Broadwell inst a dramatic IPC improvement

[zappian]

Its just a step in power efficiency , intel is trying to get into the mobile market with their 14nm manufacturing process.

 

It isnt however that interesting for hardcore pc gamers.

 

I dont need less power consumption myself , my cpu is rated at 54w.

 

http://wccftech.com/intel-14nm-broadwell-cpu-architecture-analyzed-5-ipc-increase-haswell-2nd-generation-fivr-20-compute-units/
 

[SeanBond]

I don't think this will surprise anyone. It's been a 5% increase the last few generations.

[Archangel1994]

Nobody expected a drastic IPC improvement, it's a tick (die shrink) not a tock (architectural change). An IPC improvement of 5% is still nice for people that are still on sandy/ivy/westmere or older

[cesrai]

And it's to push the iGPU hard very hard, it's will know and skylake will have better IPC.

[Dabombinable]

Nobody expected a drastic IPC improvement, it's a tick (die shrink) not a tock (architectural change). An IPC improvement of 5% is still nice for people that are still on sandy/ivy/westmere or older

Harpertown and Wolfdale are examples of where a Broadwell CPU will make a massive difference. Considering an X5450 has only half the power of an i5 4440 whilst consuming enormous amounts of power, as well as producing loads of heat.

[harrynowl]

Harpertown and Wolfdale are examples of where a Broadwell CPU will make a massive difference. Considering an X5450 has only half the power of an i5 4440 whilst consuming enormous amounts of power, as well as producing loads of heat.

I'd argue Lynnfield/Arrandale as well. 

 

Harpertown was the server codename, Yorkfield was desktop.

[patrickjp93]

Broadwell is and isn't a dramatic improvement. With TSX-NI to work in upper SKUs, we'll see some Multithreading improvement over Haswell. Also, floating point calculations take 40% less time due to the shrink shaving the FPU instruction time from 5 cycles to 3.

But in integer where almost everything was already 1 cycle per instruction even for vectors, you're not going to see big IPC increases anymore without more advanced compilers and even wider instructions (like 512-bit AVX for Skylake Xeons).

Serial processing is reaching its limits. Instruction level parallelism already has reached its limits. We need to dive into data-level parallelism and just write better code.

[Analog]

Its just a step in power efficiency , intel is trying to get into the mobile market with their 14nm manufacturing process.

 

It isnt however that interesting for hardcore pc gamers.

 

I dont need less power consumption myself , my cpu is rated at 54w.

 

http://wccftech.com/intel-14nm-broadwell-cpu-architecture-analyzed-5-ipc-increase-haswell-2nd-generation-fivr-20-compute-units/

 

 

Not sure if you needed to make a topic on this, it's like common knowledge and has been said hundreds of times over and over. 

[Speedbird]

(not) Big surprise. It was the same thing with Ivy Bridge.

[Django]

Broadwell is and isn't a dramatic improvement. With TSX-NI now working in the U SKUs and above, we'll see some Multithreading improvement over Haswell

Are there some BDW-U with TSX? I haven't seen them on the Intel Ark and for the i7-5500U I have played with it is also not available according to HWiNFO64.

 

[patrickjp93]

Are there some BDW-U with TSX? I haven't seen them on the Intel Ark and for the i7-5500U I have played with it is also not available according to HWiNFO64.

I think it's reserved for the quad cores (M, H, and X) since the bug was found so late into Core M's manufacturing and the U-series ramp up.

[FellTheSky]

 i dont think we are going to see a jump like sandy bridge again, also intel has no need, since amd is so far behind

[patrickjp93]

i dont think we are going to see a jump like sandy bridge again, also intel has no need, since amd is so far behind

Haswell was another Sandy jump. The problem is software has fallen so far behind. Serial processing has plateaued, and code needs to be more parallel.

[TheFalseHuman]

Wait, I thought only the mobile Broadwell chips are out, not the desktop (or higher) versions... Or are the desktop versions already announced/released?

[Sintezza]

Not much of an improvement on ipc, is indeed something i allready knew

[FellTheSky]

Haswell was another Sandy jump. The problem is software has fallen so far behind. Serial processing has plateaued, and code needs to be more parallel.

 

haswell was not another sandy bridge jump, i dont think you are quite aware of how much of a jump sandy bridge was.

 

http://www.bit-tech.net/hardware/2013/06/12/intel-core-i5-4670k-haswell-cpu-review/6

 

look at the i5's all very close to each other. To put some perspective. 5ghz sandy = 4.7ghz ivy = 4.4ghz haswell approximately

 

In the other hand, look at the bloomfield i7 920, poor bastard doesnt have a chance against the 2500k, let alone the 2600k. Also, its practically impossible to overclock it to sandy bridge lvls.

The 2500k was superior to the 920 in every aspect, including price (220$ vs 300$). For haswell to be such a jump the 4670k should have beated the 2600k the same way the 2500k did

[patrickjp93]

haswell was not another sandy bridge jump, i dont think you are quite aware of how much of a jump sandy bridge was.

http://www.bit-tech.net/hardware/2013/06/12/intel-core-i5-4670k-haswell-cpu-review/6

look at the i5's all very close to each other. To put some perspective. 5ghz sandy = 4.7ghz ivy = 4.4ghz haswell approximately

In the other hand, look at the bloomfield i7 920, poor bastard doesnt have a chance against the 2500k, let alone the 2600k. Also, its practically impossible to overclock it to sandy bridge lvls.

The 2500k was superior to the 920 in every aspect, including price (220$ vs 300$). For haswell to be such a jump the 4670k should have beated the 2600k the same way the 2500k did

I know exactly what the Sandy jump was, and Haswell is an equal jump if your code is properly parallelized. The problem is all the software comparing Sandy and Haswell is still not using the parallel instruction sets which arrived for Sandy. That 4th ALU is a huge performance boost if you compile code which can use it.

If we compiled all existing benchmarks using the Intel C++ compiler (and inserted a fair dispatcher) you'd see Haswell kicking the snot out of Sandy Bridge.

[KamiKatze]

Next estimated upgrade at this pace of development is prolly around 2020 for me.

[Dabombinable]

Next estimated upgrade at this pace of development is prolly around 2020 for me.

Same here, I've gone back to my i5 rig and haven't noticed a difference at all in overall performance (BF4 and Crysis 3 are exceptions). I'll probably just do an upgrade every 5-10 years since computers are no longer making the leaps and bounds they did around 2000.

[KamiKatze]

Same here, I've gone back to my i5 rig and haven't noticed a difference at all in overall performance (BF4 and Crysis 3 are exceptions). I'll probably just do an upgrade every 5-10 years since computers are no longer making the leaps and bounds they did around 2000.

I'm okay with that. Swap out a gpu once in a while, add some ram maybe. A bright future for pc gamers. But games keep getting more demanding, i mean crysis 3 is still nowhere near fluid from 1 gtx 980

[Dabombinable]

I'm okay with that. Swap out a gpu once in a while, add some ram maybe. A bright future for pc gamers. But games keep getting more demanding, i mean crysis 3 is still nowhere near fluid from 1 gtx 980

lol, I can play Crysis 3 at Ultra settings on my GTX 970 with my main rig (the overclocking rig had issues though-CPU bottleneck and the PCIe x16 v1.1 slot) and although I don't keep track of the FPS (only did that on the old rig to see how bad the bottleneck was-29% BTW) it was completely smooth at 1080p. I don't know how your managing to get issues. (And I'm using tech that is a lot cheaper-excluding the $500+ G1 gaming).

[KamiKatze]

lol, I can play Crysis 3 at Ultra settings on my GTX 970 with my main rig (the overclocking rig had issues though-CPU bottleneck and the PCIe x16 v1.1 slot) and although I don't keep track of the FPS (only did that on the old rig to see how bad the bottleneck was-29% BTW) it was completely smooth at 1080p. I don't know how your managing to get issues. (And I'm using tech that is a lot cheaper-excluding the $500+ G1 gaming).

I'm really picky. If its not pinned at 60 FPS, or at least in the 50-60 FPS range, it's not fluid to me. And my 970 draws 35-45 on ultra with full 8x AA on crysis 3. in the demanding rainy part at least. But thats okay, I'm getting my 2nd 970 tomorrow!

[Dabombinable]

I'm really picky. If its not pinned at 60 FPS, or at least in the 50-60 FPS range, it's not fluid to me. And my 970 draws 35-45 on ultra with full 8x AA on crysis 3. in the demanding rainy part at least. But thats okay, I'm getting my 2nd 970 tomorrow!

Enjoy the performance increase. My mobo is stuck in the stoneage with its second x16 PCIe slot running v2.0 @2x, 4x if I disable both v3.0 PCIe 1x slots and therefore my internet. I can't do jack in the way of large performance upgrades, but I am looking at buying my first full ATX motherboard since my 1999 Jetway dual CPU beast (socket 370 and slot 1).

[Trik'Stari]

How many "broadwell sucks!" threads are we going to be seeing?

 

Or at least, for how long will we be seeing them?

[Dabombinable]

How many "broadwell sucks!" threads are we going to be seeing?

 

Or at least, for how long will we be seeing them?

Broadwell doesn't suck when you look at the reduction in power consumption. Otherwise there just isn't much point for it yet in the desktop market unless it had a significantly larger increase in performance.