First complete review of Haswell i7-4770K

[NTMBK]

Possible thread derail, and I apologize, but I wish my FX8350 was an APU for the same reason. Love the CPU horsepower, but the fact that I have to throw in a discrete GPU card that is going to suck down another 30W idle for 2D desktop stuff just bums me out.

Hopefully steamroller powered APUs will have FX SKUs so people don't have to step back from so much CPU power just to get an IGP with their AMD rigs as they do now with the Trinity vs Piledriver FX SKUs.

Do the AM3+ motherboards not have an IGP? My old AM2 one does, but it was an nForce and not a standard chipset. Obviously its going to suck compared to a modern APU, but it should be good enough for 2D.

 

[SiliconWars]

Maybe for someone that builds their own system. But do you think OEMs are going to put expensive ram in budget apu box? Heck, sometimes they still dont even put in dual channel.

Memory is so overpriced right now that it's worth buying some better stuff for what you pay.

 

[AtenRa]

Possible thread derail, and I apologize, but I wish my FX8350 was an APU for the same reason. Love the CPU horsepower, but the fact that I have to throw in a discrete GPU card that is going to suck down another 30W idle for 2D desktop stuff just bums me out.

Hopefully steamroller powered APUs will have FX SKUs so people don't have to step back from so much CPU power just to get an IGP with their AMD rigs as they do now with the Trinity vs Piledriver FX SKUs.

It would be much better if they would produce an 8 core + small iGPU(128 Radeon cores) for a high end FM2. It would even be smaller (die size) than 8 core BD/PD at the same 32nm.

 

[grimpr]

It would be much better if they would produce an 8 core + small iGPU(128 Radeon cores) for a high end FM2. It would even be smaller (die size) than 8 core BD/PD at the same 32nm.

Come on, AMD must merge the sockets and chipsets and come out on FM3 with APUs and FXs, one platform to rule them all.

 

[AtenRa]

Come on, AMD must merge the sockets and chipsets and come out on FM3 with APUs and FXs, one platform to rule them all.

I agree but i dont see it coming soon.

 

[guskline]

Come on, AMD must merge the sockets and chipsets and come out on FM3 with APUs and FXs, one platform to rule them all.

Makes sense to me. I've felt all along that PileDriver was the end of the FX line due to the beating Bulldozer took. I think steamroller will be an APU on a new chipset primarily to address the power issues. It has been very silent from AMD about the Steamroller. Every article about it being a socket AM3+ has it's roots in articles before CEO Rory took the helm. I could be wrong but it seems a LONG time since there has been any mention of Steamroller "officially" from AMD. Perhaps, just perhaps, CEO Rory has the engineers focusing on APUs with Steamroller cores.

 

[SiliconWars]

AMD needs to lessen their process disadvantage so they can't really afford 30-40W GPU's on to their top end chips right now. Doing so would be the final retreat from the true high end.

Kaveri is likely to see hexa-cores though, and that might be enough to start pushing back into i5 revenue space.

 

[Remobz]

Guys how soon will games that take advantage of 6 cores start being more or less common?

I am wondering if any advantage upgrading to Haswell right now. Or upgrade to IVY and keep it for a year or two.

 

[grimpr]

Makes sense to me. I've felt all along that PileDriver was the end of the FX line due to the beating Bulldozer took. I think steamroller will be an APU on a new chipset primarily to address the power issues. It has been very silent from AMD about the Steamroller. Every article about it being a socket AM3+ has it's roots in articles before CEO Rory took the helm. I could be wrong but it seems a LONG time since there has been any mention of Steamroller "officially" from AMD. Perhaps, just perhaps, CEO Rory has the engineers focusing on APUs with Steamroller cores.

We'll know that when the first info of Desktop FX SKUs with integrated pcie 2.0/3.0 arrive since the FM platform doesnt come with NB, Kaveri arrives later this year and rumours have it on FM2+, new electricals, i think its the right time to merge the FX Steamroller 28nm cores with pcie ondie and package them on FM2+.

 

[guskline]

We'll know that when the first info of Desktop FX SKUs with integrated pcie 2.0/3.0 arrive since the FM platform doesnt come with NB, Kaveri arrives later this year and rumours have it on FM2+, new electricals, i think its the right time to merge the FX Steamroller 28nm cores with pcie ondie and package them on FM2+.

I agree. Doesn't bode well for AM3+ sockets!

 

[inf64]

I agree. Doesn't bode well for AM3+ sockets!

Doesn't bode well for FM2 socket either .
With progress and speed (kaveri) you have to sacrifice the socket or else you won't have neither(progress/speed). There is a slight chance Kaveri might work on FM2 but this is just speculation at the moment.

 

[SPBHM]

Do the AM3+ motherboards not have an IGP? My old AM2 one does, but it was an nForce and not a standard chipset. Obviously its going to suck compared to a modern APU, but it should be good enough for 2D.

the official AM3+ chipsets don't have any IGP (970,990FX), BUT there is no limitation to using old chipsets...
most cheap AM3+ boards use the old 760G, so they do have a basic IGP (but it's really bad, it makes the HD2000 looks good).

IGP on AM3+ doesn't make much sense anyway... AMD is pushing their APUs, and the memory bandwidth on Am2/3 for the IGP is a huge limitation (since the IGP is in a separate chip, connected using HT)...

 

[galego]

This is just more strawman argumentation.

There is no such thing as making code that doesn't follow the law, all you've done is make code that has different parameters that define a different limiting value.

I find this whole vector of "must debunk Amdahl" laughable, as if there is anything to debunk and as if debunking Amdahl actually accomplishes anything. Reminds me so much of the so-called "1GHz barrier". Bunch of nonsense.

Amdahl's law will only fail to make any sense if the individual attempting to apply Amdahl's law has failed themselves to properly characterize the workload and application in question.

Do your part of the job correctly and the math naturally falls into place, don't do it correctly and suddenly you find yourself grasping for straws and magic to explain your seemingly "Amdahl Law busting!" data set.

The law is derived from a set of assumptions which apply to some kind of code but don't apply to other kind of codes. As a consequence there exists code that does not satisfy the law or, what is the same, there is code whose speedups cannot be explained by the law. That is different from your claim about specific pieces of code that satisfy the law having different Amdahl parameters.

Generalizations of the Amdahl law have been proposed

http://arxiv.org/pdf/cs/0209029

That doesn't make a big difference. Not more than 5-10%.

Here you are repeating to me something that I said before:

Subtract about a 5-10% performance from running the AMD APU with memory below its stock speed.

 

[tweakboy]

Who cares,, bring on the Ivy E processors 4930k I will be getting. That will beat Haswell, only the iGPU is much faster on the haswell which I dont care about since I use graphics dedicated card. thx gl

 

[Sleepingforest]

The law is derived from a set of assumptions which apply to some kind of code but don't apply to other kind of codes. As a consequence there exists code that does not satisfy the law or, what is the same, there is code whose speedups cannot be explained by the law. That is different from your claim about specific pieces of code that satisfy the law having different Amdahl parameters. Generalizations of the Amdahl law have been proposed http://arxiv.org/pdf/cs/0209029

First of all, that paper doesn't seem to be received well, or perhaps even accurate; it doesn't seem to get cited anywhere.

Secondly, do you even understand what is being proposed in that generalization? It says that Amdahl's law holds true unless you have a very specific case in which the speedup is theoretically indefinite (you can continue to parrallelize and get better results) given that the dimensions of the problem being solved increases with the number of cores the problem is being solved on. Otherwise, Amdahl's law holds true: the maximum speedup possible is roughly equal to the inverse of the portion of code which remains serial.

Can you name any code which falls under the first category (the dimensions of the problem can expand indefinitely)?

 

[Tuna-Fish]

I agree but i dont see it coming soon.

Personally, I expect it to come about when they switch to DDR4. Maintaining different sockets has a lot of costs. Switching sockets has a lot of costs, too. DDR4 is a transition that they will have to do eventually, and when they do, they have to design a new socket. At that point, might as well make it an unified one for their whole lineup. (Oh, and DDR4 is sufficiently electrically incompatible that they won't be able to do the AM2/3 kind of switch.)

 

[ShintaiDK]

Its always better to make a new socket, since you can overcome the design limitations associated with an old socket.

 

[Arachnotronic]

@ mikk

So you showed us that 1 year old 5800K has noticeably better iGPU than yet to launch 4770K with HD4600 inside. Richland will offer +15% so yes it will walk away (as I mentioned Richland in my post ). 1.15x1.07=1.23 or 23% better performance on average for just clock bumped Trinity (which is what Richland is). There will be games where Trinity's +20% will turn into +40% on Richland. AMD is positioned nicely on desktop I have to say.

AMD's best, bet-the-company-on-fusion part in a power-unconstrained environment will be +40% faster than Intel's nowhere-near-top-tier iGPU?

Say hello to Haswell GT3e. Intel will not only command premium pricing for it, but it will quite handily mop the floor with Richland in an AIO with superior CPU and GPU, while consuming less power.

AMD's graphics advantage just died. Broadwell will read the eulogy.

AMD needs to go all in on Kabini and defending its turf against Baytrail-M for low cost laptops/convertibles, or it is well and truly screwed.

 

[Cerb]

Its always better to make a new socket, since you can overcome the design limitations associated with an old socket.

But think of all those people that spent more money on an AM3+ motherboard, justifying it by thinking they'd get many great CPU upgrades out of it...

Any honest observer should realize that AMD has had long socket support because they needed to do it for the comparable Opterons, so there would be little advantage to making a whole new socket. Any CPUs without Opteron variants going into sockets with some guaranteed support for X years will have no need to be tied down like that, across generations.

 

[grimpr]

AMD's best, bet-the-company-on-fusion part in a power-unconstrained environment will be +40% faster than Intel's nowhere-near-top-tier iGPU?

Say hello to Haswell GT3e. Intel will not only command premium pricing for it, but it will quite handily mop the floor with Richland in an AIO with superior CPU and GPU, while consuming less power.

AMD's graphics advantage just died. Broadwell will read the eulogy.

AMD needs to go all in on Kabini and defending its turf against Baytrail-M for low cost laptops/convertibles, or it is well and truly screwed.

Yes yes, big words and sounds from financial speculators on stock market forums with an axe against AMD and an Intel stock portfolio, we heard them before a thousand times just before they'd hit the sewage drain.

 

[IntelUser2000]

Just walk away

It looks like the games that had extremely high advantage for 7660D over HD 4000 may have been related to HD 4000 drivers' "low clock bug".

Some games in the older HD 4000 drivers don't run Turbo, but only at the base frequency of 650MHz. Being able to take advantage of Graphics Turbo means potential gain of 50%+ alone. The new tests on the HD4000 is probably tested with the drivers that fix the "low clock bug", hence the much better gains.

 

[ShintaiDK]

But think of all those people that spent more money on an AM3+ motherboard, justifying it by thinking they'd get many great CPU upgrades out of it...

Any honest observer should realize that AMD has had long socket support because they needed to do it for the comparable Opterons, so there would be little advantage to making a whole new socket. Any CPUs without Opteron variants going into sockets with some guaranteed support for X years will have no need to be tied down like that, across generations.

We already know future single socket Opterons are APUs. Meaning no AM3+.

Socket compability doesnt mean anything if the platform cant be updated anyway with proper BIOS support and validation. Not to mention all the performance and effieciency penalties.

But again, AMD is dead in the server space. So completely irrelevant today.

 

[R0H1T]

AMD's best, bet-the-company-on-fusion part in a power-unconstrained environment will be +40% faster than Intel's nowhere-near-top-tier iGPU?

Say hello to Haswell GT3e. Intel will not only command premium pricing for it, but it will quite handily mop the floor with Richland in an AIO with superior CPU and GPU, while consuming less power.

AMD's graphics advantage just died. Broadwell will read the eulogy.

AMD needs to go all in on Kabini and defending its turf against Baytrail-M for low cost laptops/convertibles, or it is well and truly screwed.

You do realize that the current APU lineup doesn't have GCN ? The graphics in there is what 2~3yrs old & a full node behind Intel so before you start cheering for Intel again try to make a meaningful comparison because this one point is always brought up in Atom vs ARM debate !

 

[Cerb]

But again, AMD is dead in the server space. So completely irrelevant today.

But, they're going to have ARMs, and stuff (and just like everyone else, won't be able to explain why we should want their added, "stuff," in hard terms)...and keep investors holding on with it just a little bit longer. I guarantee you it will be great for Cloud(tm)! Oh, and ideal for your Big Data(tm) processing needs!

Anyway, though, point was it was a happy accident for the lucky few that could take advantage of it that they have had long-lived sockets, because it was useful for some important Opteron customers, not that long-lived sockets offered any particular advantages for them, or that they would want them if they didn't have some non-consumer business reasons to keep them around. If they weren't already designing for some systems that warranted in-place upgrades, we likely would not have had the AM2->AM2+->AM3->AM3+ gradual change, as it came largely from Opterons keeping one socket for more than one hardware generation, and their luck at DDR2 signaling being moderately compatible with DDR3 (come to think of it, didn't they request that?).

 

[Arachnotronic]

You do realize that the current APU lineup doesn't have GCN ? The graphics in there is what 2~3yrs old & a full node behind Intel so before you start cheering for Intel again try to make a meaningful comparison because this one point is always brought up in Atom vs ARM debate !

The gap continues to shrink at the high end.

That being said, I think "Temash"/"Kabini" will do *very* well in the low end, and I don't think "Baytrail-M" really has a shot of being competitive in graphics there.

There's no doubt that AMD's GCN is still a superior architecture, but how long does this superiority last? How much room does AMD have to beef up GPU performance in "Kaveri" given that it is moving from 32nm -> 28nm, which doesn't exactly buy a lot of die space/power savings at a given performance design point?

These are the tough questions, and I suspect that's why AMD is going to go all in on low power cores to make sure that it defends the high volume/low end stronghold.

Time will tell, though.