Haswell model specs leaked

Page 11 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Homeles

Platinum Member
Dec 9, 2011
2,580
0
0
VRM.

Anand Shimpi has explicitly stated that embedded DRAM will be coming to at least one flavor of Haswell, so I've chalked that one up to being a done deal:
Now to the things that Intel didn't let loose at IDF. Although originally an option for Ivy Bridge (but higher ups at Intel killed plans for it) was a GT3 part with some form of embedded DRAM. Rumor has it that Apple was the only customer who really demanded it at the time, and Intel wasn't willing to build a SKU just for Apple.
Haswell will do what Ivy Bridge didn't. You'll see a version of Haswell with up to 128MB of embedded DRAM, with a lot of bandwidth available between it and the core. Both the CPU and GPU will be able to access this embedded DRAM, although there are obvious implications for graphics.
http://www.anandtech.com/show/6355/intels-haswell-architecture/12
 

Fjodor2001

Diamond Member
Feb 6, 2010
3,989
440
126
@ShintaiDK: Did you read that? Previously you've said that eDRAM wouldn't even come in Broadwell. :awe:
 

cbn

Lifer
Mar 27, 2009
12,968
221
106
HD4000->GT2 is expected to be less improvement than HD3000->HD4000.

Can you post the source?

HD3000 = 12 EUs, 850 Mhz base clocks with 1.35 Ghz turbo -->http://ark.intel.com/products/52214/Intel-Core-i7-2600K-Processor-8M-Cache-up-to-3_80-GHz
HD4000 = 16 EUs, 650 Mhz base clocks with 1.15 Ghz turbo -->http://ark.intel.com/products/65523
HD4600? (GT2 for Haswell) = 20 EUs, Clocks??

Clocks shouldn't drop on the Haswell GT2, so I would actually expect a greater increase going from Ivy Bridge GT2 (HD4000) to Haswell GT2 (HD4600??).

With that said, I do have to wonder how memory bandwith will factor in.
 

ShintaiDK

Lifer
Apr 22, 2012
20,378
145
106
VRM.

Anand Shimpi has explicitly stated that embedded DRAM will be coming to at least one flavor of Haswell, so I've chalked that one up to being a done deal:

http://www.anandtech.com/show/6355/intels-haswell-architecture/12

@ShintaiDK: Did you read that? Previously you've said that eDRAM wouldn't even come in Broadwell. :awe:

If you read its a pure guess by Anand. (And its 2 months old for that matter.)

As for why Intel isn't talking about embedded DRAM on Haswell, your guess is as good as mine.

Sofar nothing points to that Haswell or Broadwell should have any eDRAM, besides rumours. But again, first it was a L4 cache...
 
Last edited:

Fjodor2001

Diamond Member
Feb 6, 2010
3,989
440
126
Can you post the source?

HD3000 = 12 EUs, 850 Mhz base clocks with 1.35 Ghz turbo -->http://ark.intel.com/products/52214/Intel-Core-i7-2600K-Processor-8M-Cache-up-to-3_80-GHz
HD4000 = 16 EUs, 650 Mhz base clocks with 1.15 Ghz turbo -->http://ark.intel.com/products/65523
HD4600? (GT2 for Haswell) = 20 EUs, Clocks??

Clocks shouldn't drop on the Haswell GT2, so I would actually expect a greater increase going from Ivy Bridge GT2 (HD4000) to Haswell GT2 (HD4600??).

With that said, I do have to wonder how memory bandwith will factor in.

You're missing a central piece: IB brought a new iGPU arch. So EUs and frequencies cannot be directly compared with HD3000.

HD3000->HD4000 brought around 30-50% increase (http://www.anandtech.com/show/5771/the-intel-ivy-bridge-core-i7-3770k-review/9 and onwards). HD4000->GT2 is expected to bring ~30% increase last I heard, but I could not find any source for that now. If you find some source confirming or denying that, please let me know.
 

cbn

Lifer
Mar 27, 2009
12,968
221
106
You're missing a central piece: IB brought a new iGPU arch. So EUs and frequencies cannot be directly compared with HD3000.

HD3000->HD4000 brought around 30-50% increase (http://www.anandtech.com/show/5771/the-intel-ivy-bridge-core-i7-3770k-review/9 and onwards). HD4000->GT2 is expected to bring ~30% increase last I heard, but I could not find any source for that now. If you find some source confirming or denying that, please let me know.

Thank you.

So if Haswell GPU is fundamentally the same architecture as Ivy Bridge and we are looking at a 25% in EUs. (Up from 16 in IB to 20 in Haswell) then would have to see an increase in GPU clocks on Haswell to get a similar jump as we did for SB--> IV.

http://www.anandtech.com/show/6355/intels-haswell-architecture/12

Haswell builds on the same fundamental GPU architecture we saw in Ivy Bridge. We won't see a dramatic redesign/re-plumbing of the graphics hardware until Broadwell in 2014 (that one is going to be a big one).

P.S. How does you or anyone else feel about DRAM speed at the 20EU level? I haven't done much reading on the Llano and Trinity APUs but can remember seeing a lot of comments about those processors being affected by slow DRAM (on the GPU side). I would assume that as Intel begins to scale its GPU it could run into the same issue.
 

cbn

Lifer
Mar 27, 2009
12,968
221
106

mikk

Diamond Member
May 15, 2012
4,234
2,292
136
Your first screenshot doesn't make any sense because memory scaling can change from game to game. I have seen other benchmarks where higher DDR3 speeds to 1866 or 2133 resulted in a nice bump. Regarding GT2 performance improvements, cpu-world reported 15-25% faster than GT2 Ivy Bridge. This should be accurate. I expect another 5% for both Ivy Bridge and Haswell due to a better driver stack (15.31). HD4000 performs almost 50% faster than HD3000 at the same frequency (in GPGPU benchmarks more). Since Haswell graphics contains only a slightly tweaked Gen7, this is not possible. GT3 is another matter though due to EU and slice doubling. Broadwell with Gen8 and bigger uarch changes should bring us a bigger iGPU jump, not really a surprise from the Intel engineer.
 

Fjodor2001

Diamond Member
Feb 6, 2010
3,989
440
126
Your first screenshot doesn't make any sense because memory scaling can change from game to game. I have seen other benchmarks where higher DDR3 speeds to 1866 or 2133 resulted in a nice bump. Regarding GT2 performance improvements, cpu-world reported 15-25% faster than GT2 Ivy Bridge. This should be accurate. I expect another 5% for both Ivy Bridge and Haswell due to a better driver stack (15.31). HD4000 performs almost 50% faster than HD3000 at the same frequency (in GPGPU benchmarks more). Since Haswell graphics contains only a slightly tweaked Gen7, this is not possible. GT3 is another matter though due to EU and slice doubling. Broadwell with Gen8 and bigger uarch changes should bring us a bigger iGPU jump, not really a surprise from the Intel engineer.

Yes, but how should Broadwell be able to deliver significantly better GFX performance without solving the memory bandwidth problem?

IB is already bottlenecked in some cases now. Haswell GT2 will increase performance by another 20-25% or so. Then Broadwell is expected to be a huge jump according to both Anand and the Intel engineer. To me that sounds like at least a 50-100% performance increase compared to Haswell. How can that be possible if the memory bandwidth problem is not solved?
 

mikk

Diamond Member
May 15, 2012
4,234
2,292
136
Bottlenecked to some degree, but not that much. Have a look to Trinity Desktop. Even with slow dualchannel DDR3 memory, performance can be doubled with faster graphics. We don't know Intel's plan for Broadwell's memory subsystem, it might get some stacked memory for the faster variants. Too early yet. It will be interesting to see how GT3 Quadcores (without the edram) are performing compared to GT2.
 

Fjodor2001

Diamond Member
Feb 6, 2010
3,989
440
126
Bottlenecked to some degree, but not that much. Have a look to Trinity Desktop. Even with slow dualchannel DDR3 memory, performance can be doubled with faster graphics. We don't know Intel's plan for Broadwell's memory subsystem, it might get some stacked memory for the faster variants. Too early yet. It will be interesting to see how GT3 Quadcores (without the edram) are performing compared to GT2.

But isn't the GT3 supposed to be downclocked? I.e. it's primary purpose is to save power, not to increase performance? So they have 40 EUs to be able to clock them low to save power while still providing decent GFX performance.

Also, with Broadwell Intel will have twice the amount of transistors on the same die area compared to IB/Haswell. We already know whey will not increase the CPU core count. So what could they use the additional transistors for? More EUs, eDRAM, anything else?
 

ShintaiDK

Lifer
Apr 22, 2012
20,378
145
106
But isn't the GT3 supposed to be downclocked? I.e. it's primary purpose is to save power, not to increase performance? So they have 40 EUs to be able to clock them low to save power while still providing decent GFX performance.

Also, with Broadwell Intel will have twice the amount of transistors on the same die area compared to IB/Haswell. We already know whey will not increase the CPU core count. So what could they use the additional transistors for? More EUs, eDRAM, anything else?

You might see the VRM ondie instead of onpackage. Plus there will be a free transistor overhead for Skylake to use.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,786
136
VRM isn't coming on-die with Broadwell, its still very much in research phase. If anything, its coming with Tock generation too, not Tick.

Yes, but how should Broadwell be able to deliver significantly better GFX performance without solving the memory bandwidth problem?

Because from Haswell, it'll get on-package memory? I bet with Broadwell, mobile parts getting the better iGPU will be done even more extensively. So we might see GT4 with select mobile Broadwell chips while desktop still ends up with GT2.

Yes, but not low lower TDP (for the desktop CPUs).

Of course. Desktop users would cry if Intel cut clocks and performance further to lower TDP. And like I've said before, in the non-SFF/AIO market, only the enthusiast portion is growing. So there's no point for them to lower TDP.

Same with graphics.

I'm finding more and more every day how "entrenched" in enthusiast desktop Anandtech forums are. Desktop is less than 35% of volume for Intel on PCs. So while Desktop changes are unimpressive(which is a side effect of Intel focusing on mobile chips), that's not the point of Haswell.

If you look at it in terms of a mobile point of view, Haswell should dwarf what Ivy Bridge brought.

better integration

On package VRM and memory. That's BIG. Again nothing for desktop because it will either a) not have it b) the gains are insignificant

IPC: A mere 3% for Ivy Bridge versus 10%+ for Haswell plus AVX2 and TSX. It should be even more important for why IPC is a lot better gain than little IPC + clocks. Because there's overclocking. And you can do voltage/clock play to significantly lower while having same performance in other areas if Intel wishes to.
 
Last edited:

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,786
136
Also, with Broadwell Intel will have twice the amount of transistors on the same die area compared to IB/Haswell. We already know whey will not increase the CPU core count. So what could they use the additional transistors for? More EUs, eDRAM, anything else?

I can guess.

So Intel increases GT2's EU count by 20-25% every generation:

8-10-12-12-16-20

GT3 doubles GT2's EU, and GT4 probably doubles GT3's EU count.

40 EUs(in Haswell GT3) x 1.2 x 2 = 96 EUs

Yeah, I can see that alone taking up all the transistor count increase. Since its a new arch you will probably see enhancements in other areas as well.

Computer Bottleneck said:
HD3000 = 12 EUs, 850 Mhz base clocks with 1.35 Ghz turbo -->http://ark.intel.com/products/52214/...up-to-3_80-GHz
HD4000 = 16 EUs, 650 Mhz base clocks with 1.15 Ghz turbo -->http://ark.intel.com/products/65523
HD4600? (GT2 for Haswell) = 20 EUs, Clocks??

Ivy Bridge's EU has a peak FLOPs rate of 2x Sandy Bridge per clock, and ~1.5x sustained. Along with 33% increase in EUs, that corresponds to 2x increase in sustained performance. That's because Ivy Bridge significantly increases co-issuing of instructions. Haswell keeps that part the same.
 
Last edited:

Fjodor2001

Diamond Member
Feb 6, 2010
3,989
440
126
I can guess.

So Intel increases GT2's EU count by 20-25% every generation:

8-10-12-12-16-20

GT3 doubles GT2's EU, and GT4 probably doubles GT3's EU count.

40 EUs(in Haswell GT3) x 1.2 x 2 = 96 EUs

Yeah, I can see that alone taking up all the transistor count increase. Since its a new arch you will probably see enhancements in other areas as well.

Yes, that could be a possibility. But going from 20->96 EUs without any solution to the memory bandwidth problem does not seem like a balanced design. So perhaps there is some truth to the eDRAM rumor after all, although some people here does not believe in that.

Also, it's interesting to note that a 96 EU chip with 4 CPU cores would probably allocate something like 70% of the die area to the iGPU, 20% to the CPU cores (including caches), and 10% to the rest like memory controller.

Then you're no longer buying a CPU with iGPU. You're buying a GPU with iCPU...
 

mikk

Diamond Member
May 15, 2012
4,234
2,292
136
But isn't the GT3 supposed to be downclocked? I.e. it's primary purpose is to save power, not to increase performance? So they have 40 EUs to be able to clock them low to save power while still providing decent GFX performance.

Also, with Broadwell Intel will have twice the amount of transistors on the same die area compared to IB/Haswell. We already know whey will not increase the CPU core count. So what could they use the additional transistors for? More EUs, eDRAM, anything else?


For ULT models with GT3 yes, because 15W envelope is a big limitation there. I don't expect much faster ULT models, Anandtech wrote 30% which sounds realistic. The bigger jump will be on the bigger 47/57W GT3 models side. CPU-world wrote 50-100% which is based on Intel estimates. In best case a doubling compared to GT2 Ivy Bridge. We will see.
 

bunnyfubbles

Lifer
Sep 3, 2001
12,248
3
0
the 8 core Xeon E5-2690 is 2.9GHz base 3.8GHz Turbo and is 135W TDP

while 8 core E5-2687W is 3.1GHz base 3.8GHz Turbo and is 150W TDP

and then there's the 3970X which recently established a 150W TDP for the i7s, however I think that might have just been paving the way for an 8 core part as the 70X is little more relative to the 130W TDP 60X

that being said, unless Intel ups the TDP, I don't see them squeezing a 3.4-3.5GHz and 4-4.1GHz Turbo 8 core part into 150W, so I'd expect clockrates to be more in line with the 2687W, which really wouldn't matter because the likely buyers will just overclock the chip anyway.

The real kicker will be price. Coming in at the standard ~$1000 mark would seem too good to be true as the closest companion 8 core Xeons are already running ~$2000 and there's still the 3970X @ $1000...

At any rate I'd probably be interested @ no more than $1200, but if its $1000 I'll be a first day adopter most likely.
 

Fjodor2001

Diamond Member
Feb 6, 2010
3,989
440
126
the 8 core Xeon E5-2690 is 2.9GHz base 3.8GHz Turbo and is 135W TDP

while 8 core E5-2687W is 3.1GHz base 3.8GHz Turbo and is 150W TDP

and then there's the 3970X which recently established a 150W TDP for the i7s, however I think that might have just been paving the way for an 8 core part as the 70X is little more relative to the 130W TDP 60X

that being said, unless Intel ups the TDP, I don't see them squeezing a 3.4-3.5GHz and 4-4.1GHz Turbo 8 core part into 150W, so I'd expect clockrates to be more in line with the 2687W, which really wouldn't matter because the likely buyers will just overclock the chip anyway.

The real kicker will be price. Coming in at the standard ~$1000 mark would seem too good to be true as the closest companion 8 core Xeons are already running ~$2000 and there's still the 3970X @ $1000...

At any rate I'd probably be interested @ no more than $1200, but if its $1000 I'll be a first day adopter most likely.

The funny thing is that they could probably make an 8 core Haswell without iGPU, and it wouldn't occupy much more die area than a 3570K (perhaps 30% more). So if CPUs were priced based on die area alone, it could probably be sold for $300-400 or so.

But then there's the marketing aspect of this unfortunately that stops this from happening... :\
 
Last edited:
sale-70-410-exam    | Exam-200-125-pdf    | we-sale-70-410-exam    | hot-sale-70-410-exam    | Latest-exam-700-603-Dumps    | Dumps-98-363-exams-date    | Certs-200-125-date    | Dumps-300-075-exams-date    | hot-sale-book-C8010-726-book    | Hot-Sale-200-310-Exam    | Exam-Description-200-310-dumps?    | hot-sale-book-200-125-book    | Latest-Updated-300-209-Exam    | Dumps-210-260-exams-date    | Download-200-125-Exam-PDF    | Exam-Description-300-101-dumps    | Certs-300-101-date    | Hot-Sale-300-075-Exam    | Latest-exam-200-125-Dumps    | Exam-Description-200-125-dumps    | Latest-Updated-300-075-Exam    | hot-sale-book-210-260-book    | Dumps-200-901-exams-date    | Certs-200-901-date    | Latest-exam-1Z0-062-Dumps    | Hot-Sale-1Z0-062-Exam    | Certs-CSSLP-date    | 100%-Pass-70-383-Exams    | Latest-JN0-360-real-exam-questions    | 100%-Pass-4A0-100-Real-Exam-Questions    | Dumps-300-135-exams-date    | Passed-200-105-Tech-Exams    | Latest-Updated-200-310-Exam    | Download-300-070-Exam-PDF    | Hot-Sale-JN0-360-Exam    | 100%-Pass-JN0-360-Exams    | 100%-Pass-JN0-360-Real-Exam-Questions    | Dumps-JN0-360-exams-date    | Exam-Description-1Z0-876-dumps    | Latest-exam-1Z0-876-Dumps    | Dumps-HPE0-Y53-exams-date    | 2017-Latest-HPE0-Y53-Exam    | 100%-Pass-HPE0-Y53-Real-Exam-Questions    | Pass-4A0-100-Exam    | Latest-4A0-100-Questions    | Dumps-98-365-exams-date    | 2017-Latest-98-365-Exam    | 100%-Pass-VCS-254-Exams    | 2017-Latest-VCS-273-Exam    | Dumps-200-355-exams-date    | 2017-Latest-300-320-Exam    | Pass-300-101-Exam    | 100%-Pass-300-115-Exams    |
http://www.portvapes.co.uk/    | http://www.portvapes.co.uk/    |