Haswell will rival graphics performance of today's discrete cards!

[CPUarchitect]

Anyways I wanted to bump my question from the other page - is it possible that Haswell is fully homogenous?

I.e. software rendering for everything

Not likely. At least not yet. Some seemingly reliably sources give the Haswell graphics hardware a separate name: Denlow.

That said, Hawell (and it's shrink Broadwell) might very well be the last heterogeneous Intel architecture. The AVX2 instruction set adds 'gather' and 'fused multiply-add' support, two powerful operations which used to set the GPU apart from the CPU. Since it will make Haswell's CPU cores capable of high performance throughput computing, it's not unlikely that Denlow will actually be assisted by these AVX2 enabled cores.

Even NVIDIA is exploring pure software rasterization! It's only a matter of time before the added flexibility is more valuable than theoretical performance.

The biggest hurdle to make CPUs capable of taking over all graphics tasks, is power consumption due to the out-of-order execution pipeline. But that can be solved by executing very wide vectors over multiple clock cycles (just like a GPU does), and clock gating the out-of-order logic while it's underutilized. Intel already revealed that AVX can be extended up to 1024-bit. The earliest something like that can be expected is 2015 though, and the software has to follow suit as well, so it's still quite speculative. In any case it needs no question whether graphics will one day become 100% programmable.

 

[jones377]

Not likely. At least not yet. Some seemingly reliably sources give the Haswell graphics hardware a separate name: Denlow.

That said, Hawell (and it's shrink Broadwell) might very well be the last heterogeneous Intel architecture. The AVX2 instruction set adds 'gather' and 'fused multiply-add' support, two powerful operations which used to set the GPU apart from the CPU. Since it will make Haswell's CPU cores capable of high performance throughput computing, it's not unlikely that Denlow will actually be assisted by these AVX2 enabled cores.

Even NVIDIA is exploring pure software rasterization! It's only a matter of time before the added flexibility is more valuable than theoretical performance.

The biggest hurdle to make CPUs capable of taking over all graphics tasks, is power consumption due to the out-of-order execution pipeline. But that can be solved by executing very wide vectors over multiple clock cycles (just like a GPU does), and clock gating the out-of-order logic while it's underutilized. Intel already revealed that AVX can be extended up to 1024-bit. The earliest something like that can be expected is 2015 though, and the software has to follow suit as well, so it's still quite speculative. In any case it needs no question whether graphics will one day become 100% programmable.

And when every Intel CPU comes bundled with your GPU software rasterizer you stand to become the next Bill Gates, right?

 

[Riek]

Are talking discrete . Because if your not your trolling . SB had the best IGP ever till LLANO came out . This is not a discussion about Discrete graphics . and don't pretend ypu don't understand that . You are using red herring. That put intel 1/2 generation behind they had the lead earlier this year. Thats really 1/2 a generation as IB will be out . UNless trinity comes out befor Ivy. I never comparred Intel discrete to NV amd because Intel has no discrete. Hell I should start on how NV has nothing to compare with AMD intel CPUs.

This is a haswell thread . IGP are in . But discrete is out. Lets bring IBMs latest release in to the thread.

Are you going to insist AMD is 3 generations ahead of Intel on IGPs . Funny because In jan of 2011 that wasn't the case . Intel was the Leader. THATS a FACT . I got points taken in this topic. You going to say Intel didn't have the Best IGP ever in january of 2011?

You should have to back it up with proof . SHOW your proof.

Zacate was released in Q4 2010. It has lower gpu performance then HD3000 (better then HD2000). uses less power, has better image quality , supports Dx11 and openCL and fits within a power budget of <<18W on an inferior process node.

Intel will support Dx11 and openCl in Q2 2012. So yeah more then a year ahead.

 

[Arkadrel]

Isnt Haswell set for mid 2013?
We re still in 2011.... theres a 2year wait until Haswell gets here.

Meanwhile Trinity is set to launch early 2012, and have +50&#37; performance (claimed by amd) on Llano A-3850.

If Trinity gives +50% performance, we could see it perform like a
5670 ~ 9800 GT level card (Im curious about memory bandwidth though).

Haswell is gonna face whatever comes after Trinity.
By then DDR4 should be around to, hopefully makeing it abit easier to get iGPUs performance up.

 

[n0x1ous]

From Anandtech's review, I would have to disagree with you: http://www.anandtech.com/show/4476/amd-a83850-review/5. It depends on what games you wish to play, and on quite a lot of them, Llano is playable, HD3000 is not. But this isn't really THE source of our disagreement, because I am sure you've already seen that review.

Where we really differ on opinion is this, the root of why we disagree:

Your argument is still based on an assumption about the target audience, and this assumption is unsubstantiated (and the both of us would have a hard time proving that assumption). You are assuming that anybody who is not obsessed with hi-res, max settings, full AF/AA are "more casual players" and assume they would not be playing modern games like those in the AT review I linked to above. Just from a few friends I have, I know that assumption is simply not true. At any rate, since I am not the one making the assumption, I have nothing to prove.

And that's where we differ. Like in my first response to you, we simply have no data that proves that, so we can't make that assumption. And that definition of a "gamer" is a rather narrow one. You mean to say that the friend I mentioned in my response to you is not a hardcore gamer (even though he plays the same modern games as us, and just as frequently) simply because he gets along fine with 1024x768 or 1366x768?

Before I got out of college, I also did not bother with max settings or hi-res - in fact, I remember having to play Morrowind on 800x600 back in 2003, because that was the only way to get acceptable performance using our low-powered PC. Now, I play everything maxed out as much as possible. Does that mean I suddenly became a "hardcore gamer" only when I started having money of my own? I was a hardcore gamer long before I had money to waste on a gaming PC - I was in PC gaming since the XT processor when I started playing the original Prince of Persia and Steel Thunder (a tank simulation) on floppy disks.

That's really the meat of our difference of opinion. You assume that people who are satisfied with low-res and non-maxed-out settings are just casual gamers and are interested only in older/low-end games, I am not making any assumptions. And without such an assumption, you can see why Llano is a good enough product.


Irrelevant to the discussion of Llano's worth/value as a product. This is up to the OEMs to decide. This affects Llano's effect to AMD's bottomline, not Llano's value to its target market (end-user / consumer effect). Let's not mix the two issues up and focus our debate on Llano's value to end users. I would rather talk about how a product affects us, than how a product would affect the company's bottomline.

I am not asking you to stop posting about it I wouldn't want you to confuse what I am saying with that. I simply have no interest in talking about it, so I am not likely to respond to such topics further, but you are completely free to do so, as others here may want to talk about it.

Cheers.

Agree with some of your points. We can't prove this either way. You do present the example of your friend, but a couple people you know who would be a good fit for Llano hardly a trend makes. I know a few people who fit it the way I see it. IE : Don't know jack about PC's and play all their games on consoles only or play farmville on their PC's.

Certainly there are going to be some people who Llano is perfect for like the friend of yours and im sure many others. I just think that AMD is going to find that giving a decent GPU to everyone doesnt mean everyone of those people are going to use it or even care about it. I would be willing to bet that you "know" (not necessarily friends with) that don't care about graphics on their PC at all and just use it for browsing, email, Facebook. I am glad that the baseline performance of a PC has been increased by AMD with the Llano chip and in a few years down the road I believe it will be sufficient to play more 3d games at respectable performance levels, but I dont think we are quite there yet.

Also, i never said in my earlier post that hardcore or pc gamers need to have everything maxed out to be a real hardcore gamer, but if someone could get better visuals and/or performance on their 360/ps3 because they don't have a discreet GPU then I think many of them would stick to a console.

 

[jvroig]

Agree with some of your points. We can't prove this either way. You do present the example of your friend, but a couple people you know who would be a good fit for Llano hardly a trend makes. I know a few people who fit it the way I see it. IE : Don't know jack about PC's and play all their games on consoles only or play farmville on their PC's.

Well our points are now a bit closer to each other, that's progress

Re: "a couple people you know... hardly a trend makes". That's true! That's just the thing, I wasn't proposing it was a trend - I was merely countering your absolute, negative statement that nobody will want Llano's performance, specifically:

And yet it doesn't matter, because anyone who is going to play any type of 3D game on their PC is not going to be happy with Llano's performance.

That's all I was doing - not proposing a trend, not even saying Llano is important to everybody - merely countering this absolute, negative statement. That is why in my last post I said "At any rate, since I am not the one making the assumption, I have nothing to prove."precisely because I made no assumptions myself, but merely showed that the assumption made is wrong because it is too absolute but has no factual basis to support such a strong, absolute statement (it's ok, for example, to say that "the 2600k absolutely walks all over a 1100T, everyday, twice on a Sunday", because despite being so strong and absolute, we have tons of empirical data to support it)

Certainly there are going to be some people who Llano is perfect for like the friend of yours and im sure many others.

I'm glad you see it this way, because your earlier post (quoted above) seemed to indicate otherwise. Had you stated it this way, you would have gotten no argument from me, because I agree with this (not just this particular statement, but the entire paragraph as well)

Also, i never said in my earlier post that hardcore or pc gamers need to have everything maxed out to be a real hardcore gamer, but if someone could get better visuals and/or performance on their 360/ps3 because they don't have a discreet GPU then I think many of them would stick to a console.

Then maybe I have misinterpreted your earlier post (again, the one I quoted above, post #74 of this thread), because when I read it, I interpreted it to mean exactly that. It happens, I suppose, misinterpreting posts happen in the internet.

As for consoles, it's an issue more complicated than I can address, but Llano isn't supposed to take market share from consoles anyway - just from low-end graphics cards that have traditionally been high-volume. Whether Llano ends up luring people away from consoles or not probably doesn't matter, as long as Llano ends up replacing low-end video card sales.

 

[jvroig]

Meanwhile Trinity is set to launch early 2012, and have +50&#37; performance (claimed by amd) on Llano A-3850.

Early 2012?

Bulldozer is yet to arrive Q3 2011, and Trinity is already supposed to launch early 2012? That looks a tad too optimistic. I have not kept up with the news, so I am not saying you are wrong. It is just surprising and looks like a very optimistic estimate from AMD.

 

[ieatdonuts]

^^ what jvroig said. I'll be damned if they can get Trinity out the door in early 2012 with Bulldozer ending up the way it has. However, it is possible - because it's on 32 nm process, it takes an already designed and manufactured Bulldozer core and matches it with a Cayman IGP. Only problem that lies in there is the shrink 40-32 nm of Cayman. And they had a working demo laptop in June of this year....so who knows....

As for Haswell, if it isn't fully homogeneous then there would have to be some sort of way for the CPU to assist a basic GPU because otherwise all this AVX2 and FMA etc. would be rather useless.

Even NVIDIA is exploring pure software rasterization! It's only a matter of time before the added flexibility is more valuable than theoretical performance.

Wow, that looks really interesting. I'm gonna read it. I hope this happens, too much unnecessary bloat in graphics rendering.

 

[bryanW1995]

Early 2012?

Bulldozer is yet to arrive Q3 2011, and Trinity is already supposed to launch early 2012? That looks a tad too optimistic. I have not kept up with the news, so I am not saying you are wrong. It is just surprising and looks like a very optimistic estimate from AMD.

AMD ALWAYS hits their launch target dates. Don't you dare say otherwise, or I'll get hector ruiz hired as your CEO!

 

[jvroig]

AMD ALWAYS hits their launch target dates. Don't you dare say otherwise, or I'll get hector ruiz hired as your CEO!

Sorry, sorry, I take it back, don't shoot!

This is not to say I don't want Trinity to appear ASAP - god knows if AMD can quickly replace Llano SKUs with SKUs that have a more powerful CPU component, that would be better for a lot more people. It just caught me by surprise considering desktop BD was delayed and Trinity is based on BD, so I can only assume that whatever problems BD encountered, the Trinity team must have also, so I don't know how it affects their supposed timelines.

 

[bryanW1995]

If they get the issues fixed pre-launch then it might not impact trinity at all, but a too-fast switch to fm2 would probably cause some grumbling amongst the faithful. Personally, I would only care about which one clocks higher, but there is quite likely a good number of people who already have am3+ and are just waiting for sept 19.

 

[n0x1ous]

Well our points are now a bit closer to each other, that's progress

Re: "a couple people you know... hardly a trend makes". That's true! That's just the thing, I wasn't proposing it was a trend - I was merely countering your absolute, negative statement that nobody will want Llano's performance, specifically:

That's all I was doing - not proposing a trend, not even saying Llano is important to everybody - merely countering this absolute, negative statement. That is why in my last post I said "At any rate, since I am not the one making the assumption, I have nothing to prove."precisely because I made no assumptions myself, but merely showed that the assumption made is wrong because it is too absolute but has no factual basis to support such a strong, absolute statement (it's ok, for example, to say that "the 2600k absolutely walks all over a 1100T, everyday, twice on a Sunday", because despite being so strong and absolute, we have tons of empirical data to support it)


I'm glad you see it this way, because your earlier post (quoted above) seemed to indicate otherwise. Had you stated it this way, you would have gotten no argument from me, because I agree with this (not just this particular statement, but the entire paragraph as well)


Then maybe I have misinterpreted your earlier post (again, the one I quoted above, post #74 of this thread), because when I read it, I interpreted it to mean exactly that. It happens, I suppose, misinterpreting posts happen in the internet.

As for consoles, it's an issue more complicated than I can address, but Llano isn't supposed to take market share from consoles anyway - just from low-end graphics cards that have traditionally been high-volume. Whether Llano ends up luring people away from consoles or not probably doesn't matter, as long as Llano ends up replacing low-end video card sales.

I overstated/overreacted by saying "nobody" wants/needs this. I think we are in agreement that it will be great for some users. I just don't think (in its 1st iteration anyway, trinity should be much better) its quite all its cracked up to be so far for gaming. For an HTPC however, it would be spot on.

 

[jvroig]

I overstated/overreacted by saying "nobody" wants/needs this. I think we are in agreement that it will be great for some users. I just don't think (in its 1st iteration anyway, trinity should be much better) its quite all its cracked up to be so far for gaming. For an HTPC however, it would be spot on.

Completely agree. Amazing what a calm, rational conversation can end up in :thumbsup: Had we jumped at each other guns blazing, I doubt we would have found common ground at all.

FWIW, if somebody were to tell me Llano is "perfect", I'd be all up in his business. It's quite adequate, maybe even good, for its target market, but it certainly is far from being the perfect solution, especially for gaming. Trinity does seem to hold better promise (better CPU and better GPU, what's not to like), but my main complaint against Llano is price. When it launched, the price disappointed me. Not sure how cheap I was expecting it to be, maybe ~$30 cheaper (combined mobo+chip price) to make it a great deal and a no-brainer. Oh well.

 

[Jionix]

Being It been officially made a llano thread. I will assume were talking Ondie grapgics here. With that said . AMD isn't anywere near 3 generations Ahead. Read ATs review of SB graphics . He doesn't over praise it. But He does say its the first IGP ever that you can play many games at at reasonable frame rates at lower RES. AMD had nothing that could match it till llano came out . So please tell were AMD is three generations ahead. IF your talking discrete . INtel hasn't got a product . But for about 12 billion they could have one dam quik. That would be the dumbest thinf intel could ever do . Intel may buy Imagination tech tho . As they own 25% of the company now and apple owns 15%.

As soon as AMD put Fusion into market, they jumped 3 generations ahead on IGP technology. Intel DOES NOT RENDER PROPERLY. They do not support DX 11. They won't, theorectically, until Haswell, which is at least a year off still.

By that time, AMD will have Trinity, maybe even something else on the block to release at the same time. AMD will be generations ahead, with better performance, and CORRECT RENDERING.

Intel doesn't even have the basics of filtering implamented correctly right now. And the only reason HD 3000 is 1/3rd the speed of Llano instead of being 1/6th, is because Intel is getting away with corner-cutting rendering, and falsely getting a preceived speed bump because of it.

Intel could double their EU in their chips right now, and suddenly be breathing down Llano's neck. DOESN'T MATTER. Until they fix their drivers and actually render games correctly, the speed of Intel IGP is a complete farce.

 

[bryanW1995]

That's true, but up until a few years ago "speed" and "intel igp" were never even in the same sentence. Intel has proven time and time again that if you throw enough $$$ and engineers at an issue you typically come out ahead over time. AMD has a chance here to capture true mindshare as well as marketshare, they need to capitalize on this and build their reputation while it's still better than intel's in this arena.

 

[AnandThenMan]

Are talking discrete . Because if your not your trolling . SB had the best IGP ever till LLANO came out . This is not a discussion about Discrete graphics .

You're. You are. BTW, is your keyboard broken? Your posts are often riddled with all manner of errors.

Now, do you realize that discreet, mobile, on-die, Fusion, whatever, they are based on the same basic tech? Why are you so desperate to exclude discreet in this discussion? Because AMD has the discreet technology, they are able to leverage that into other form factors. Intel doesn't have this advantage, and their GPUs end up being very poor as a result. This extends to drivers as well, AMD and Nvidia have extensive experience with making highly sophisticated drivers, in large part because they push out high end discreet products.

And on the subject of drivers, Intel is dreadful here. Image quality is horrible (Anandtech.com seems to completely ignore this fact in reviews) and stability is not that great either.

 

[Idontcare]

You're. You are. BTW, is your keyboard broken? Your posts are often riddled with all manner of errors.

Now, do you realize that discreet, mobile, on-die, Fusion, whatever, they are based on the same basic tech? Why are you so desperate to exclude discreet in this discussion? Because AMD has the discreet technology, they are able to leverage that into other form factors. Intel doesn't have this advantage, and their GPUs end up being very poor as a result. This extends to drivers as well, AMD and Nvidia have extensive experience with making highly sophisticated drivers, in large part because they push out high end discreet products.

And on the subject of drivers, Intel is dreadful here. Image quality is horrible (Anandtech.com seems to completely ignore this fact in reviews) and stability is not that great either.

I'm not big on the grammar nazi BS.

Whether or not he posts your versus you're, the fact of the matter is that given the context of the topic, combined with your intelligence of deduction, you can and do figure out exactly what it is/was that he is/was attempting to communicate to you and the rest of the community.

Now climb down off your high horse before someone, in a bit of cosmic justice, publicly berates you over the fact that it's discrete, not discreet.

The adjective discreet means prudently self-restrained or tactful. Discrete means distinct or separate.

http://grammar.about.com/od/alightersideofwriting/a/discreetgloss.htm

 

[AnandThenMan]

Now climb down off your high horse before someone, in a bit of cosmic justice, publicly berates you over the fact that it's discrete, not discreet.


http://grammar.about.com/od/alightersideofwriting/a/discreetgloss.htm

I welcome being corrected.

 

[AtenRa]

Meanwhile Trinity is set to launch early 2012, and have +50% performance (claimed by amd) on Llano A-3850.

When AMD said that Trinity will have 50% more performance over Llano ?? any links ??

 

[ieatdonuts]

Y'know you could google that - I googled "AMD Trinity 50&#37;" and found this:

http://www.zdnet.com/blog/computers...performance-boost-over-llano-due-in-2012/6177

 

[bryanW1995]

Is it just me, or does anybody else wonder if any NBA players ever see that "Trinity" is now the name of a new apu and just kind of smile?

 

[Arkadrel]

Completely agree. Amazing what a calm, rational conversation can end up in :thumbsup: Had we jumped at each other guns blazing, I doubt we would have found common ground at all.

FWIW, if somebody were to tell me Llano is "perfect", I'd be all up in his business. It's quite adequate, maybe even good, for its target market, but it certainly is far from being the perfect solution, especially for gaming. Trinity does seem to hold better promise (better CPU and better GPU, what's not to like), but my main complaint against Llano is price. When it launched, the price disappointed me. Not sure how cheap I was expecting it to be, maybe ~$30 cheaper (combined mobo+chip price) to make it a great deal and a no-brainer. Oh well.

The A-3850 is like 139$ on newegg.

What would be the alternative from Intel? i3-2100 + Nvidia GF 430?

I3-2100 is around 125$ (on newegg) and a Nvidia GF 430 ~35$ (with rebates) (160$ total).

The A8-3850 is probably a faster overall, than the i3-2100+nv GF430.

Pricewise it doesnt seem so bad.

 

[AtenRa]

Y'know you could google that - I googled "AMD Trinity 50%" and found this:

http://www.zdnet.com/blog/computers...performance-boost-over-llano-due-in-2012/6177

AMD said 50% more GFLOPs not 50% more performance.

http://www.brightsideofnews.com/new...-trinity2c-promises-10tflops-apu-by-2020.aspx

 

[Nemesis 1]

As soon as AMD put Fusion into market, they jumped 3 generations ahead on IGP technology. Intel DOES NOT RENDER PROPERLY. They do not support DX 11. They won't, theorectically, until Haswell, which is at least a year off still.

By that time, AMD will have Trinity, maybe even something else on the block to release at the same time. AMD will be generations ahead, with better performance, and CORRECT RENDERING.

Intel doesn't even have the basics of filtering implamented correctly right now. And the only reason HD 3000 is 1/3rd the speed of Llano instead of being 1/6th, is because Intel is getting away with corner-cutting rendering, and falsely getting a preceived speed bump because of it.

Intel could double their EU in their chips right now, and suddenly be breathing down Llano's neck. DOESN'T MATTER. Until they fix their drivers and actually render games correctly, the speed of Intel IGP is a complete farce.

Hay . Since my last post here I have pretty much enjoyed this topic as it went back towards topic. You can talk all day long about DX11 and How Fusion Llano coming out suddenenly leeps 3 generations ahead . SB . Jan . 2011 . Undisputed best ever IGP until LLANO . Because LLANO is DX11 scrathes arse, So when IB appears with DX11 will intel magicly jump 4 generations ahead because Intels compiler tech is 4 generations ahead of AMDS. Or don't compilers have a starring role in open CL AVX and What ever the proper term is for larrabbee instruction set . If you believe Intel isn't going to intertwine these in the compilers your sadly mistaken . The pre fix of Vex will play a starring rule in this if you have read the info on AVX 2 you would see this .

 

[AnandThenMan]

...So when IB appears with DX11 will intel magicly jump 4 generations ahead because Intels compiler tech is 4 generations ahead of AMDS.

When Haswell comes out, AMD will be 6 generations ahead of Intel. Yea, I can make stuff up too.