Haswell model specs leaked

[boxleitnerb]

http://chinese.vr-zone.com/43824/in...luding-4770k-4670k-for-overclocking-12112012/

84W TDP, no increase in clocks, not even a single MHz
Call me disappointed, very disappointed.

 

[NTMBK]

I'm dubious- the model numbers are pretty much identical to IB with 4xxx instead of 3xxx, and they have matching clocks. Seems a bit too close to be true.

If it is true, then Haswell graphics are going to be seriously bandwidth starved. No wonder they needed Crystalwell.

 

[Saylick]

I am suspicious about that TDP too. I was expecting something more or less in line with IVB's TDP, especially considering that Intel's had plenty of time to optimize the power efficiency for Haswell AND these specs indicate the same clock speed. From what I recall, the desktop SKUs will be running GT2 iGPUs which contain modified version's of IVB's shaders, so I do not expect a 7W increase solely due to the iGPU.

 

[FreshJR]

I think its a little sketchy. Currently ivy bridges TDP is 17 - 77w depending on the model.
Intel is all about power efficiency with Haswell, so why did the range jump up to 35-88w?

Source: http://ark.intel.com/products/family/65504

The lower power versions were already rumored to be 7-10 watts I thought (for laptops).

Although clock speed != performance so I'm still excited either way, I hope they don't drop the ball on this one.

Does anyone know if 2013 A,B,C,D mean 1rst, 2nd, etc quarter?

 

[NTMBK]

I am suspicious about that TDP too. I was expecting something more or less in line with IVB's TDP, especially considering that Intel's had plenty of time to optimize the power efficiency for Haswell AND these specs indicate the same clock speed. From what I recall, the desktop SKUs will be running GT2 iGPUs which contain modified version's of IVB's shaders, so I do not expect a 7W increase solely due to the iGPU.

They've added a lot of hardware to the cores themselves. AVX2, more ports, etc. I'm not overly surprised by a TDP rise, but there should be IPC rises too.

I think its a little sketchy. Currently ivy bridges TDP is 17 - 77w depending on the model.

Source: http://ark.intel.com/products/family/65504

Intel is all about power efficiency with Haswell.

Why would we get such high TDP products without a performance increase, and also no low TDP products available for purchase. Although clock speed != performance. Im still excited either way, I hope they dont drop the ball on this one.

These seem to only be desktop SKUs, so I'm not surprised to see no 17W models.

 

[inf64]

It's not a high TDP and you guys forget you get better performance at the same clock with Haswell, both on CPU and GPU parts of the core. And why do you think GPU will be BW starved? They have the same fast 8MB L3 cache connected to memory (just as in IB's case;GPU in IB does not scale with DDR3 speeds that much if at all IIRC,unlike Llano or Trinity's GPUs).

So 9% higher TDP for 10-15% more CPU performance and 20-50%(?) better GPU performance. This equates to better perf./watt than IB.

 

[Lepton87]

http://chinese.vr-zone.com/43824/in...luding-4770k-4670k-for-overclocking-12112012/

84W TDP, no increase in clocks, not even a single MHz
Call me disappointed, very disappointed.

I don't know why people expected higher clocks with haswell, it's a WIDER core with more EXECUTIONS units build on the SAME process. And IPC is supposed to be improved. If they wanted to deliver higher clocks they would have to deepen the pipeline which would make it really hard to increase IPC.

 

[boxleitnerb]

I don't know why people expected higher clocks with haswell, it's a WIDER core with more EXECUTIONS units build on the SAME process. And IPC is supposed to be improved. If they wanted to deliver higher clocks they would have to deepen the pipeline which would make it really hard to increase IPC.

Sandy Bridge improved IPC AND clocks. Ivy Bridge didn't improve much, barely at all. I read somewhere that Ivy was not fully optimized for the 22nm process but Haswell is. So I expected Intel to catch up a bit.

 

[FreshJR]

I forgot about the fact that they moved many of the motherboard components onto the chip itself.

The higher TDP makes sense now, and yea clockspeed != performance.

Think about it. Would you rather have a 3.6ghz pentium 4 or a 1.6 ghz i3? We will just have to wait a see. I'm expecting amazing results since Haswell is a TOCK, on intels timeline.

 

[boxleitnerb]

Oh please, no one is talking about these extremes.

10% higher IPC is okay, but CPU-wise one would expect a total of at least 20% higher performance as a product of IPC and clocks. We have seen that with Nehalem/Lynnfield and with Sandy Bridge. Compared to that, Ivy and Haswell are just disappointing.

 

[OBLAMA2009]

those clockspeeds are going to be a serious marketing problem. they pretty much hit the wall two years ago

 

[NTMBK]

And why do you think GPU will be BW starved? They have the same fast 8MB L3 cache connected to memory (just as in IB's case;GPU in IB does not scale with DDR3 speeds that much if at all IIRC,unlike Llano or Trinity's GPUs).

The reason IB's graphics doesn't scale much with memory is because it's terrible. It doesn't take much bandwidth to feed the EUs if there aren't many of them and they aren't doing much work. Given that Haswell is meant to be a big jump in graphics performance, that's not going to hold any more. If they're aiming to match or beat Trinity graphics performance, then they're going to need more bandwidth. The L3 cache will help, but not that much.

 

[amsterdamxxx]

There is no competition at this stage, AMD are dead in the water so we get stagnant performance increases.

 

[Lepton87]

Sandy Bridge improved IPC AND clocks. Ivy Bridge didn't improve much, barely at all. I read somewhere that Ivy was not fully optimized for the 22nm process but Haswell is. So I expected Intel to catch up a bit.

Sandybridge clocks just a tiny bit better then westmere, probably because it has deeper pipeline, but it does not have more executions units.

 

[inf64]

The reason IB's graphics doesn't scale much with memory is because it's terrible. It doesn't take much bandwidth to feed the EUs if there aren't many of them and they aren't doing much work. Given that Haswell is meant to be a big jump in graphics performance, that's not going to hold any more. If they're aiming to match or beat Trinity graphics performance, then they're going to need more bandwidth. The L3 cache will help, but not that much.

I'm sure reviewers will use different memory speeds in their tests so we can see how new GPU scales with them . I'm not convinced we will see some huge gains with higher memory clocks. The main point is that GPU will be decently faster while core is being done on the same node as IB.

 

[boxleitnerb]

Sandybridge clocks just a tiny bit better then westmere, probably because it has deeper pipeline, but it does not have more executions units.

If we look at the desktop, the fastest CPU of the performance segment was the i7-880 at 3.06 GHz. The 2600K, its successor, clocked 10% higher and had 10-15% higher IPC, too.

My point is, we haven't seen these seizable jumps since Sandy Bridge.

 

[inf64]

If we look at the desktop, the fastest CPU of the performance segment was the i7-880 at 3.06 GHz. The 2600K, its successor, clocked 10% higher and had 10-15% higher IPC, too.

My point is, we haven't seen these seizable jumps since Sandy Bridge.

i7 880? What happened to lga1366 CPUs? You had i7 975 which was 3.33/3.6Ghz 4C/8T part ,very close to 3.4/3.8Ghz on SB. Performance wise it was on the tail of 2600K as you can see here(SB was 6% faster in apps and 7% faster in games,overall).

 

[Lepton87]

If we look at the desktop, the fastest CPU of the performance segment was the i7-880 at 3.06 GHz. The 2600K, its successor, clocked 10% higher and had 10-15% higher IPC, too.

My point is, we haven't seen these seizable jumps since Sandy Bridge.

There was a westmere Xeon at 4.5GHz, too.

 

[NTMBK]

I'm sure reviewers will use different memory speeds in their tests so we can see how new GPU scales with them . I'm not convinced we will see some huge gains with higher memory clocks. The main point is that GPU will be decently faster while core is being done on the same node as IB.

There's already definite gains to faster memory on Trinity- it's graphics performance will scale with memory bandwidth as far as DDR3-2133, whereas IB's graphics seems bandwidth saturated at ~DDR3-1600.

 

[boxleitnerb]

i7 880? What happened to lga1366 CPUs? You had i7 975 which was 3.33/3.6Ghz 4C/8T part ,very close to 3.4/3.8Ghz on SB. Performance wise it was on the tail of 2600K as you can see here(SB was 6% faster in apps and 7% faster in games,overall).

They belong to a different segment. Back then Intel started to split their enthusiast and performance lines up like it is today. TDP and price differ quite a bit on the 975 and the 880. You would have to compare socket 1366 CPUs with socket 2011 CPUs in the same price and TDP range.

There was a westmere Xeon at 4.5GHz, too.

And that is relevant for the average consumer, how?

 

[jones377]

The higher TDP can be explained by the higher IPC nature of the core, but most importantly by FMA3 which doubles float throughput. This will come at a power cost. I would expect Haswell to use less power than Ivy Bridge when running legacy binaries (anything compiled for IVB or earlier) and only come close to hitting TDP when running something like Linpack compiled with FMA3, while at the same time getting roughly double the FLOPS. 7 extra watts for twice the FLOPS, not a bad trade off.

It was the same with Bulldozer. I seem to remember people getting ~20W higher power consumption when running Linpack with FMA4 vs the old SSEx binaries and getting ~2x the FLOPS.

At least clockspeeds didn't regress with Haswell, but why oh why does Intel insist in disabling VT-d on the K models?!?!?

 

[inf64]

@ NTMBK
Yeah my point was that IB didn't see much benefit from higher DDR3 clocks(unlike Trinity which was Radeon based). Is intel officially supporting higher specced DDR3 modules with Haswell(compared to IB)?

They belong to a different segment. Back then Intel started to split their enthusiast and performance lines up like it is today. TDP and price differ quite a bit on the 975 and the 880. You would have to compare socket 1366 CPUs with socket 2011 CPUs in the same price and TDP range.

I'm aware of the split in 2 segments intel did back then,but what was the price of 2600K at launch(300$?) and what was the price of top i7s for 2 other(older) sockets(965/975 and 880)? IIRC they were all cheaper than 2600K and perf./$ increase when going from either of them to new SB was not that large(look at the hardware.fr chart I linked).

 

[boxleitnerb]

The 965X and 975X had an MSRP of $999, the 880 was at $583 (they dropped the $583 performance SKU and stayed at $294-ish since then). The 2600K was much cheaper compared to these.

 

[Lepton87]

They belong to a different segment. Back then Intel started to split their enthusiast and performance lines up like it is today. TDP and price differ quite a bit on the 975 and the 880. You would have to compare socket 1366 CPUs with socket 2011 CPUs in the same price and TDP range.



And that is relevant for the average consumer, how?

It shows the potential of the silicon. SB didn't really improve much in terms of the clocks. Remember that only 6 cores where build using westmere architecture(Xeons were exceptions), apart from early dual cores with immature process. You know how extensive validation Xeons get? So comparing i7 880 on 45nm to SB on 32nm is pointless because IVY and Haswell AFAIK will use the same process.

 

[inf64]

The 965X and 975X had an MSRP of $999, the 880 was at $583 (they dropped the $583 performance SKU and stayed at $294-ish since then). The 2600K was much cheaper compared to these.

Ah yes it was because of the Extreme/X model number,you are correct. But still around the same clocks the increase was not stellar ,the main difference between the platforms was the (relatively fast) iGPU in SB core which was a big plus.