[ 3 / biz / cgl / ck / diy / fa / g / ic / jp / lit / sci / tg / vr ] [ index / top / reports / report a bug ] [ 4plebs / archive.moe / rbt ]

2016/06/28: We're halfway there! 2016/06/15: We've run out of space for images and need to upgrade our storage. Warosu is looking for a sponsor to support this effort. Email "haha at warosu" if interested.

/g/ - Technology

View post   

[ Toggle deleted replies ]
File: 75 KB, 560x268, AMD_BulldozerVsSteamroller.jpg[1].jpg [View same] [iqdb] [saucenao] [google] [report]
27282894 No.27282894 [Reply] [Original] [archive.moe] [rbt]


>According to Mark Papermaster, the improvements should yield up to 30% performance increase, but our sources inside the company beg to differ.

>"Steamroller is not Bulldozer Enhanced. F*** no. The layout might look the same but our LEGO blocks are completely different. When all is said and done we should get 45% improvement and this goes to show how the Bulldozer was f***** design. This is all what Bulldozer was supposed to be."

>> No.27282919

I like how they admit Bulldozer was a shitty design.

>> No.27282932

There's a first for everything.
Intel had its fair share of failed CPU designs.

>> No.27282943

AMD confirmed for bro-tier

>> No.27282978

So similar, yet so different.
A decode for each Integer unit and 2 decodes on the single buffed FP unit should make wonders happen for the IPC. Bulldozer could only do 16 instructions per clock, that's 8 less than Sandy's, and 2 less than Phenoms.

This should flip the tables.

>> No.27282980
File: 95 KB, 600x450, slide-1-728.jpg [View same] [iqdb] [saucenao] [google] [report]

Good new on the low end of things too. New Jaguar CPU cores to replace the Bobcat cores used in the low end APUs like current Brazos models.
Specs tout 10% faster IPC, 15% faster clock speeds, and 2 more cores at a lower TDP. These chips should have a 15watt max TDP. ULV Trinity will be taking over the 17 watt placement.
The design also boasts improved power gating, support for more instruction sets, and a 128bit FPU.

Considering that a little E-350 Brazos APU could fully handle 1080 Bluray playback without stuttering at all I'd say these chips look pretty good.

>> No.27283011

Can't wait to see how Steamroller performs when it comes out.

>> No.27283015

> Given that AMD had restricted resources, the company could not address the issues inside the memory controller and the execution cores, so Steamroller features significant improvements to execution cores, while memory controller is heading for a revamp with the 2014 architecture dubbed Excavator. 2014 is also the year when we are going to see the complete fusion between the GPU and CPU architectures, at least as far as AMD is concerned.

This is more important in my opinion, this should take care of the bottleneck their on-die GPU have.
Intel GPU don't have this bottleneck because their memory controller is several magnitudes stronger than AMD's.

They kinda stagnated on this for some reason, this shouldn't have happened since they're the ones that introduced the IMC

>> No.27283022

im planning on building a computer soon,and i was going to use an i3 2120 (budget limits) would i be better of waiting for these to come out? AMD seem to have amazing prices when compared to intel.

>> No.27283036

>Furthermore, AMD reorganized the L2 cache, which is now dynamic and can change its size per core. Gone are the days when every 2-core block had fixed amount of L2 cache and that was that. Steamroller is taking full advantage of the fact that AMD has exclusive, not inclusive cache architecture, and the L2 will stretch and shrink depending on the load on a particular core

After the magic they pulled with Trinity's battery life I can't wait to see this in action, the cache is like 40% of the entire bloody die.

>> No.27283044

If AMD can release a CPU at better price then Intel that renders faster in programs like 3ds Max then I'm sold.

If they can bring to the market dual socket CPUs you can actually overclock I'll probably give kiss them

>> No.27283066

Just imagine all of this implemented with non planar transistors

>> No.27283067

Meanwhile, Haswell is predicted to go _up to_ 10% increase in single core performance over Ivy.

>> No.27283073

I don't expect anything, but it would be really nice if AMD could at least provide a little competition to Intel.

>> No.27283075

I can see it now
>amdfags: wait for steamroller
>another fail cpu
>amdfags: huuurrr bios update

>> No.27283087

Not 40%, but it's pretty huge, the dieshot shows the dedicated 4 blocks of L2 cache almost the same size as a single module, which is 19.42mm^2

>> No.27283088

people seem to be forgetting that amd themselves said they're not competing in the high-end market anymore

>> No.27283098

Why would I care for $300+ CPU?

>> No.27283107

AMD? 45% increase?

they must have been sitting in the cave for years with bulldozer to think they've come across this impressive 45% increase in performance

>> No.27283121

Because you want the best!?

>> No.27283133

If I want the best I'd be buying POWER7 CPU, not x86 "we can't into parallelism" trash.

>> No.27283134

It's all marketing bullshit form AMD again, just like Bulldozer, how many slides did they put out about their imaginary performance increase

>> No.27283167

To be fair, this does go after Bulldozer's shortcomings, the shared hardware resources which were made to lower die size at the cost of IPC.

>> No.27283170


Enjoy your shit-tier price-to-performance ratio.

>> No.27283177

I must be so blind as to never have heard of power7

and because I haven't must mean its shit

>> No.27283179



>> No.27283192

to be fair, how many more excuses are we going to have to hear just because there's a severe case of amd fanboyism

>> No.27283200

Obviously that failed, they even admit it.
So now they're eating a bit of die size(with fixed decodes) to make up for the IPC.

2 steps forward, 1.5step back I guess.
Non conventional CPU architectures are a bitch.

>> No.27283220
File: 476 KB, 1946x997, piledriver block.png [View same] [iqdb] [saucenao] [google] [report]

Trinity APUs show around a 15% increase in IPC over Bulldozer in some benches while lacking L3 cache.
Thats a fairly impressive increase in that short of a time span. Vishera should easily be 10-15% faster than Bulldozer across the board.
If Steamroller is a 15% improvement over Piledriver then we're around 30% better than Bulldozer. This isn't hard to believe.

Improvements have been made all around the block design. Things like resonant clock mesh, and the maturation of GloFo's 32nm process have helped eliminate the absurd current leakage that Bulldozer suffered from. Improvements to the fetch, decode, and scheduler blocks have greatly improved prediction and execution by reducing miss penalties.
Steamroller is just a continuation of the incremental improvements that were made to the Piledriver block.

Point is that what they're saying here isn't far fetched at all.

>> No.27283227

This is nice to hear. Now, if AMD will just open its APU drivers so its Linux performance outpaces that of its Intel counterparts, they can have all of my money.

>> No.27283239

The dynamic L2 should help quite a bit, powering off the hungry cache should allow higher Turbo boost numbers, which is AMD's bread and butter for IPC.

>> No.27283243

So how much does AMD pay their marketing team and how do I apply?

>> No.27283259

You don't, they fired all of them.

>> No.27283268



>> No.27283321

I hate it, but unless AMD opens its graphics drivers for Linux, they won't get another cent from me. Despite the underwhelming CPU progress from Haswell, Intel HD 5000 graphics will still outperform anything from AMD simply because of those drivers.

>> No.27283329

>10~20% improvement per generation for two refreshes in row

Is this the new normal, or is Intel purposely fucking with its customers now that AMD doesn't have competitive products?

>> No.27283368

AMD has fully open spec for their GPUs like Intel.

It's that they've relegated the fully open driver development responsibility to X.org and X.org has been dragging its heel on Radeon driver development.

It's not helped by the fact that AMD doesn't seem to be giving additional funds to X.org or hiring its own engineers to implement the open-source drivers in conjunction with X.org.

>> No.27283398
File: 732 KB, 1479x4838, AMD.png [View same] [iqdb] [saucenao] [google] [report]


>> No.27283423

It is a product of Intel's lack of competition on the high end and lack of need of better CPUs. The market is in GPUs now, and they are hurrying along to outpace AMD's on-die graphics development.

>> No.27283446

Why doesn't Intel support coreboot?

>> No.27283475

Then they should contribute as Intel does to its free driver development. The issue is that they compete with themselves by releasing Catalyst for Linux in conjunction with Radeon. They should invest in free driver development to stamp out Intel's rapidly growing influence and jump ship on fglrx when Radeon finally performs just as well or better.

>> No.27283489

They came up with UEFI and wish to propogate it instead.

>> No.27283524

Because, Intel, like a good company doesn't care about shitty freetards.

>> No.27283553

Those shitty freetards released a BIOS replacement which is faster and more flexible than anything else right now.

>> No.27283570


>> No.27283595

Intel has contracts with American Megatrends, Phoenix and other proprietary BIOS manufacturers.

>> No.27283623

So... wait for Haswell?

My PC uses an overclocked 1090t. I won't need a CPU upgrade for a while, but when I do, I guess I'll go for Haswell since it will already be far more powerful than AMD and more powerful than Ivy Bridge.

>> No.27283629

If it was better then why does nobody but lolAMD care about it?

>> No.27283681

marketing, pr, misinformation, bribes

>> No.27283700

a great read about AMD's internal failures. thanks for posting.

>> No.27283835

>AMD's troubles with Barcelona (K10) and Shanghai (K10.5) were the sole result of a mismatched memory architecture, which saw L3 cache operating at the same clock as the memory controller, which was up to 50% slower than the regular clock

Only Dirk Meyer could pull this kind of shit and cover it up.

>> No.27283864

L/a/dies and /g/entoomen, please welcome for your internet browsing pleasure, from the world wide web, 4chan image boards!

We comin' to yo' board and then we trolling’ ova',

F8, but betta' yet, you betta' boot up

Backtrack says they wanna exploit with' a hacker,

Click, clack, hack, that's how I computer,

Hey, make way for the GCC
Hey, the freedoms we will bring
Hey, make way for the GCC
Bootin' gentoo, ‘cause we the Linux thing

>> No.27283923

Wtf did i just read

>> No.27284045

song parody of some kind

looks like a ke$ha song

>> No.27284062

thenks from me too

I am an "amd fanboi" in a way that I have athlon 2 and was waiting for bulldozer fix, but apparently company is slightly doomed.

>> No.27284089


>> No.27284107

they still have gpus and non-performance cpus

>> No.27284130

im just gonna wait for graphene.

>> No.27284143

Speaking of AMD GPUs.
What in the fuck does Nvidia think they're going to do for their next lineup of cards? AMD has been making solid advances all over. A 7950 can put a 660ti absolutely to shame, nearly matching a 670. Nvidia already had to cut compute hardware on Kepler to keep up with Southern Islands. They can't exactly do that again.
What would they do if AMD made the same move to cut power consumption and push clocks even higher?

>> No.27284152

Why do you think nVidia keeps pushing CUDA?

>> No.27284154

Graphene is supposed to enter usage around 2016.
HP Labs says that their memristors will be ready in 2013 though, so thats something.

>> No.27284171

I heard about that. Aren't they some knockoff of actual memristors or something along those lines? If they actually managed to perfect the technology, SSDs might not be doomed after all.

>> No.27284180

Oh boy I laughed. Nvidia is a horrible company and needs to burn.

It's good to see AMD management didn't fuck with ATi when they bought it out. I wonder how much money from the GPU department is used keeping their CPU division afloat.

>> No.27284185

>SSDs might not be doomed after all
You mean write cycle lifetime right?

Its already insanely high anyway right? Like several hundred Terabytes.

>> No.27284220

I was talking about the inverse relationship between capacity and performance which is expected to begin due to limitations of current SSD implementation.

It's like I keep saying; if AMD will just fund Radeon a bit more (or just open up Catalyst on Linux) and outperform nVidia and Intel in the Linux market, their APUs will be the absolute love of the free software fan. That is especially true if they keep up their support of Coreboot.

>> No.27284232

>Aren't they some knockoff of actual memristors or something along those lines?
HP were apparently the first ones to figure out how to make them.

Really nice keynote on them. Its about 47 minutes long and totally packed full of information.

The neatest thing about them is how power efficient they are. They can build stacks of them multitudes high and not have to worry about cooling.
Oh yeah, and theres that little trick they have where they can also perform fucking logic.

>> No.27284234

Grab an AM3+ motherboard and a Phenom II x4 965. You'll get a decent quad-core and have the ability to upgrade in the future. 965s are $90 at microcenter or $110 (I think) at NewEgg so they're still cheaper than the i3. And the 965 might run better in certain programs since it's a quad-core.

Many people love to spout that GAYMEN DOESN'T USE MORE THAN TWO CORES but that's just flatout BS at this point. Look at the recommended requirements for any recent game. The difference? Minimum = dual-core, recommended = quad-core. And BF3 has been proven to use 4+ cores. The future is to optimize games for more cores.

Plus the 965 is still a nice little chip, especially if you want to OC.

>> No.27284246

I believe it when we have independent benchmarks.

Are there any Trinity ultrabooks yet?

>> No.27284257

Why not snag a Trinity when they are released in boxed form in October? They are shown to outperform Bulldozer to the point of matching Phenom, and the dual-graphics feature might be helpful for certain games. Also, it actually gives one a reason to buy highly clocked, low-latency RAM.

>> No.27284273

The only trinity notebooks I can find are released by budget manufacturers. Intel hit AMD hard when signing new contracts with laptop OEMs.

>> No.27284275

ULV Trinity isn't out yet that I know of.
Regular notebooks and prebuilt desktops have been out for a while though.

>> No.27284286


>> No.27284323

Aha! I didn't catch that.

The specifications are out by now, but I have not seen them anywhere. Trinity in general is difficult to find.

>> No.27284411

I will be sure to check it out, but I heard bad things about HP's research. I can only hope that this is the real deal.

>> No.27284459

So how do you think the balance between ipc and clock speed will be this time?

Around 4ghz this time and no significant heat/power issues this time?

>> No.27284461

Guise, I haven't been following intel processors at all, now I look at article in wiki and it's a complete mess of sockets and bullshit.
What is the newest intel processor without unnecessary shit like GPU for around ~250$?

>> No.27284500

>amdfags trying to justify their buyers remorse

>> No.27284531

There are no mainstream Intel CPUs that don't include an on-die GPU. The general consensus around here for that price is anything in the i5-3570-ish range. For the record, all Ivy Bridge desktop CPUs use socket LGA1155.

Check out ark.intel.com for detailed descriptions of their various processors and processor lines. It can be a touch disorganized at times, but it helps when looking up socket type, etc.

>> No.27284538

>intelfag paying 2-3x what we did for our processors

>> No.27284550

Intel has a better cpu at every price point
nice try

>> No.27284592

are there performance differences between chipsets?
I don't care for raid and ssd caching, so I should be fine with B75

>> No.27284609

AM3+ is dead after Piledriver

>> No.27284637

Phenom-965 begs to differ.

>> No.27284937
File: 768 KB, 1280x720, [HorribleSubs] Moyashimon Returns - 09 [720p].mkv_snapshot_01.09_[2012.08.31_03.04.28].png [View same] [iqdb] [saucenao] [google] [report]

>he didn't get a 960T for $120 and turned it into a 6 core

>> No.27284953

>unlocking defective cores
How about no.

>> No.27285359


>> No.27285413


>AMD has done away with hand-drawn processor schematics (which define how all the internals are laid out and interconnected) in favor of using a computer-assembled design (a more automated design approach).

Daily Reminder: Every AMD CPU so far has been hand crafted by a professional AMD CPU Forgemaster who loves you so much that he drew every single little detail of the chip.

>> No.27285427

>he thinks that just because they are locked means they are completely defective

>> No.27285488

Got a 140eur 8120 this week which runs at 4.7GHz, if any of you Intel fans can show something like this for such a price for virtualisation I might consider it in the future. Bonus points: It runs on a old motherboard for their previous socket.

>> No.27285489

Almost. The compacted designs are BASED on the hand-drawn versions. If I do remember correctly, AMD had a lot of automated stuff in Bulldozer. Whatever works and makes it faster does it for me.

Jim Keller knows what's up. He made the Athlon 64, and AMD just hired him back last month and he's currently working on Steamroller/Excavator/Kaveri.

>> No.27285673

Im not sure if he can salvage whats left.

Arent most of the current CPU team a bunch of new guys who don't really give a shit?

>> No.27285695

they should hire the gentlemen behind the kepler optimizations

>> No.27285706

Clearly you missed on the news on Steamroller

>> No.27285745

>Bulldozer fucking sucks
>Releasing the next line of processors
>"This is what Bulldozer was supposed to be"

>Steamroller fucking sucks
>Releasing the next line of processors
>"This is what Steamroller was supposed to be"

>> No.27285817

You're missing Piledriver

>> No.27286063

Bulldozer was a pile of ass because the design goes back to 2005 (or even further). Since AMD fucked up Barcelona and had to do a Phenom II to get that design working and the whole screwing over by Intel during Athlon 64, they probably didn't have the resources to get BD out when it was supposed to. I guess they just bit the bullet and released it even though it was already dated and rushed.

>> No.27287392

The Athlon64 creator is back at AMD, and he is already on Steamroller/Excavator.

>> No.27287420

Intel cut them off and the R&D stopped. They barely had to pay any damages.

>> No.27289625

all of my money.
it's over, intel is finished.

Name (leave empty)
Comment (leave empty)
Password [?]Password used for file deletion.