of your proposed system. If things sounds too good to be true, they usually are.

[Tomsen]

The only time hyperthreading gets in the way is if you have more threads than physical cores and those threads use the same resources. You fill up the physical cores before the logical ones get involved. I forget if it's 0-based or 1-based indexing, but if you have a quad core I7, you don't get HT in the way until you have 5 threads, and threads 0/4 or 1/5 want to use the same instructions on the same resources.

Hyper-threading could get in the way that other software (non-related to the game) might take up execution ressources the game could have used. There are many ways for it to get in the way. That is just how Windows handle hyper-threading (just like AMDs CMT).

It must certainly is not the only way..

// From phone

[Sintezza]

This topic only comes down to 4k being GPU limmited in most cases.

Thats it. .

 

i could get into the discussion about SMT and CMT,

and how both intel and amd architectures work.

But that totaly makes no sense, because its totaly irrelevent to this topic.

[winningsince1337]

AMD is still shit

[patrickjp93]

stupidly easy you say, perhaps. But how complex is it to build onto this system?

How many hours did it take you to build this system?

Was this something YOU built, or something that was halfway done already that you just pieced together and completed?

From what i can gather from your signature, you're a student. And no disrespect, but i would think the people who actually make games, do have a bit better understanding of how to optimize and build around code then you might have. After all, its what they do for a living, and DX11 has been out long enough for the game devs to find out how to do so.

So again, no disrespect, but if it was this easy/didnt have any massive caveats, why has it not been done on a large scale?

My guess, is the time it takes over not bothering. As simple as that.

Them optimizing their game to hardware wont give them any money at all, it would just cost them money and more time to make the game.

That and bug-fixing. How prone is such a system to bugs?

How easy is it to untangle these bugs without setting off a chain reaction?

I am by no means any coder or game maker, but i do understand enough to see that there got to be a caveat of your proposed system. If things sounds too good to be true, they usually are.

It's a system I built, and the key to any system is you build it to be flexible and extensible from the beginning. I'm a student, but I'm a better coder than most people with 3 years of jobs experience. 6000 lines of the Linux kernel right now are mine. 4000 lines of GCC and Clang are mine. 300 lines of virtual box are mine. 80 lines of the open source AMD graphics driver for Linux are mine. The moment you dismiss people around you based on their age or occupation to gauge intelligence, you've lost an opportunity.

I have disassembled and decompiled the code in most games that have come out in recent years, and I can tell you game programmers do not know the first thing about optimizing code. It doesn't help Microsoft's C/C++ compiler is way behind GCC and Clang in optimization of non-trivial code. It may sound arrogant, but I can program circles around most game programmers because most of them are college dropouts or graduated from a non computer science degree.

I have experience both theoretical and practical building parallel code. The reason it doesn't happen on a large scale is because consumer developers are lazy about learning new concepts or programming standards. That, and sales teams want the smallest executable with the widest support in the least amount of development time. Quality is an afterthought, and too much is considered a detriment to future sales. Multithreaded asynchronous coding is not remotely difficult. You just have to be willing to learn it. I also have an HPC for dummies blog on here which you may be interested in. It's filled with optimization concepts.

[patrickjp93]

Hyper-threading could get in the way that other software (non-related to the game) might take up execution ressources the game could have used. There are many ways for it to get in the way. That is just how Windows handle hyper-threading (just like AMDs CMT).

It must certainly is not the only way..

// From phone

All those background services are incredibly light weight and won't do much.

[Prysin]

It's a system I built, and the key to any system is you build it to be flexible and extensible from the beginning. I'm a student, but I'm a better coder than most people with 3 years of jobs experience.

Ok, life lesson my friend.

 

That attitude, is what will get you in troubble with your peers. I tried it, doesnt work. What work you think you've done, anyone with 3 years of work experience, would have done that or way more. I know it may seem like everyone just sits at their asses, but assuming you know so much more then people who actually are working. That is what gets you labled as "cocky new kid" pretty darn fast.

 

I am sure you know your stuff. But since you appear to be a student, a word of advice from someone who have gone through school and have been working for a few years... We got a apprentice in our company, he is REALLY good, but he has that attitude of "i know so much, and i know i'm good"... there is A LOT of small corrections and ways of doing our work that he doesnt know, or doesnt realize. Mind you, i work as an electrician, so my type of work is more on the physical side of things. But never think the work you have done compares to what those who have been working for even a few years...

 

In comparison... Before i started working full time some 4-5 years ago, i thought that the 10 miles of cables i had laid down was A LOT.... and that a 50sqmm cable was large... Lets just say, that is nothing. I have laid like 100s of miles of cables within just those 4 years and cable sizes of 240sqmm isnt uncommon. It is rather common... When you think something is alot before starting to work, it really is not...

 

Our apprentice learnt that, when we handed him 3 reels of 95 meters of cable wheighing a modest 22 pounds pr meter..... he got a bit more humble after that.

[Tomsen]

All those background services are incredibly light weight and won't do much.

You are limiting it to background services..

It is still take up execution ports, it could also have influence in regards to the different data structures (IE limiting the amount of entries the thread can have in the different queues/windows, register files, cache, and so on). It certainly is not as big as an issue as it used to be, as Intel have improved hyper-threading over generation, but some might experience worse performance when having hyper-threading enabled (The older the "version" of hyper-threading, the more apparent the problem is).

Haswells hyper-threading should not really have such a big penalty on this.

[QueenDemetria]

 

Intel vs amd on LTT.

+1 for Demi gif.

[fatfinguh]

off: Sry...had to do it.

 

on: You're right though. We're talking 4K here, even if it is irrelevant. Nobody with a budget for 4K will ever consider an AMD CPU(i mean those who buy 2015).

 

That's what this article is meant to question. If direct X 12 will ultimately result in more draw calls to the FX series line, and 4k gaming is not fully dependent on CPUs, then why not? Will DX 12 close the gap between FX and the 4790k? According to AMD, it will. At the same time, the 5690x will get a huge bump as well. Linus did a video in December about 4K gaming with and AMD rig and an R285. Based on the most popular games at the time, it wasn't beyond reach.

[samcool55]

I'm really wondering how things will go with cpu's because of DX12.

I mean i3's and i5's are fast enough for most gpu's with Dx11, and with dx12 probably getting mainstream quicky, why would gamers even upgrade? 

Unless games get a LOT more cpu intensive, there won't be any need for a better cpu for years, most people's upgrade cycles were getting longer and longer, and this will probably make them even longer.

 

Dx12 is a fantastic piece of tech, and it's needed to make the games of tomorrow better. But in the long term it will have a bad financial effect, and that might be quite big. Don't forget that the expensive cpu's are the ones where intel and AMD make big money with. They can't make more profit on an i3 than an i5.