Haswell Core Count

[AtenRa]

So don't underestimate Intel. They know what the market needs. If it's evolving slower than you expected, maybe you should adjust your expectation and also see how revolutionary AVX2 and TSX really are.

It is more than a year since SB and AVX are available in the market. Could you tell me how many desktop applications can take advantage of AVX ???

I dont say we dont need AVX and AVX2 but we have Multithreaded Desktop CPUs (Pentium 4 HT) from 2002, a decade have past and there are people here that say we dont need more than 4 cores in Desktop. Im sorry but that is pathetic, programs are lagging behind hardware advances more and more the past few years.

If AVX will take a decade to be mainstream then AVX2 on Haswell is not something you people should get your hopes up and running.

 

[LOL_Wut_Axel]

@Wut Axel

I honestly have no idea why you made that long post comparing the 3820 to the 2600K, which has been beat to death back in Jan. In this thread I was comparing the 3770K to the 3930K which is the 4 core and 6 core CPUs of choice right now. Your post does not seem to add anything useful to that discussion.

Interesting that you say that, given it was you that started this.

Let me quote you again:

3770K = $349, which is $87 per core.
3930K = $599, which is $100 per core. But for the little extra cost, you are getting more PCI-e lanes, double RAM bandwidth and more L3$ per core.

Sorry, but there is nothing wrong with that pricing if you ask me.

It was you arguing in this thread that those three things is what made it worth it. I debunked it using the 3820 as the prime example because double the memory bandwidth doesn't increase performance in real-world desktop programs against the 2600K/3770K and neither does having 0.5MB more of L3 cache/core (BTW, bad argument if you were saying more L3 cache/core for the 3930K because it has the same 2MB L3 cache/core as the 3770K). I was just correcting you in that it's the extra cores that make it worth it, not the other three things.

 

[Magic Carpet]

It is more than a year since SB and AVX are available in the market. Could you tell me how many desktop applications can take advantage of AVX ???

I dont say we dont need AVX and AVX2 but we have Multithreaded Desktop CPUs (Pentium 4 HT) from 2002, a decade have past and there are people here that say we dont need more than 4 cores in Desktop. Im sorry but that is pathetic, programs are lagging behind hardware advances more and more the past few years.

If AVX will take a decade to be mainstream then AVX2 on Haswell is not something you people should get your hopes up and running.

Can't disagree with that.

 

[frozentundra123456]

The REAL question being asked here is: "When Can I have 6 cores for the prices of 4?" snip
QUOTE]

NO, I dont think anyone suggested that. What people want is a mainstream six core chip at prices proportionate to a quad core, even without all the extras that SBE or IVBE will bring, if necessary. You dont seem to be able to understand that. And in fact, I believe that progress was made so that now we can get a quad core for less than the early dual cores. Where is that progress toward six cores?

However, considering the savings from the die shrink to 22nm, Intel could probably sell a six core IVY or haswell for the same cost as a sandy bridge quad core. In fact, on the desktop, I would much rather see intel's efforts put into more cores at a reasonable price or better CPU performance instead of improved graphics, especially for Haswell. IVB seems to have sufficient graphics for most users, and if you really want to game or use graphics intensive programs, just add a discrete card.

 

[ShintaiDK]

More cores for what? Idling? Not to mention the cost of creating 6-8 cores designs just to please a tiny crowd. As it is today you can get a 6 core Thuban or Westmere/SB-E. Bulldozer and i7 Lynnfield/SB/IB just gives you the illusion of 8 threads.

 

[pelov]

In fact, on the desktop, I would much rather see intel's efforts put into more cores at a reasonable price or better CPU performance instead of improved graphics, especially for Haswell. IVB seems to have sufficient graphics for most users, and if you really want to game or use graphics intensive programs, just add a discrete card.

But that's precisely the reason why we won't get 6 core chips. 4 cores and 4 cores / 8 threads is enough and more than enough. Graphics is generally the most intensive task most people do nowadays so it makes more sense to address that then add more threads/cores if it's unnecessary.

 

[IntelUser2000]

It is more than a year since SB and AVX are available in the market. Could you tell me how many desktop applications can take advantage of AVX ???

I guess we overestimated how much floating point extensions have on consumer applications. Look for example how poor Bulldozer and AMD CPUs do in general do compared to Intel ones, but then when we look at workstations, the situation completely reverses: http://www.anandtech.com/show/5553/the-xeon-e52600-dual-sandybridge-for-servers/12

Ignoring the Xeon E5 results, the AMD chips perform significantly better on floating point intensive HPC applications. In HPC, AVX and FMA are all very important performance features.

http://software.intel.com/en-us/blogs/2011/06/13/haswell-new-instruction-descriptions-now-available/

AVX2’s integer support is particularly useful for processing visual data commonly encountered in consumer imaging and video processing workloads. With Haswell, we have both Intel® Advanced Vector Extensions (Intel® AVX) for floating point, and AVX2 for integer data types.

Isn't that interesting?

 

[pelov]

On Intel's description for AVX2, they say the extension of Integer SSE to 256-bit would have big impact for consumer video editing applications. So it basically means the FP extensions are useless for a vast majority of cases.

I think that's proven to be true if you look at current AVX implementation and will continue into AVX2 with video editing being the major, and perhaps only, widely used application(s) to take advantage. AVX2 being used for gaming is a pipe dream but in HPC it makes far more sense.

I posted this in the other thread, but it bears repeating here; Bulldozer's AVX performance is actually pretty good so long as all threads are tasked. In HPC and video editing this is exactly what happens

Compared to non-AVX in the same benchmark

BD actually handles AVX better than Intel does. Still, though, it only has 4 FPUs so though it fares better than Intel's 4 FPUs with respect to AVX, it isn't allowed to stretch its legs and use that advantage because it's still only got 4 :/ Anyway, most FP-related tasks a user comes across that require good performance are games and we've got a co-processor for that.

 

[Hulk]

LGA2011 is workstation/server class. Hint hint...

Also people needing more than 4 cores on their desktop accounts for what? 0.01%?

Depends on how you define "needing."

Is speeding up video transcoding "needing?"
Is transcoding a video while playing a game "needing?"

Perhaps most people don't even need a dual core?

 

[Edrick]

It was you arguing in this thread that those three things is what made it worth it. I debunked it using the 3820 as the prime example because double the memory bandwidth doesn't increase performance in real-world desktop programs against the 2600K/3770K and neither does having 0.5MB more of L3 cache/core (BTW, bad argument if you were saying more L3 cache/core for the 3930K because it has the same 2MB L3 cache/core as the 3770K). I was just correcting you in that it's the extra cores that make it worth it, not the other three things.

I stand corrected. I completely forgot the 3930K does not have the extra L3$ per core that the 3960 and 3820 has. My apologies.

 

[Edrick]

But every new instruction hardware set, requires a re-implementation in 99% of the cases. Moar cores on a base modular build up of locking/threading access does not.

I agree with the new instruction set. It generally does require a re-compile with possibly some code changes. But generally this is nothing too difficult and since most large software houses release new versions yearly/bi-yearly, we can usually expect a turnaround of 2 years or less (in most cases). But sometimes it wont even make a difference, so they dont bother.

I disagree with the modular build point. I do not know of many applications that will scale to X number of cores without some sort of code change or re-compile. I know it is possible (to a certain degree), but very complex and expensive to do. And with Intel only upping the core values every 4-5 years, why bother?

 

[2is]

I don't live in the States, I'm not really that into actual market prices. I'm going off what Intel says the processors should be sold at. It doesn't make any difference to my actual point though.



A lot of professionals working with graphics or design would very much like more cores at an affordable price. I would very much like more cores at an affordable price. Saying there's no need for it therefore Intel shouldn't try to provide it, is not very forward thinking.

I would absolutely love a Ferrari at an affordable price... Fact is that 8 core processors are not main stream, they are niche/enthusiast markets and you will pay enthusiast prices until they become mainstream. And please don't even mention bulldozer. 8 half-cores don't count.

 

[BenchPress]

It is more than a year since SB and AVX are available in the market. Could you tell me how many desktop applications can take advantage of AVX ???

Can? Plenty.
Will? Few.

It's a matter of cost vs. gain. AVX1 is an abomination. It only doubles the width of floating-point SIMD instructions, doesn't offer anything to facilitate SPMD processing (which vectors this wide begs for), nor did Intel provide any increase in cache bandwidth to sustain a higher throughput. So it's hard to program and the speedup is low. Hence few developers have bothered to use AVX1.

AVX2 is completely different. It offers 256-bit SIMD integer instructions, fused multiply-add, gather support, and vector-vector shift. That's finally everything we need for auto-vectorization of loops with independent iterations (SPMD programming model). It's straightforward to use, and will make many generic 32-bit code loops run eight times faster!

 

[BenchPress]

Anyway, most FP-related tasks a user comes across that require good performance are games and we've got a co-processor for that.

No we don't. The GTX 680 is pathetic at general-purpose computing.

Games are severely limited by the lack of throughput computing power offered by today's CPUs, and clearly even the latest GPU is rubbish for GPGPU.

Hence AVX2 will revolutionize gaming. It brings us SPMD throughput computing, without the limitations the GPU runs into. The future is homogeneous computing, not heterogeneous computing.

 

[dinker99]

Show me the software that craves 6 cores ?

seti .. folding .. encoding .. a photoshop plugin ?

4 cores is more than enough for the time being .. let the software tinkers get a chance to catch on ... it will take another 5 years minimum (i think).

I work with Maxon C4D, as this has no GPU acceleration I need more cores

 

[Mr. Pedantic]

I would absolutely love a Ferrari at an affordable price... Fact is that 8 core processors are not main stream, they are niche/enthusiast markets and you will pay enthusiast prices until they become mainstream. And please don't even mention bulldozer. 8 half-cores don't count.

Except by telling me not to expect so much, you're essentially telling me it's unreasonable for me to want it to be mainstream sooner.

 

[greenhawk]

You EXPAND your thinking to live outside your US bubble, thanks.

While I would generally agree with the bubble statement, I find it sillier still that if there is such a large price difference between local and US prices, why not mail order? been doing that for years. Far easier than going off about local prices.

 

[greenhawk]

The REAL question being asked here is: "When Can I have 6 cores for the prices of 4?"

had to laugh at that. it follows very closly with the marketing that AMD gives as to why they pushed for more cores

 

[greenhawk]

Else why buy the new products?
(along with the new sockets and chipsets).

I would say people that upgrade for 20% performance increase are not really caring about value for money, they want the latest and greatest and so money takes a back seat.

People some times wake up to this cycle and skip every second, just so they can afford better gear every second release.

 

[greenhawk]

A lot of professionals working with graphics or design would very much like more cores at an affordable price. I would very much like more cores at an affordable price. Saying there's no need for it therefore Intel shouldn't try to provide it, is not very forward thinking.

There is a need for performance, and intel deliver options for it.

Just because there is no competition at that level means intel can charge what ever they want for it.

might as well be going off at AMD for being bad competition, but still be one of those people that do not buy there products to help fund then to be a good competition against intel.

 

[greenhawk]

Is it so horrible to want to pay less?

Nope, but the desire can head in the direction of being labeled the same as the people that have not grown up who complain, or the ones that go down the path of "it is too expencive to buy, I'll just steal it instead".

It is all marketing exists for, convice people that they have a need for something they actually could have lived without. Said desire then either gets them to buy the item (and they get a rush of acheivement) or they do something silly about not being able to afford it.

Personally, once you get to the point of

a) acknowledging the qualities of something
b) knowing you do not need it

your life gets soo much better as you have far less to get your blood pressure up about.

 

[greenhawk]

Are we going to have to discuss why quads cost more than dual cores next?

proberly head towards the older one of how hyper threading counts as real processors

 

[Lonbjerg]

3930k is 500 USD.
+ a cheap mobo - 250/300 USD.

800 USD.

Now let's try that overseas here in EU:

3930k = 900 USD
Granted there are cheap mobo's for 350 USD in my territory, but most are 400 USD.

That equals 1250/1300 USD averagely.

That's more than 50% the cost from EU.
A cheap mobo + 3550k costs me 500-600 USD.
(Depending on mobo/features).


So no, it's not just "SIMPLE CHEAP JUST A BIT MORE" for everyone.
And this is discounting that if i live in the US - there's CRAZY deals on mobo + mainstream i5/i7 deals in most big retailers suddenly.


HENCE, EMPHASIS on MAINSTREAM.

You EXPAND your thinking to live outside your US bubble, thanks.

PowerYoga:
Your statement is ignorant and dumb.
By your logic we'd all still be single core'ing and with 1mbit internet.
Give the people power/tools and they'll use it.

I personally can't afford to mush out for a x79 (or x58 which saves me 100 usd) - just so i can have easier time on my programming/virtual machine hobby projects.

4 Cores have been standard since Nehalem - that's 4 years coming haswell. I was hoping they'd notch it up a tad.
(You bet they would, if bulldozer wasn't a dud).

You are barking up the wrong tree.
Import taxes and VAT = you should talk to the EU politicians.
So you calling other people dumb...talk about irony!

 

[greenhawk]

The folks who say "who needs more than 4 cores" arent very forward thinking...

possibly, but they are thinking in the now.

I was one of the people saying dual was better than single back when dual required two cpu's, the rest came around to my thinking.

That being said, I have not seen a huge number of things that benifited my daily tasks to make quad massivly better over dual until recently (still running a q6600, so it took a long time).

It seems the best thing since the q6600 has been turbo mode which disables unneeded cores. Not sure what feature will be next but I do not have the gut feeling it will be going to 6 cores for the masses (like the local wall mart / ect will find it any easier to get mom and pop to but a quad over a dual).

 

[2is]

Except by telling me not to expect so much, you're essentially telling me it's unreasonable for me to want it to be mainstream sooner.

Well that's your interpretation of what I said, which is fine. Fact is that it isn't mainstream and that's because it doesn't need to be. Anyone that "needs" that much power can have it, it'll just cost them. Anyone that wants it is in the same boat. It's not unreasonable for intel to want to make a larger profit on their top teir products. It's not unreasonable to want it, but it's also not unreasonable for Intel to not give it to you, yet.