Yes another Haswell thread. Let's have a look at tock-to-tock IPC.

[SunRe]

Cool thread, actually Haswell might start looking even better: http://forums.anandtech.com/showpost.php?p=35123392&postcount=8

I managed to extract ~22% speedup clock-for-clock compared to Ivy Bridge generation using AVX2 (including usage of both FMA and Gather instructions)

 

[mrmt]

My thesis is, had they instead of focusing on expanding and development the market around performance, instead of beating AMD, they would be in a far better position today. They shot themselves in the foot.

I think you are overestimating what Intel can do here.

The most demanding OS ever was Windows Vista, and that was 2006. Windows 7 and Windows 8 both reduced system requirements. Games? What a Joke. We're in 2013 and stuck with 2005 hardware.

At the same time you can't really compare 2005 processors with what we have today. Both in raw performance and power efficiency we have a very different landscape in the CPU market. If companies wanted to write more demanding apps for their customers, the user base is there, but nobody does.

At the same time you have a booming mobile market. First laptops, then tablets and phones. I don't think Intel has much choice here except ride on this trend, as they are trying to do now.

 

[guskline]

Hulk: Thank you for the analysis and data. Very informative.

 

[Hulk]

Yes they will. And thats excactly why they have failed.
They own a dying market. A market where there is very solid profit for Intel.

My thesis is, had they instead of focusing on expanding and development the market around performance, instead of beating AMD, they would be in a far better position today. They shot themselves in the foot.

Now i will go and asemble my Haswell system for my boy. He doesnt even want to open the package that just arived. He doesnt even want to look at it ! Man - and here his father does everything for him He just want to get rid of the wireless and have a 1g cable. Damn kids theese days, they dont care for tech.

And perhaps thats exactly the point. Intel just lost a new generation.

Interesting thesis. But do you really believe the desktop market would be in a better position if more compute was available? The logical conclusion to your thesis would be that people are migrating to mobile devices because there isn't enough compute available on the desktop? Which would also lead to the conclusion that people need more compute.

I guess it's possible but I find it more plausible that people are migrating to mobile devices because they are better than ever. They are enough compute, battery life, good form factors, and pricing to make them a viable alternative. In addition much of these mobile devices are new to most people, while desktops have been around for decades. It may be that once people have their desktop, laptop, tablet, and phone, all of those markets will equalize to some extent.

You can still get more compute, more expand-ability, and generally more flexibility, upgrade-ability, and more for your dollar with a desktop system and I think that will be the case for some time. I have a few desktop systems and I expect to have them around for quite some time.

I have a strange feeling that when the dust settles it's going to again be just Intel and AMD. AMD knows what it's like to compete with Intel. The ARM guys haven't felt the pressure yet. I hope that's not the case and we have a competitive market for a long, long time.

 

[Cogman]

I think we are going to see a few Haswell like generations from intel. My bet is they will focus mainly on energy efficiency and GPU performance with minor increases in general performance. I think that is where they NEED to go for the future.

Big improvements to CPU performance are just not worth it right now. At the corporate level, most bottlenecks come from disk and network traffic. At the consumer level, the CPU is fast enough for their favorite games. So rather, I see them looking at plugging up their current holes. They are leaking like crazy when it comes to the mobile market. ARM based tablets and phones are destroying a lot of peoples desire for Intel based laptops and tablets.

Haswell looks like it was specifically designed to plug that hole. I imagine the next few generations of intel processors will be primarily focused on shoring up the mobile market.

 

[TuxDave]

Standard project management in engineering. How do people build bridges and know in advance they aren't going to collapse? Or buildings? Or airplanes?

Like anything in engineering, it can be done if it is done correctly. But simply claiming you've done it doesn't mean you did it correctly (see bulldozer, "IPC does not go down").

Nonsense! I just pick and choose whatever feels good that day and hope for the best!

But seriously, as IDC mentioned, it's in any CPU company's best interested to establish a methodology that allows them to predict the performance of any feature they add. It won't be perfect but it's far better than flying blind.

 

[meloz]

(delete)

 

[Vesku]

So will IvyBridge -> Haswell help kill off synthetic benchmarks being used and cited?

http://www.hardocp.com/article/2013/06/01/intel_haswell_i74770k_ipc_overclocking_review/3

Edit: Haswell looks to have more gain in the aggregate of actual software than the aggregate of synthetics.

 

[BenchPress]

At the same time you can't really compare 2005 processors with what we have today. Both in raw performance and power efficiency we have a very different landscape in the CPU market. If companies wanted to write more demanding apps for their customers, the user base is there, but nobody does.

That's only because multi-threading and SIMD parallelism has been too hard for the average developer to take advantage of. Clock frequency increases and IPC improvements are nice since they don't take developer effort, but they've been very modest in comparison to the theoretical performance improvements from multi-core and SIMD technology.

Haswell is the first architecture that really tries to address the fundamental issues. AVX2 adds essential instructions for easy SIMD vectorization by the compiler, and TSX greatly simplifies and optimizes multi-threaded synchronization. Of course the user base for Haswell really isn't there yet, and the lack of TSX support in the K-series isn't helping either. There's also a need for some middleware tools/libraries that use AVX2/TSX, although I've seen a fair bit of that in the making already.

My point is that it's not a lack of demand for more demanding applications. Consumers just haven't seen those applications yet because it has been too big of an investment for developers to exploit parallelism without AVX2 and TSX. I am certain that in several years from now Haswell will be remembered as the architecture that has really put the multi-core and vectorization revolution into motion.

Just think about how powerful GPUs are at certain workloads (which is being held back by heterogeneous overhead). That same technology is finally finding its way into the CPU cores. At the same time GPUs are adopting CPU technology, such as call stacks, exceptions, and virtual memory. So eventually they'll unify and eliminate the heterogeneous overhead.

 

[Ancalagon44]

That looks like a strawman to me?

I think i have explained why the focus was wrong?
If yes - You can refer to the arguments there. They might be wrong but we can discuss from there.

You are saying that, by going after IPC, Intel gave up higher overall performance?

Completely wrong.

Look at K8 vs P4 - AMD went after IPC, Intel went after clockspeed. We all know which was faster.

Look at Bulldozer and derivatives vs Intel Core anything - again, Intel is faster at just about everything, despite "only" having 4 cores and "stupidly" going after higher IPC.

IPC should be the first thing anyone ever optimizes, due to A) inherent scaling problems with multiple cores, except in ideal situations like graphics, and B) power consumption and heat problems when trying to boost clock speed at the expense of IPC, a la both P4 and Bulldozer.

 

[dmens]

Nonsense! I just pick and choose whatever feels good that day and hope for the best!

But seriously, as IDC mentioned, it's in any CPU company's best interested to establish a methodology that allows them to predict the performance of any feature they add. It won't be perfect but it's far better than flying blind.

But... muh netburst!

 

[crashtech]

Thanks, OP, I'm enjoying this talk about tocks.

 

[Hulk]

You are saying that, by going after IPC, Intel gave up higher overall performance?

Completely wrong.

Look at K8 vs P4 - AMD went after IPC, Intel went after clockspeed. We all know which was faster.

Look at Bulldozer and derivatives vs Intel Core anything - again, Intel is faster at just about everything, despite "only" having 4 cores and "stupidly" going after higher IPC.

IPC should be the first thing anyone ever optimizes, due to A) inherent scaling problems with multiple cores, except in ideal situations like graphics, and B) power consumption and heat problems when trying to boost clock speed at the expense of IPC, a la both P4 and Bulldozer.

Well stated. I agree that IPC should be the cornerstone of any good microprocessor architecture. What good are more cores if the first one isn't excellent? What good is pumping up the clockspeed of the core at the expense of heat and electricity if the core isn't inherently efficient at processing instructions?

 

[BenchPress]

IPC should be the first thing anyone ever optimizes, due to A) inherent scaling problems with multiple cores, except in ideal situations like graphics...

It's a common misconception that GPUs have many cores. The GTX 680 has 'only' 8 cores, with 6 vector units each, which are 32 elements wide. NVIDIA multiplies that all together to claim it has 1536 cores, but really there are only 8. For comparison, mainstream Haswell CPUs have 4 cores, with 2 (floating-point) vector units each, which are 8 elements wide, and run at over 3 times the clock frequency.

Also note that the GTX 680 reaches only a relatively small market, while Haswell is aimed at a very wide market. My point is that Intel does put a lot of effort into multi-core and vectorization. Haswell's TSX and AVX2 technology underscores that. That said, IPC should indeed not be sacrificed for more cores like AMD did with Bulldozer.

On the other hand I think it would be bad to optimize for IPC first and add more cores and SIMD capabilities as an afterthought. They really need equal attention and I think Intel achieved a home run with Haswell. Keep in mind that extracting more ILP is getting ever harder, while TLP and DLP are up for grabs. Rest assured that we're a long way from hitting the "inherent scaling issues" of multi-core and vectorization. The real issue was the lack of efficient synchronization primitives, which is addressed by TSX, and not having wide vector equivalents of every scalar operation, which is addressed by AVX2. So clearly Intel is very forward-thinking.

I wouldn't be surprised if their next major architecture barely improves IPC but instead features 8 cores with 4 vector units of 16 elements. The transistor budget required for that could come from ditching the IGP and doing graphics on the CPU cores. For power efficiency, the vector units could be split into two clusters running at half the base frequency, each dedicated to one thread and covering memory latencies with AVX-1024 instead of Hyper-Threading.

 

[sxr7171]

1. The increase of 16.4% is more than I expected and welcome.

2. They'd basically have to return to soldering that IHS and 5GHz would be pretty easy to release.

3. Laptops sell more than desktops and workstations. They are certainly following the money. Good thing for a company to do.

4. If I could get 4.8GHz like I get on my Ivy with just a H80, then yeah I'd upgrade. But as it stands I'm not hearing that will be possible. Well unless one gets into delidding. It's possible a delidding service may emerge.

 

[krumme]

We are on quad core on the 7 year. Go 7 before core2 quad, and have a look at performance.

Performance is slowing on the desktop like never before. In historical perspective its practically at a standstill.

You dont want to hear it - why is that?

You must be the weakest lead users in history. [redacted]

No wonder. You talk like "The increase of 16.4% is more than I expected and welcome" - as if 16.4% increase on 2 years is anything to talk about. LOL.

No wonder its so easy for Intel to sell the leftovers - and thats exactly what Haswell is - for this market, when the supposed lead users, is completely deprived of any leader capabilities.

If this community instead was populated with real agressive leadusers/firstmovers - it would influence Anands sweet Intel talk. And instead of quad cores we would have 8 cores since at least IB. Instead we not only get to pay for "unlocking" but dont get TSX for the K models.

[redacted]

I don't even know where to begin. If you keep attacking other users you will no longer be welcome to post here.
-ViRGE

 

[mavere]

...

Excellent parody account.

 

[Ancalagon44]

On the other hand I think it would be bad to optimize for IPC first and add more cores and SIMD capabilities as an afterthought. They really need equal attention and I think Intel achieved a home run with Haswell. Keep in mind that extracting more ILP is getting ever harder, while TLP and DLP are up for grabs.

That's true, but I still think Intel/AMD should increase IPC while they can. To me, it is the low hanging fruit. Fix that first, because it is "easier" than TLP.

Rest assured that we're a long way from hitting the "inherent scaling issues" of multi-core and vectorization.

What? We have been hitting inherent scaling issues ever since the Pentium D! Synchronization always adds a cost, not to mention, most tasks are simply not easily parallelizable. Yes, scaling issues are lower with vectorization, but with multi core CPUs, scaling is and always has been an issue, except in ideal situations.

 

[Abwx]

Intel increasing IPC by 10% is , generaly , not as impressive ,
and by far , as AMD doing the same %age.

They always rely on in house new instructions once they have
trouble competing in the performance front , it will be much
harder for AMD to then implement thoses ISAs on their own
designs and still have them run efficently on an architecture
that is completely different , that s why i m not impressed
at all with intels so called architectural enhancements....

 

[Smoblikat]

Isnt IB 77w TDP and Haswell 84w TDP?

 

[VirtualLarry]

Isnt IB 77w TDP and Haswell 84w TDP?

And yet, supposedly, Haswell is more efficient in terms of performance/watt. IOW, it increased in performance more than it increased in TDP.

 

[Puppies04]

And yet, supposedly, Haswell is more efficient in terms of performance/watt. IOW, it increased in performance more than it increased in TDP.

Keep in mind that they just increased the iGPU size by a fair chunk.

 

[ShintaiDK]

Isnt IB 77w TDP and Haswell 84w TDP?

Haswell got the FIVR ondie. So while the CPU TDP goes up slightly, the platform TDP goes down.

 

[crashtestdummy]

Yes they will. And thats excactly why they have failed.
They own a dying market. A market where there is very solid profit for Intel.

My thesis is, had they instead of focusing on expanding and development the market around performance, instead of beating AMD, they would be in a far better position today. They shot themselves in the foot.

I think you're overestimating the benefit of higher powered processors to the vast majority of users. There are very few applications that need high end computing power. It used to be that a 25% increase in processor power meant that using the computer felt completely different. It's just not the case anymore. In fact, if Haswell doubled computing power, I doubt that it would even be noticeable to most people. There's no "killer app" for high-end desktops. On the other hand, if you give my laptop 25% more battery life, that is a difference I can notice.

That is also why computing is moving to tablets and phones. These devices are now fast enough for 95% of what most people want to do now, so they have no need for a laptop, let alone a clunky desktop machine. If I did not need a computer to generate content (i.e. papers, data analysis, presentations, etc.), I would probably use a tablet as my primary device as well.

Now i will go and asemble my Haswell system for my boy. He doesnt even want to open the package that just arived. He doesnt even want to look at it ! Man - and here his father does everything for him He just want to get rid of the wireless and have a 1g cable. Damn kids theese days, they dont care for tech.

And perhaps thats exactly the point. Intel just lost a new generation.

You sound like the father of 25 years ago disappointed that his son isn't interested in building a hot rod. Your son is disinterested not because the Haswell is underpowered, but because it doesn't fundamentally do anything he hasn't seen before. Sure, maybe if it had a little more speed you could get a couple more FPS out of some game (though games are mostly GPU limited now), but why does he care, really?

 

[Abwx]

Haswell got the FIVR ondie. So while the CPU TDP goes up slightly, the platform TDP goes down.

Not true , the plateform still has a SMPS that switch
the full CPU power down to 2.4V , a value low enough
to not damage the CPU s IVRs that are unable
to work with a 12V rail as you re wrongly assuming it.