AMD CPU speculation... and expert conjecture

[sarinaide]

I prefer "hasbeen"

 

[juanrga]

Never heard that Haswell will work better on Linux, but Steamroller compiler optimizations were already released: march=bdver3

Well, Haswell CPU and GPU code was being added to GCC from about late 2011. I dont know what "released" means.

march=core-avx-i-2 IIRC for Haswell.

"Released" looks as correct english for me. I did mean something as

AMD has released the code optimization for Steamroller.

http://www.overclock.net/t/1317332/phoronix-amd-drops-steamroller-bdver3-compiler-support/10#post_18542944

and the same when

as various compiler optimizations and CPU driver fixes were released

for Bulldozer

http://en.wikipedia.org/wiki/Bulldozer_%28microarchitecture%29

 

[juanrga]

I prefer "hasbeen"

I like "hasbeen". I still prefer "failwell" because resembles the typical "faildozer" that you could read in so many forums when AMD released Bulldozer.

 

[mayankleoboy1]

AMD has released the code optimization for Steamroller.

http://www.overclock.net/t/1317332/phoronix-amd-drops-steamroller-bdver3-compiler-support/10#post_18542944
and the same when

I myself posted here the original Phoronix article the day it was released.

Never heard that Haswell will work better on Linux

Well i said this because generally in linux world, programs are compiled wiht -O2 and something like -mtune= SSE4.1 or something, which means the compiler produces a generic code, as well as specialised code which is executed at runtime depending on the CPUID of the processor.
Whereas in Windows world, we get an generic binary that will run on all x86 hardware, with IIRC atleast SSE1 support. So you could say that GCC builds are faster.
But then again, some people also say that MSVC has better optimisations than GCC. But then again, a great many windows devs use outdated MSVC versions that dont have all the improvements.
So its a mixed bag of assumptions.

I wonder what happened to a few people that were speculating haswell at 20% ... lol

Reality killed the wishful thinking. I would have loved that Failwell did improve so much, not because i love intel, but CPU perf has been stagnant-ish for some time now.

I wonder if people who say SR has 20% improvement will cry later

 

[JAYDEEJOHN]

http://diybbs.zol.com.cn/11/11_106489.html
This been posted?

 

[Cazalan]

I prefer "hasbeen"

I like "hasbeen". I still prefer "failwell" because resembles the typical "faildozer" that you could read in so many forums when AMD released Bulldozer.

Haswell isn't anything remotely close to the Bulldozer fail.

Minor IPC gain, large graphics gain, and better power savings is what they were after. It has little benefit for desktop enthusiasts but the laptops will love it.

 

[Cazalan]

So, what have I missed the last 3 months?

Nothing. Lots of announcements but no hardware on the shelves.

 

[de5_Roy]

http://diybbs.zol.com.cn/11/11_106489.html
This been posted?

haven't seen it yet. but looks quite promising. i didn't understand most of the parts but it looks like (what google translate gave) -
excavator test chip, 4 ghz.
4 modules, 8 'cores', 4 gcn compute units.
28nm process...?
cus are directly connected to the modules.
(very) high floating point performance. opposite of bd, iirc.
Speculated that multi-threading technology (Speculative Multithreading, SpMT) is developed speculatively execute multiple threads to thread-level parallelism, superscalar processor performance. By additional hardware units, such as thread synchronization unit (Thread Synchronization Unit, TSU), the thread context Sheet (Thread Context Table TCT) and thread memory history table (Thread Memory History, TMH), extends the transactional memory system to improve instruction set based on the wave scalar system the structure (the Wave Sealar ISA) to achieve WaveCache simulator's performance. It also proposed a new two thread-level transaction commit mechanism.

 

[mayankleoboy1]

http://diybbs.zol.com.cn/11/11_106489.html
This been posted?

Saw it on S|A forum.
From what i understand, it is something like running FP math on the GPU, so there is no dedicated FP hardware on the CPU core. I dont pretend to understand if something like this can even be done while remaining compatible with existing x86 code.

 

[mayankleoboy1]

So, what have I missed the last 3 months?

You missed some lovely arguments between fanbois :lol:

 

[mayankleoboy1]

Minor IPC gain, large graphics gain, and better power savings is what they were after. It has little benefit for desktop enthusiasts but the laptops will love it.

I wonder where the hell Anand got those "around 10-15% improvement" numbers from. Straight from intel PR ?

 

[dogman_1234]

So, what have I missed the last 3 months?

Nothing. Lots of announcements but no hardware on the shelves.

Oh...okay.

*grumbles to self about AMD*

So, what have I missed the last 3 months?

You missed some lovely arguments between fanbois :lol:

Yay!

/sarcasm

 

[Cazalan]

Minor IPC gain, large graphics gain, and better power savings is what they were after. It has little benefit for desktop enthusiasts but the laptops will love it.

I wonder where the hell Anand got those "around 10-15% improvement" numbers from. Straight from intel PR ?

"You can expect CPU performance to increase by around 5 - 15% at the same clock speed as Ivy Bridge. "

Wide range. The preview was very clear the major benefits will be from AVX2 and GT3.

 

[The Q6660 Inside]

I prefer HasBurst™, the improvement for a tock is just disappointing compared to Nehalam/Gulftown->Sandy Bridge.

 

[Cazalan]

I prefer HasBurst™, the improvement for a tock is just disappointing compared to Nehalam/Gulftown->Sandy Bridge.

Nahalem -> Sandy would be a tick and a tock

 

[JAYDEEJOHN]

About link, it may be 20nm, and yea, not sure how its going to be implemented for fp, currently, latency will kill it, but it looks theyre going down the same path as Intel, just a lil slower in widening.
I saw some more numbers for Haswell, and mostly its slightly faster, leaving room for catch up to AMD, even at less than what some think AMDs speedups will be, they will be closer.
The next nodes or so, this will be interesting, by then, power numbers may start to become good enough, or, over 8 hrs.
By interesting, will Intel and AMD then go for faster, will they go wider (more cores).
Or, will it all come crashing down, and just shrinks occur, lessening power, lessening silicon while improving margins........

 

[The Q6660 Inside]

I prefer HasBurst™, the improvement for a tock is just disappointing compared to Nehalam/Gulftown->Sandy Bridge.

Nahalem -> Sandy would be a tick and a tock

Exactly, the level of improvement that Sandy had from Gulftown was much better than the improvement Haswell had from Ivy Bridge.

 

[Cazalan]

I prefer HasBurst™, the improvement for a tock is just disappointing compared to Nehalam/Gulftown->Sandy Bridge.

Nahalem -> Sandy would be a tick and a tock

Exactly, the level of improvement that Sandy had from Gulftown was much better than the improvement Haswell had from Ivy Bridge.

Funny I just read this old review and I think it's mainly the overclocking that gave Sandy it's famed status.

http://www.legitreviews.com/article/1722/11/

"Legit Bottom Line: If you do a blend of gaming, video, and photo editing the 15% - 25% performance gains are worth the price. If you are purely a gamer then just upgrade your video card and wait to see what the next generation brings to the fight. "

 

[JAYDEEJOHN]

Found this interesting:
I also asked Huang how investors should think about margins longer term, assuming a snap-back in Tegra revenue, given that Tegra brings lower gross margin than the company’s GPU parts such as “Kepler.”

“It will have an effect, but Tegra is not that much lower, it’s close to 50%” gross margin, Huang pointed out to me. “But as that takes off, we’d be very happy to have the volume” in Tegra sales, he added.
http://blogs.barrons.com/techtraderdaily/2013/05/10/nvda-bulls-raise-targets-on-second-half-product-cycle/?mod=BOLBlog

 

[jdwii]

I said like 4 pages back haswell is a disappointment then i got some Intel fanboys who can't see the truth and say otherwise. Honestly there's still no reason to upgrade from Sandy from a Intel standpoint. For laptops i would never even get a Intel unless its their I7 or I5 with Nvidia or Amd graphics and i'm not even talking about a pure gamers perspective, since i think it makes more sense to have a Laptop and a desktop, A laptop to do medico gaming such as a A10(not Intel graphics) and then a beast desktop. Anyways a A10 is pretty darn efficient for a laptop and i'm sure richfield is going to improve on this. As for the low-end market such as tablets Jaguar should be used much more than Intel's overrated 22nm design. I'll say it again Atom and Celeron have a horrible name(history) to it and their always stripped much more than the Amd counterpart.

I know their is odd businesses that choose to spend 1200$ on a Laptop which has the power of a 750$ desktop but mobile, but it just doesn't make sense unless you're always on the move and doing extremely heavy tasks that A10 or a I3 can't do. Most people on the move just need excel and powerpoint and of course long battery life for this a A10 or a I3 would be a better choice from a power consumption perspective.

 

[cowboy44mag]

I have a question that I'm not sure has been answered here before yet or not. Has AMD confirmed Steamroller FX will be on the current chipset or will it require something new? If you were to order the best FX motherboard right now it would be AM3+ 990FX SB 950. In only a couple weeks we will be half way through 2013. If Steamroller FX is going to release Q4 2013 or Q1 2014 I would have to assume AMD knows what chipset it will be on. I think that the AM3+ 990FX motherboards still have enough to offer for at least the first gen Steamroller processors.

 

[jdwii]

I have a question that I'm not sure has been answered here before yet or not. Has AMD confirmed Steamroller FX will be on the current chipset or will it require something new? If you were to order the best FX motherboard right now it would be AM3+ 990FX SB 950. In only a couple weeks we will be half way through 2013. If Steamroller FX is going to release Q4 2013 or Q1 2014 I would have to assume AMD knows what chipset it will be on. I think that the AM3+ 990FX motherboards still have enough to offer for at least the first gen Steamroller processors.

I believe it is going to be compatible

http://www.theinquirer.net/inquirer/news/2208525/amd-sticks-with-socket-am3-for-steamroller

But the next CPU excavator won't be compatible with Am3+.

 

[cowboy44mag]

I have a question that I'm not sure has been answered here before yet or not. Has AMD confirmed Steamroller FX will be on the current chipset or will it require something new? If you were to order the best FX motherboard right now it would be AM3+ 990FX SB 950. In only a couple weeks we will be half way through 2013. If Steamroller FX is going to release Q4 2013 or Q1 2014 I would have to assume AMD knows what chipset it will be on. I think that the AM3+ 990FX motherboards still have enough to offer for at least the first gen Steamroller processors.

I believe it is going to be compatible

http://www.theinquirer.net/inquirer/news/2208525/amd-sticks-with-socket-am3-for-steamroller

But the next CPU excavator won't be compatible with Am3+.

Yea, I had read that article, but haven't seen any newer info. Is theinquirer.net a reliable source? Some of the sites you find online are total worthless, but others are usually right on the money. If Steamroller FX is compatible AM3+ has had a good lifespan for a motherboard.

 

[juanrga]

I prefer "hasbeen"

I like "hasbeen". I still prefer "failwell" because resembles the typical "faildozer" that you could read in so many forums when AMD released Bulldozer.

Haswell isn't anything remotely close to the Bulldozer fail.

Minor IPC gain, large graphics gain, and better power savings is what they were after. It has little benefit for desktop enthusiasts but the laptops will love it.

I dont' call that "minor" but nonexistent, because is less than the error. The improved graphics are useless:

What is even worse for Haswell in this state is that at stock, the Core i7-4770K barely improves gaming performance representing only 1% improvement over the 3770K. The upgrade to Haswell is only worth if you are upgrading from either Clarkdale or Nehalem architecture’s otherwise from the review results, i’d advise even Sandy Bridge users to stick with their current platforms. The only notable improvement in Haswell processors would be its HD 4600 graphics chip which offers around twice the performance over HD 4000 but for a chip that costs $300+, it would be mad to not use a discrete graphics solution.

And they used 2600 MHz RAM!!! Repeat the tests using 1600 MHz.

You can find higher power consumptions and poor performance per watt (The early rumours about Intel having power problems by moving the VRM on-chip are now confirmed).

http://wccftech.com/intel-haswell-core-i7-4770k-review-live-china/

Plus the usb3 issues, the PSU incompatibilities...

 

[mayankleoboy1]

Instead of pure CPU power use, a better metric would be the platform power use. With Haswell that is going to be better, specially for notebooks. And the whole philosophy behind Haswell is that its a notebook oriented part.