Core i7-4770K is performance crippled

[ShintaiDK]

This is not so good:

http://www.pcpro.co.uk/news/382267/intel-haswell-hotter-and-slower-than-expected

It appears ES samples sent to system integrators overclocked much better than the production CPUs they got.

That type of story had been around for ages with previous chips. And it never holds water.

 

[grimpr]

And what hinders intel to offer a k version with TSX and the eDRAM / l4 cache?

The only reason is marketing. Such a chip could be faster than upcoming IB-E...even in multi-threaded scenarios vs a 6-core.

They just need to move to LGA2011 exclusively, kill the mainstream LGA11xx line and leave the market to AMD, problem solved, dog satisfied pie intact. :biggrin:

 

[ShintaiDK]

And what hinders intel to offer a k version with TSX and the eDRAM / l4 cache?

The only reason is marketing. Such a chip could be faster than upcoming IB-E...even in multi-threaded scenarios vs a 6-core.

Is it actually possible to fit the eDRAM into the LGA1150 chip with the IHS on?

 

[blackened23]

This is not so good:

http://www.pcpro.co.uk/news/382267/intel-haswell-hotter-and-slower-than-expected

It appears ES samples sent to system integrators overclocked much better than the production CPUs they got.

Oh great, completely circumstantial evidence with no real proof. Nothing to see here, move along.

 

[ShintaiDK]

They just need to move to LGA2011 exclusively, kill the mainstream LGA11xx line and leave the market to AMD, problem solved, dog satisfied pie intact. :biggrin:

AMDs desktop sales is collapsing alot faster than their mobile sales. So the trend is the same for both companies, except that Intel can keep its revenues stable.

The question might be, when is the desktop dead?

 

[Phynaz]

Intel is doing that thing they did with Willamette (ala Rambus) and Prescott (hardly any clockspeed or IPC improvement) where they completely lose focus on delivering a product that the consumer can understand.

I'm guessing that all this product segmentation is being requested by the OEM's so they can hit their desired price points.

And while you see this as the beginning of the downfall of Intel, you do have to admit it has worked very well for Intel so far, along with many, many other companies. It's why the Toro lawn mower you but at Home Depot isn't sold anywhere else - Lowe's get's a slightly different model, Amazon gets a different model, etc.. And while we know that "past results do not guarantee future performance" markets don't change quickly at all. If all the market segmentation ends up being a negative to Intel's financials you can bet they will reduce the number of SKU's.

As far as TSX being disabled on K CPU's...I have no idea why they would do this. You have mentioned in the past it being a validation issue. One less thing that could possibly malfunction when a chip is being run highly out of spec.

 

[StrangerGuy]

They disable TSX for the same reason they installed a huge gap between the IHS and the die: to limit performance. It seems intel is more focused on limiting performance than they are providing the best product possible in any given price range. The market responds by moving away from x86 in droves.

I don't know what drugs were they on when they decided Celerons and Pentiums won't have AVX enabled.

"Hmm, this already questionable feature is too good for the peasants that makes up the majority of the market...So let's make it harder for the entire market to adopt so we can cut our nose to save our i7 face!"

 

[blckgrffn]

It's simply aggravating.

I will be doing the Haswell K route because they gimped the OC on the non K parts and also because that is the CPU that Microcenter offers with the mobo discount.

How Intel thinks that their instruction sets will be utilized when they make a confusing mess of SKU's that have a cocktail of what is available is beyond me.

I could see if they had two segments. Core ix has all the goodies, pentium and celeron a consistent subset. But no! Jeebus, the mess that is the mobile chips features is ridiculous with their mish-mash of enabled features... then we get to the desktop.

So IVBe is going to come out and not have all the features that a "mainstream" core i5, at 1/3 the price, has? WTF.

It is my impression that walking away from your core market, even as it dwindles in the scope of your gross sales and markets, is foolhardy. HP should have continued to make the best printers on the market. Given their resources, there can be no excuses. Apple should have 30" cinema displays, epic Mac Pros, and big focus on their core (final cut) software that makes them untouchable in that segment. Given their resources, there can be no excuses.

Same for Intel. They should absolutely dominate the desktop CPU market with clear choices and best in class everything. Given their resources, there can be no excuses.

I am a big believer that taking care of those segments that made you great is important for PR - and when you take a misstep in one of your other business lines, the loyalty of the customers served in those segments may overwhelm their rational decision making ability and help you retain those sales. Once you get soft in the segment you were once the clear choice in, doubt begins to creep in about your ability to execute in markets where your track record is shorter.

Yeah, I know. My opinions are the basis for the reality we all live in 😉

 

[NTMBK]

Is it actually possible to fit the eDRAM into the LGA1150 chip with the IHS on?

Intel designed the socket size- if it is not possible, then it's their own fault.

 

[Mallibu]

It's always fun to see a hyperbole laden post here at what is now AMD zone. Let me get this straight - TSX, a new feature that has never been implemented before, is not present on the 4770k.

This means the obvious conclusion being, the 4770k is performance crippled and is the worst processor ever produced in decades. This is, of course, despite the fact that the 4770k consistently has higher IPC than the prior 3770k which it replaced. Which, oddly enough, doesn't have TSX. 🙄 Am I getting this correct, benchpress?

Indeed. So much hyperbole in this post. By that logic, every cpu in the world is crippled except 4770

 

[BenchPress]

It's always fun to see a hyperbole laden post here at what is now AMD zone. Let me get this straight - TSX, a new feature that has never been implemented before, is not present on the 4770k.

This means the obvious conclusion being, the 4770k is performance crippled and is the worst processor ever produced in decades. This is, of course, despite the fact that the 4770k consistently has higher IPC than the prior 3770k which it replaced. Which, oddly enough, doesn't have TSX. 🙄 Am I getting this correct, benchpress?

Don't blame me of hyperbole; I never said it's the worst processor ever produced in decades. I totally agree the 4770K isn't bad at all in comparison to the 3770K, but that's not the point. The 4770 has more features, which could very well make it faster. Paying extra for disabled features is crazy and a real issue in how Intel treats its customers.

This forum never ceases to amaze. And I suppose you have absolutely no quantifiable proof (IE BENCHMARKS) showing TSX to have a tangible benefit, no? Just press releases? Yeah, figured.

Please. The burden of proof is as much on people who claim it isn't an important feature than on those who claim it is. But in any case it doesn't take away the fact that paying Intel to fuse out features in a flagship product is crazy. What's next, no Hyper-Threading, no 64-bit support?

So unless you have any proof of the contrary, TSX seems like a pretty big deal because it allows to use fine-grained thread synchronization with low overhead, whereas previously only coarse-grained locking was feasible but it results in high idle time. That is a known fact. Rarely any software scales linearly with the number of cores, even when not bandwidth limited. Also, hardware transactional memory and lock elision have been a big topic in multi-core research for many years, so who are you to claim it's suddenly irrelevant when Intel implements this feature for consumer CPUs for the first time but then decides not to enable it on the K series?

 

[mrmt]

And what hinders intel to offer a k version with TSX and the eDRAM / l4 cache?

The only reason is marketing. Such a chip could be faster than upcoming IB-E...even in multi-threaded scenarios vs a 6-core.

Cost and price. How many people would be willing to shell out $500+ USD for those SKUs?

 

[blackened23]

You're the one stating that the 4770k is performance crippled. Since you obviously have zero quantifiable data showing the benefits of TSX, the burden of proof is on us to prove you wrong, yeah? Is that it? Am I understanding you here?

 

[beginner99]

Is it actually possible to fit the eDRAM into the LGA1150 chip with the IHS on?

Good point. You are probably right. They can't make a socketed version of it (or at least not with a certain required investment) and hence offering such a SKU would not be cost effective.

Cost and price. How many people would be willing to shell out $500+ USD for those SKUs?

see above. But if a socketed version would already I think price for end-user would not be an issue. of course it would not be a top-selling part.

 

[mrmt]

see above. But if a socketed version would already I think price for end-user would not be an issue. of course it would not be a top-selling part.

Wouldn't it? What about the 3930K and the 4930K that would sell around the same price? Would you pay $500 for extra L4 cache or for two more cores?

 

[beginner99]

Please. The burden of proof is as much on people who claim it isn't an important feature than on those who claim it is.

No it's not. What your saying is like the religious fanatic telling the scientist to proof that god does not exist. In fact science works the opposite way. You have to proof that your theory is true because you are claiming 4770k is crippled. You can't come up with a random idea and tell tell others to disproof it. no, no. That is not how it works...🙄

So unless you have any proof of the contrary, TSX seems like a pretty big deal because it allows to use fine-grained thread synchronization with low overhead, whereas previously only coarse-grained locking was feasible but it results in high idle time. That is a known fact.

Your knowledge is wrong as I have indicated twice. fine-grained locking can be done in software and if you already have it , TSX doesn't help that much.

http://www.anandtech.com/show/6290/...ll-transactional-synchronization-extensions/4

And this time actual read the information people offer you instead of clinging to your misguided beliefs. the parallels to a religious fanatic are well, disturbing.

Rarely any software scales linearly with the number of cores, even when not bandwidth limited. Also, hardware transactional memory and lock elision have been a big topic in multi-core research for many years, so who are you to claim it's suddenly irrelevant when Intel implements this feature for consumer CPUs for the first time but then decides not to enable it on the K series?

Again. You came up with an idea. That 4770k without TSX is crippled and can perform worse than an a 4770 even when overclocked. That is your idea and hence it is up to you to come up with data to support it. Only then it becomes "truth" and is accepted.

So you finally get it know? omg

 

[KompuKare]

I don't know what drugs were they on when they decided Celerons and Pentiums won't have AVX enabled.

"Hmm, this already questionable feature is too good for the peasants that makes up the majority of the market...So let's make it harder for the entire market to adopt so we can cut our nose to save our i7 face!"

Yes, I still do not get the thinking here except for laziness. That is, if they want to limit AVX2 performance on Celerons and Pentiums fair enough but then they should be willing to put some extra work in and figure out a way to bring those instructions to their low end (sans Atom of course).

By all means have AVX2 running slower on those low end parts. Maybe by using microcode to emulate AVX2 on the AVX(1) part or something. Or put wait states in. Anything is better than just fragmenting the market with half-hearted support. A lot more work for sure, but if they are that keen to disable features they should be willing to put some effort in...

 

[beginner99]

Wouldn't it? What about the 3930K and the 4930K that would sell around the same price? Would you pay $500 for extra L4 cache or for two more cores?

obviously depends on the benchmarks. Maybe that L4 cache is useless in consumer workloads, maybe it helps a ton in gaming. I don't know and AFAIK no one was yet able to determine this.

 

[BenchPress]

Highly unlikely. Where in a graphics driver would fine grained locking of shared data structures be required?

Are you kidding? Everywhere. All the dependencies between draw calls have to be respected, while also allowing the application to update resources, from multiple threads, simultaneously, and servicing some asynchronous queries. Also, graphics drivers have only milliseconds to get things done to achieve a high frame rate and minimal lag. So a lot of things happen in a short time frame. Hence fine-grained locking is something that would certainly benefit graphics drivers.

TSX is a data center function.

It is no more data center specific than Hyper-Threading. Any multi-threaded application benefits from TSX.

 

[Exophase]

Sure, AVX2 is a major feature on its own but when you don't have efficient multi-threading it can't be put to good use.

I thought the popular rhetoric was that AVX2 enables all loops w/independent iterations can be vectorized now vs usually not vectorized before, and that most software spends a large amount of time in these loops. What about that would be contingent on multi-threading or TSX in particular?

Intel's ISA fragmentation is dumb and probably hurts them more than it helps, but if anything we'll see fewer people write software that uses the feature it because of it. Either way I doubt that software would have caused a huge difference any time soon, it's just a matter of wanting to target something a very very very small percentage of users have vs only a very very small percentage. People aren't exactly going to be rushing out the doors to upgrade their desktops to Haswell any time soon.

 

[LogOver]

Not sure if someone mentioned it but it is also possible that Intel removes some functionality from K-series as it can limit (or prevent) overclocking. It might be that with TSX on it overclocks much worse than when it off.

BTW, Intel has published new price list which includes R-series CPU (Iris Graphics)
http://files.shareholder.com/downlo...une_03_13_Recommended_Customer_Price_List.pdf
i7-4770R - $392
i5-4670R - $310
i5-4570R - $288

So Intel is asking additional $90-100 for GT3 + eDRAM. I guess the main reason we wont see socket version of GT3e is that the eDRAM die doesn't fit into the socket.

 

[mrmt]

obviously depends on the benchmarks. Maybe that L4 cache is useless in consumer workloads, maybe it helps a ton in gaming. I don't know and AFAIK no one was yet able to determine this.

See the point here? Some people are saying they want more performance, but the "more performance" option is already on the market (LGA2011, It's not perfect, but it's there).

What people want is more bang for the buck, not more performance, but who doesn't? And as Intel isn't under any competitive pressure, we either wait for a few years down the road for these features to become standard, or we go to the non-perfect solution (LGA 2011).

I think it's clear that the desktop is on its dusk as a platform, nobody thinks it's profitable enough to develop a platform tailor made for the desktop.

 

[aigomorla]

Intel: I got a brilliant idea... lets put IGP in all our cpus!
US: Uhhh thats a great idea but how about limiting to only the low end cpus.
Intel: Why? Dedicated GPU's are dead... our HD gpu is console grade!! console grade is the best.
US: no... we all get titans.... there is no reason for a IGP in a titan system.
Intel: Well thats your fault... who told u to get dedicated GPUs.. our HD is rock solid that it has to be included in ALL of our processors.
US: but that will just add to heat issues... and also can u remove the vrms off the cpu?
Intel: but that would give u guys better overclocking... no.. we will only do that on our 1000 dollar cpus...
US: but ...
Intel: FINE if u want it like that... wait til the server line where we will charge you 500 for the board and 800 for the unlockable CPU.
US: ARRRRRRRRRRRRHHHHHHHHHHHHHHHHHHHHHHHHHHHHHHGGGGGGGGGGGGGHHHHHHHH

You intel fanboys asked for AMD's destruction...
Well this is just the start of AMD's destruction...

AMD id say probably brought a lot of people foot first in overclocking.
C2D is what held most people..

without AMD i think intel just has no need to compete in the overclocking area... hence we are paying for it.
:\

That is how i see Haswell

 

[beginner99]

I think it's clear that the desktop is on its dusk as a platform, nobody thinks it's profitable enough to develop a platform tailor made for the desktop.

I don't diagree because an extra socketed version would probably not have a + ROI.

But jsut assume the socketed version was already here. Then why not give it a desktpo sku number and sell it for $500? you don't loose anything and if you get a couple if people to buy it who would less have bought a 4770k you profit.

But again in this specific case it made sense. i initially just didn't think about the fact there is only a bga version of iris pro.

 

[bronxzv]

I thought the popular rhetoric was that AVX2 enables all loops w/independent iterations can be vectorized now vs usually not vectorized before

Haswell isn't changing much the situation, only packed shifts with independant count may help a bit, but it's something rarely used in practice, IMO

gather seems way too slow (the optimization guide [1], bottom of page C-5, reports around 10 clocks reciprocal thoughput for 256-bit 4 elements and 8 elements gather) to really make a difference, if you aren't careful it can even *slow down your code*, it was my 1st experience with gather

in fact the main "optimization" for my existing AVX2 code was to disable all the gather code paths (!)

[1] https://www-ssl.intel.com/content/d...4-ia-32-architectures-optimization-manual.pdf"