First complete review of Haswell i7-4770K

[Rvenger]

Intels four core chips w/HT already beat AMDs eight core chips.

...And that doesn't mean AMD is dead. They actually sell just fine.

 

[guskline]

Isn't it funny how the thread topic was First complete review of Haswell i7-4770K yet it has morphed into another AMD vs Intel slugfest?

How many days until Intel "officially" releases Haswell ?

 

[R0H1T]

Have we seen any confirmation of there being a 3-module Kaveri? The roadmaps I saw all said 4 cores i.e. 2 modules.

And I can't see any way that AMD could fit 6 Steamroller cores and a 7750 level GPU onto the same die without hitting Sandy Bridge E level of die size... This seems like wishful thinking which has been repeated so much that people assume it is true.

Like how everyone here drools at the idea of an eight core chip with HT from Intel yet they don't have any plan/roadmap to entertain these folks ? I can't see why such a chip isn't plausible given that the TDP doesn't exceed ~150W levels, with steamroller or maybe excavator cores, because AMD at four modules is good enough to compete with top of the line i5 & even beats it in certain cases, the power consumption however is still a major issue as I said above !

 

[mrmt]

3M Kaveri will get close to FX83xx performance wise (not FX6300 )

Should I quote you for this misprediction just like you do with others?

 

[NTMBK]

Like how everyone here drools at the idea of an eight core chip with HT from Intel yet they don't have any plan/roadmap to entertain these folks ? I can't see why such a chip isn't plausible given that the TDP doesn't exceed ~150W levels, with steamroller or maybe excavator cores, because AMD at four modules is good enough to compete with top of the line i5 & even beats it in certain cases, the power consumption however is still a major issue as I said above !

Intel sells eight core chips with Hyperthreading. They're the same chip as the 3930k, just with all 8 cores on the die enabled, and a Xeon sticker on the box.

It's not plausible because the die would be massive. 32-28nm doesn't give nearly enough room to make it viable.

And every roadmap about Kaveri has listed it as having 2-4 cores.

 

[Skolomerija]

What is Gustafson's law?

 

[R0H1T]

Intel sells eight core chips with Hyperthreading. They're the same chip as the 3930k, just with all 8 cores on the die enabled, and a Xeon sticker on the box.

It's not plausible because the die would be massive. 32-28nm doesn't give nearly enough room to make it viable.

And every roadmap about Kaveri has listed it as having 2-4 cores.

We don't know that yet & because its an unknown, its still in the realms of possibility, you can't rule that out completely. I'm not saying that Kaveri next year will indeed have three modules, perhaps its next revision will, but if its profitable for AMD & makes them more $ per die then it is plausible !

 

[Charles Kozierok]

Let me quote a part of this recent work (Bold face from mine):

http://www.sciencedirect.com/science/article/pii/S0743731509000884

The article proposes another generalization to Amdahl's law with speedups above those predicted by the Amdahl's law.

You quoted this from the article:

History seems to be repeating itself. Two decades ago, mainstream vendors, worried about the pessimistic implications of Amdahl's law, made parallel machines only with 2 to 8 processors, such as the IBM 7030 Stretch Data Processing System and Cray Y-MP.

Wow. Really?

You know, I have no idea who the people are that wrote this article, but given the number of errors packed into that one sentence, I'm not inclined to pay much attention to them. Let's see:

1. Mentioning the IBM 7030 and Cray Y-MP together as something designed two decades ago is bizarre. They were not contemporaries: the Y-MP was created in the late 1980s; the 7030 was created in the 1950s.
2. The two machines have almost nothing in common except for both being made with transistors. They aren't even in the same generation: the 7030 used discrete components and magnetic core memory; the Y-MP was IC-based with semiconductor memory. The Y-MP was also, oh, 1,000 times faster, not even including multiple CPUs?
3. The 7030 wasn't a parallel processor. It wasn't even a multiprocessor! It had only one CPU, not 2 to 8.
4. Cray Research was not holding back on the number of processors they put in their machines based on Amdahl's Law -- they doubled them every generation. The X-MP had 4 CPUs; the Y-MP had up to 8; its successor, the C90, had up to 16; and its successor, the T90, had up to 32.
5. Amdahl's Law was only coined in 1967, a decade after the IBM 7030 was designed. That would have made it difficult for it to make the 7030's creators pessimistic.

Hmm, I think that's all of them.

For the heck of it I glanced through the rest of the paper. It doesn't seem to be saying anything different than Gustafson's Law. IDC and I have already covered this ground: Amdahl's Law is talking about attempts to solve a particular problem more quickly, not about solving problems of indeterminate size.

 

[NTMBK]

We don't know that yet & because its an unknown, its still in the realms of possibility, you can't rule that out completely. I'm not saying that Kaveri next year will indeed have three modules, perhaps its next revision will, but if its profitable for AMD & makes them more $ per die then it is plausible !

Trinity already has a larger die than a 4 core Sandy Bridge, and a much larger die size than Ivy Bridge 4 core. (Numbers here if you're curious: http://www.anandtech.com/show/5831/amd-trinity-review-a10-4600m-a-new-hope ) Adding 50% more cores (which are also more complex, and will probably take up more die space than they used to) along with the bigger GPU on will blow that die size up, a lot. I just don't see it being profitable.

(And don't forget, AMD will use the same die for their 4C and 2C models, so the larger die will destroy profit margins on them too.)

 

[inf64]

Have we seen any confirmation of there being a 3-module Kaveri? The roadmaps I saw all said 4 cores i.e. 2 modules.





And I can't see any way that AMD could fit 6 Steamroller cores and a 7750 level GPU onto the same die without hitting Sandy Bridge E level of die size... This seems like wishful thinking which has been repeated so much that people assume it is true.

Nothing official but there is a document named "Preliminary BIOS and Kernel Developer's Guide for AMD Family 15h Models 30h-3Fh Processors" (under NDA I suppose) that has this page.
That's all we know. I guess it's possible that we will have 2M mainstream Kaveri and that 3M will be basic building block for server/high end desktop segment in 2014/2015.
http://www.brightsideofnews.com/Dat...ural-Enhancements-Unveiled/AMD_Kaveri_689.jpg

 

[Phynaz]

Isn't it funny how the thread topic was First complete review of Haswell i7-4770K yet it has morphed into another AMD vs Intel slugfest?

How many days until Intel "officially" releases Haswell ?

You read my mind.

 

[R0H1T]

Trinity already has a larger die than a 4 core Sandy Bridge, and a much larger die size than Ivy Bridge 4 core. (Numbers here if you're curious: http://www.anandtech.com/show/5831/amd-trinity-review-a10-4600m-a-new-hope ) Adding 50% more cores (which are also more complex, and will probably take up more die space than they used to) along with the bigger GPU on will blow that die size up, a lot. I just don't see it being profitable.

(And don't forget, AMD will use the same die for their 4C and 2C models, so the larger die will destroy profit margins on them too.)

Maybe you didn't see this ~
http://www.anandtech.com/show/6396/the-vishera-review-amd-fx8350-fx8320-fx6300-and-fx4300-tested

From the tables above I'll list the stats as following ~
AMD Trinity 32nm 1.303B 246mm2
AMD Vishera 32nm 1.2B 315mm2

So if I'm correct in my assumption then trinity uses something like 600~700 million transistors for the CPU part of the die, taking Vishera numbers as evidence, that means if they increase the current 246mm^2 die size by about a third then they could potentially add a single module or beef up the GPU at 28nm ! Not so much improbable as you're making it out be, granted it will cost in the range of their current FX line or possibly even more than that.

 

[R0H1T]

You read my mind.

Could it be that there are no "reliable" numbers out there for Haswell hence debating other things isn't such a bad idea in the meantime :whiste:

 

[bgt]

Trinity already has a larger die than a 4 core Sandy Bridge, and a much larger die size than Ivy Bridge 4 core.

Big dies are easier to cool:whiste:

 

[NTMBK]

Big dies are easier to cool:whiste:

No. The heat density of a 3 module Kaveri would be the same as a 2 module Kaveri, but the overall heat generated would be higher, making it harder to cool.

 

[galego]

We don't know that yet & because its an unknown, its still in the realms of possibility, you can't rule that out completely. I'm not saying that Kaveri next year will indeed have three modules, perhaps its next revision will, but if its profitable for AMD & makes them more $ per die then it is plausible !

From bsn

AMD Kaveri APU Processor will feature 4-6 Steamroller cores

http://www.brightsideofnews.com/new...core-architectural-enhancements-unveiled.aspx

 

[WhoBeDaPlaya]

Can we please leave the gorram Intel vs. AMD bickering out of this thread?
A lot of us have deployed AMD and Intel CPUs as appropriate and just want more info on Haswell.

 

[galego]

You quoted this from the article:

Wow. Really?

You know, I have no idea who the people are that wrote this article, but given the number of errors packed into that one sentence, I'm not inclined to pay much attention to them. Let's see:

1. Mentioning the IBM 7030 and Cray Y-MP together as something designed two decades ago is bizarre. They were not contemporaries: the Y-MP was created in the late 1980s; the 7030 was created in the 1950s.
2. The two machines have almost nothing in common except for both being made with transistors. They aren't even in the same generation: the 7030 used discrete components and magnetic core memory; the Y-MP was IC-based with semiconductor memory. The Y-MP was also, oh, 1,000 times faster, not even including multiple CPUs?
3. The 7030 wasn't a parallel processor. It wasn't even a multiprocessor! It had only one CPU, not 2 to 8.
4. Cray Research was not holding back on the number of processors they put in their machines based on Amdahl's Law -- they doubled them every generation. The X-MP had 4 CPUs; the Y-MP had up to 8; its successor, the C90, had up to 16; and its successor, the T90, had up to 32.
5. Amdahl's Law was only coined in 1967, a decade after the IBM 7030 was designed. That would have made it difficult for it to make the 7030's creators pessimistic.

Hmm, I think that's all of them.

For the heck of it I glanced through the rest of the paper. It doesn't seem to be saying anything different than Gustafson's Law. IDC and I have already covered this ground: Amdahl's Law is talking about attempts to solve a particular problem more quickly, not about solving problems of indeterminate size.

You take a pair of lines from an one-page historical introduction, which if eliminated from the article do not change the main body of the article, and decide to criticize that, whereas you fail to provide any real argument against the main content on the article including the equation that they derive, which is:

a generalization of Amdahl’s law and Gustafson’s law

In any case, here go some comments about your revision of the historical intro:

• The first 7030 was sold in 1961. The first Y-MP in the 1988. That is two decades: 60s vs 80s.
• The original 7030 had only a CPU (the IBM 7101), but one year latter IBM added the IBM 7951. This count to me as two processors. The authors of the article do not refer to CPUs. You do.
• The Y-MP had between 2 and 8 processors only.
• And as they note in the historical introduction it was after the introduction "of the scalable computing concept in 1988" (i.e. just after the Y-MP was designed), that supercomputers with large number of processors appeared.

Regarding Haswell:

Thermaltake Announces PSUs Ready for Intel Haswell CPUs

http://benchmarkreviews.com/index.php?option=com_content&task=view&id=22957&Itemid=47

And some prices for Haswell: the i7-4770k @ $368

http://www.itpro.co.uk/desktop-hardware/19655/intel-haswell-release-date-and-specs

 

[guskline]

Can we please leave the gorram Intel vs. AMD bickering out of this thread?
A lot of us have deployed AMD and Intel CPUs as appropriate and just want more info on Haswell.

Agreed.

 

[Phynaz]

Could it be that there are no "reliable" numbers out there for Haswell hence debating other things isn't such a bad idea in the meantime :whiste:

Then we can talk about Haswell and not make it another tiresome Intel vs AMD fanboy thread.

 

[Lepton87]

No. The heat density of a 3 module Kaveri would be the same as a 2 module Kaveri, but the overall heat generated would be higher, making it harder to cool.

Of course bigger ICs are easier to cool, provided that they use the same amount of power. If die size increase is commensurate with the increase in power consumption then they are indeed harder to cool. Also we don't know if 3 module Kaveri's heat density would be the same as 2 module, we could make that assumption if Kaveri didn't have an integrated GPU. CPU part of Kaveri might have higher power density then GPU part that would make the average heat density higher or it could be the other way around but I really doubt it.

 

[tweakboy]

Amazing graphics,, gameplay story line,, cry3 engine pownz us..

 

[Remobz]

Then we can talk about Haswell and not make it another tiresome Intel vs AMD fanboy thread.

How long before games start taking advantage of 6 cores?

Maybe wait and buy a Haswell until games really take advantage of it?

 

[Phynaz]

How long before games start taking advantage of 6 cores?

Maybe wait and buy a Haswell until games really take advantage of it?

If you're already on SB/IB and your purpose for buying a CPU is to play games then yes, you should wait.

 

[Remobz]

If you're already on SB/IB and your purpose for buying a CPU is to play games then yes, you should wait.

Sorry I should of been more clear.

I have an Intel i3 2120 and I do nothing CPU intensive. I hope to play BF3 and BF4 when I get a new PSU and computer case to complete my build.

Honestly, thinking about it now I do NOTHING CPU intensive other than wanting to play new and old games on the market.

Might as well wait indeed.