Core i7-4770K: Haswell's Performance, Previewed

[InvalidError]

[citation][nom]morpheas768[/nom]they need something to be able to claim an improvement over the previous gen chipset (7,8 series), and as for the CPU's the improvements are minor, and in fact so minor that there's practically no difference in gaming for enthusiasts.[/citation]
Enthusiasts represent less than 5% of the CPUs out there. The other 95% only needs something that is "good enough" for whatever they need to do and for office machines, point-of-sale terminals, Skype, Facebook, Twitter, watching Youtube, etc., even a Celeron 2xxx is already overkill.

Intel could make a "Netburstified" (heavily pipelined) i5/i7 to push higher clocks but for most people, this would be a waste of power and R&D effort. The simpler option is to tell those people to buy 8-cores and multi-socket Xeon CPUs if they really need that much performance today - likely cheaper than the R&D costs if Intel had to develop and manufacture a separate CPU line specifically for enthusiasts.

I doubt many enthusiasts would be interested in seeing the days of $1000 enthusiast CPUs come back. Few people would be interested in seeing Netburst's horrible performance-per-watt scaling either.

So the days of enthusiast-oriented CPUs beyond K/X variants of mainstream parts are pretty much over.

 

[InvalidError]

[citation][nom]sanilmahambre[/nom]Can there be anything like UNIVERSAL PROCESSOR SOCKET (UPS)?Again change of sockets! what am i gonna do with my expensive mobo (OLD)[/citation]
That problem is going to be solved soon enough: Haswell models with GT3 IGP are only going to be available in BGA packaging only. Soldered to the motherboard, no pesky socket to worry about!

With Broadwell, Intel's plan is to go BGA-only for all mainstream parts. Goodbye sockets... at least for most sub-$250 parts.

Since Intel will be switching to promoting embedded/All-in-One form factors with Broadwell (likely planning to test the market with Haswell-GT3), we may even end up having to say goodbye to ATX/mATX/ITX/etc. as well.

 

[Raghar]

Can you benchmark it with this? https://mega.co.nz/#!od5GlaiT!VdFGO3r334z6UPNmG3HV10NXc3oDk4wU9AuyQjKJNEc

Run pack4C.cmd with hyperthreading disabled, then enable it and run pack8C.cmd . Left numbers are actual timing in nano seconds.

 

[Blandge]

[citation]Think about it:the i7-4770T and 4765T are identical except the clock speeds. So 500 MHz seems to warrant a TDP increment of 10w.By that logic, the difference b/w the 4770S and 4770/K should also be 10w or less, not 19w.BUT! Look at the 4770S and 4330. The 4770S has more cache, a higher clock speed, higher iGPU clock, same number of cores, yet still has a 19w lower TDP! Even the 4570 is too close to the 4770S to warrant a 19w increase in TDP.My conclusion:They could have fit the 4330 and 4570 within 65w if they wanted to, and the 4670/K and 4770/K could have easily fit within 75w (or at the most 77w like Ivy).They've chosen not to. And i haven't been able to figure out why this is the case ever since the first Haswell slide leaked.[/citation]

Power (=/= TDP, but they to are closely related) is based primarily on voltage ( Frequency * Capacitance * Voltage^2, the squared is the important feature) which is not listed here. Stock voltage gets set by binning, so the chips that can run stable at the lowest voltage are set to S and T.

 

[mayankleoboy1]

[citation][nom]blazorthon[/nom]The K series CPUs don't have it even in the current series from Ivy and Sandy. How is this corporate greed? They decided that a virtualization feature was not necessary in an overclocking CPU. It is present in most cheaper models.[/citation]

It is corporate greed. Thr reason Vt-d is not in the K processors is so that users cant use these procs as acheaper alternative to their $2000 chips.
Perf of a 3770K @4.4GHZ is comparable to a stock xeon chip with 6 cores. So if these had Vt-d as well, a lot of people would not buy the Xeons.
It has been the same in SB and IB generations. Lack of competition makes the market fucked up.

This is the same reaosn why intel hasnt included TMX extensions in the Haswell K chips.

 

[ojas]

[citation][nom]mayankleoboy1[/nom]It is corporate greed. Thr reason Vt-d is not in the K processors is so that users cant use these procs as acheaper alternative to their $2000 chips.Perf of a 3770K @4.4GHZ is comparable to a stock xeon chip with 6 cores. So if these had Vt-d as well, a lot of people would not buy the Xeons.It has been the same in SB and IB generations. Lack of competition makes the market fucked up. This is the same reaosn why intel hasnt included TMX extensions in the Haswell K chips.[/citation]
You do realise that:
A) A 3770, with VT-d enabled, can hit 3.9 on Turbo Boost, and stay there till like, forever.
B) The 3930K has 6 cores, 12 thread, an unlocked multiplier and VT-d for under $600.
http://ark.intel.com/products/63697/Intel-Core-i7-3930K-Processor-12M-Cache-up-to-3_80-GHz

 

[ojas]

[citation][nom]InvalidError[/nom]All the mainstream models, which I presume means pretty much everything except K/X CPUs and maybe the top i5/i7 non-K models. Most non-enthusiast people never upgrade a CPU so losing the socket should enable cheaper and more compact form factors.[/citation]
IIRC, there won't be any broadwell for the desktop at all.

 

[ojas]

[citation][nom]InvalidError[/nom]That problem is going to be solved soon enough: Haswell models with GT3 IGP are only going to be available in BGA packaging only. Soldered to the motherboard, no pesky socket to worry about!With Broadwell, Intel's plan is to go BGA-only for all mainstream parts. Goodbye sockets... at least for most sub-$250 parts.Since Intel will be switching to promoting embedded/All-in-One form factors with Broadwell (likely planning to test the market with Haswell-GT3), we may even end up having to say goodbye to ATX/mATX/ITX/etc. as well.[/citation]
Wait, we don't know any of this. As i said on my previous reply to you, what i read last was that Broadwell isn't for the desktop at all, and Skylake will see LGA return. It makes the most sense to me, since we're complaining about the smaller performance increases every year anyway.

Now with each desktop release they'll shrink the process while simultaneously using a new architecture. We see larger performance boosts and lower power consumption (with major chipset changes), they see more sales.

 

[ojas]

[citation][nom]Blandge[/nom]Stock voltage gets set by binning, so the chips that can run stable at the lowest voltage are set to S and T.[/citation]
Hmmm. Makes sense, i guess. But...wasn't it like...the highest binned processors are the ones that overclocked the most? Or would some be more stable at higher voltages and some more stable at lower ones?

I don't know why i have a feeling that the least defective chips will be able to run over a wider range of voltages, both above and below the average...so they should be able to underclock and overclock the most.

Just a hunch, might be completely wrong here.

 

[virgult]

Will we have quad-cores on ultrabooks, then?
It's about time...

 

[nirvanaman_1985]

Power consumption under idle and load, heat under idle and load, noise under idle and load and overclockability benchmarks would be great. Also do you know if it can hardware decode H.265 or if there is programmable circuits that could decode H.265? Please test guild wars 2 and far cry 3 too.

 

[nirvanaman_1985]

please test guild wars 2 and far cry 3. Heat, power consumption and noise tests under idle and load would be great too. Maybe boot time and shutdown time in windows 7 and windows 8 would be nice too.

 

[shadowfamicom]

Would love to see more comparison to 2011 socket CPUs. Even just the 4 core i7-3820 would be an interesting comparison.

 

[shadowfamicom]

[citation][nom]shadowfamicom[/nom]Would love to see more comparison to 2011 socket CPUs. Even just the 4 core i7-3820 would be an interesting comparison.[/citation]

There also seems to be a big lack of comparisons to 2011 sockets on Tom's in general when talking about CPUs.

 

[ubercake]

[citation][nom]shadowfamicom[/nom]Would love to see more comparison to 2011 socket CPUs. Even just the 4 core i7-3820 would be an interesting comparison.[/citation]

There also seems to be a big lack of comparisons to 2011 sockets on Tom's in general when talking about CPUs.

Although they are used for most high-end GPU benchmarks...

 

[Novuake]

Hey mayankleoboy1, I run Gentoo on my FX 8350 and I generally see a 10% speed up at least on everything. Blender is the exception as it is more than twice as fast in Gentoo than the official version from Blender.org.

I actually beat a 3930k @ 4ghz with my FX 8350 at 5ghz by about 30% when we both rendered a demo file in Cycles, with me on FX 8350 in Gentoo optimized and him on Windows.

So yes, I expect Intel to downplay the importance of AVX, FMA, etc.

[citation][nom]shadowfamicom[/nom]Would love to see more comparison to 2011 socket CPUs. Even just the 4 core i7-3820 would be an interesting comparison.[/citation]

There also seems to be a big lack of comparisons to 2011 sockets on Tom's in general when talking about CPUs.

My reply to you will be the same. There are MANY comparisons between Ivy and SB-E on the internet, add 10% average to Ivy's result and you have your comparison.

 

[InvalidError]

[citation][nom]ojas[/nom]IIRC, there won't be any broadwell for the desktop at all.[/citation]
Just because most Broadwell models are BGA does not make them non-desktop. The majority of people will never upgrade the CPU on their motherboard without upgrading the motherboard and RAM anyway, even more so now that systems have a useful life exceeding 5 years for most non-gamers, well beyond most platforms' market lifespan.

Personally, I do not see BGA as a problem for mainstream PCs aside from the nightmares it may cause to motherboard manufacturers who are going to have to split hairs between board features and CPU selection. Then again, they could take a page from history and use a split motherboard design like at least one PC manufacturer used to do in the 90s: CPU, RAM, VRMs, cache, etc. on one card, expansion slots and IO ports on the bolted-down card... or in more recent history, this could be viewed as Slot-1/A on steroids.

http://www.techpowerup.com/177817/Intel-Haswell-and-Broadwell-Silicon-Variants-Detailed.html
Apparently, there may be some LGA1150 Broadwells but Techpowerup does not expect them to cover models under $200, which is almost the same thing I said except they set the bar $50 lower than I did. (I wrote the post you replied to before reading that article.)

As for Skylake bringing back LGA, this is only a rumor at this point and it could only be for a subset of models and only for a generation or two, not a permanent full-scale thing.

 

[edlivian]

i am only switching mobo platforms when there is a true chipset advancement, like ddr4 or ddr5, or 10gb/s usb 3.0 . There is nothing here that I would call a true advancement in tech.

 

[Novuake]

i am only switching mobo platforms when there is a true chipset advancement, like ddr4 or ddr5, or 10gb/s usb 3.0 . There is nothing here that I would call a true advancement in tech.

USB3 10Gbps might be happening soon without an actually new chipset if all goes well... CANT WAIT!!! I use USB way to often.

 

[blazorthon]

[citation][nom]blazorthon[/nom]They do not consume more power. They use less power. The VRM is now integrated into the CPU instead of on the motherboard, so their TDPs are higher (which I noticed you based your power consumption number on, so I'll state yet again that TDP is NOT power consumption and shouldn't be treated as if it is) and sure, the chips themselves may use more power, but system power consumption went down because Haswell's motherboards use much less power.[/citation]
To be honest, i don't think the VRM has anything to do with it. All models have:

1) An integrated VRM.
2) Turbo frequencies in the same range (especially the 4770/K and S models).
3) Identical graphics cores with almost identical clocks.

Yet the TDPs are so different. Something else is going on here. No, i'm not suggesting they'll consume more power. I think it's to do with overclocking and/or Turbo Boost, and a different TIM b/w the IHS and the silicon.

Think about it:
the i7-4770T and 4765T are identical except the clock speeds. So 500 MHz seems to warrant a TDP increment of 10w.
By that logic, the difference b/w the 4770S and 4770/K should also be 10w or less, not 19w.

BUT! Look at the 4770S and 4330. The 4770S has more cache, a higher clock speed, higher iGPU clock, same number of cores, yet still has a 19w lower TDP! Even the 4570 is too close to the 4770S to warrant a 19w increase in TDP.

My conclusion:
They could have fit the 4330 and 4570 within 65w if they wanted to, and the 4670/K and 4770/K could have easily fit within 75w (or at the most 77w like Ivy).

They've chosen not to. And i haven't been able to figure out why this is the case ever since the first Haswell slide leaked.

Nothing before Haswell has the integrated VRM.

Binning can account for all other differences.

I fail to see how the TIM can affect power consumption negatively unless it is even worse than Ivy's and even then, the impact should be minuscule and only felt if temps get high.

Turbo could have impacted this if it was greatly more aggressive, but even that is a longshot, especially considering Intel's downwards strategy with power consumption.

The fab process should have matured some since Ivy, so if anything, that should improve the silicon quality. An increase in transistors could, in theory, make binning still more important, and would in fact be likely to show us what we see.

On this binning thought train, that a similar pattern is seen in Sandy SKUs and especially Ivy Bridge SKUs is quite supportive of this possibility being true. For example, compare some of the mobile processors and desktop processors. Some of the similar models can have wildly varying TDPs. A more mature and improved fab process can easily account for some of that moving into the desktop CPUs because yields and/or familiarization with working with the fab process should have improved, so desktop CPUs wouldn't all need to get the short end of the stick compared to the mobile CPUs in binning.

Sure, at this point I'm just theorizing too, but at least from my perspective, it simply makes much more sense that this is either all or partially right, especially since Intel's own words support this from multiple angles.

 

[blazorthon]

i am only switching mobo platforms when there is a true chipset advancement, like ddr4 or ddr5, or 10gb/s usb 3.0 . There is nothing here that I would call a true advancement in tech.

USB3 10Gbps might be happening soon without an actually new chipset if all goes well... CANT WAIT!!! I use USB way to often.

A new version of USB would need a new chipset unless the companies/organizations that work on USB manage to find a way to boost performance with the exact same hardware through mere firmware/driver updates like the Firewire guys did for a while (other than using a PCIe expansion card, of course). Have you read something about this that points to such a thing happening or perhaps something else that I've missed?

 

[ojas]

[citation][nom]InvalidError[/nom]Just because most Broadwell models are BGA does not make them non-desktop. The majority of people will never upgrade the CPU on their motherboard without upgrading the motherboard and RAM anyway, even more so now that systems have a useful life exceeding 5 years for most non-gamers, well beyond most platforms' market lifespan.Personally, I do not see BGA as a problem for mainstream PCs aside from the nightmares it may cause to motherboard manufacturers who are going to have to split hairs between board features and CPU selection. Then again, they could take a page from history and use a split motherboard design like at least one PC manufacturer used to do in the 90s: CPU, RAM, VRMs, cache, etc. on one card, expansion slots and IO ports on the bolted-down card... or in more recent history, this could be viewed as Slot-1/A on steroids.http://www.techpowerup.com/177817/ [...] ailed.htmlApparently, there may be some LGA1150 Broadwells but Techpowerup does not expect them to cover models under $200, which is almost the same thing I said except they set the bar $50 lower than I did. (I wrote the post you replied to before reading that article.)As for Skylake bringing back LGA, this is only a rumor at this point and it could only be for a subset of models and only for a generation or two, not a permanent full-scale thing.[/citation]
Never meant that. When this entire broadwell rumor started, that's exactly what i though, that BGA would be confined to mobile/ultra-mobile, while the desktop would still get replaceable LGA parts, since they'd already have the socket in place.
But as more attention went to the issue, i remember Tom's in particular and other sites too saying that there won't be any desktop chips from the Broadwell generation, which is what i mean.

I don't really see BGA as a problem either, like you said, people usually end up buying a new motherboard anyway. BUT. I still don't think there'll be any desktop BGA chips, and the (very interesting) slide you've linked to supports this.

You observe yourself that it'll be a pain for motherboard manufacturers, and no i don't think they'll do a repeat of the 90's, it doesn't seem very viable or cost effective (look the Xi3 systems, that's sort of what they're doing but...you know the price). Neither will Intel suddenly return to making motherboards to Totally Annihilate the other mobo manufacturers, we already know they're going in the complete opposite direction.

And Chris observes:

We know from our talks with motherboard vendors at this year’s CES that you’ll be able to buy Haswell in LGA 1150 trim, but that its successor, Broadwell, is going to be BGA-only (meaning it’ll ship soldered onto motherboards). Now, it’s possible that Skylake, the architecture to follow Broadwell, will see Intel re-introduce an upgradeable interface.

That's consistent with what i've been reading so far, until i read what you had linked to. My interpretation of this remains the same, you're not going to be buying BGA motherboards for the desktop.

Now that slide. Haswell "-DT" parts are obviously desktop parts, -ULT ones are for ultrabooks, while -ULX is for tablets. I'm confused as hell on the -H and -MB stuff, especially because Haswell-H 2M-GT2 and Haswell-MB 2M-GT2 look exactly the same. I'm guessing that the 2M -MB chips will go into high-end gaming/workstation notebooks with discrete graphics, -H into high-end non-gaming, the GT-3 -H variant goes into high-end notebooks with integrated graphics while the 2 core -MB will be for the i3/Pentium notebooks.

I must say i'm surprised by the lack of GT3e in the charts. Perhaps it's considered a subset of GT3? I dunno, more like the other way around.

Now let's look at Broadwell. As you state, the LGA parts will likely be $200+ quad cores if the slide is accurate and these will eventually exist. Which makes sense, since LGA1150 boards would already be in circulation.
But note that these are the only CPUs marked as -D, which implies "desktop", obviously. In fact, looking at the 8MB cache, these are likely to be Core i7s only.

-H and -M models seem identical (Both seem to be i7s, again based on the L3 cache size), but i'd guess that -H will get GT3e while the -M will get GT2 so that it can be used in gaming/workstation notebooks.

-U will be for ultrabooks, most likely a hyper threaded i5, or maybe on lower models, a Core i3.
-Y will be for tablets, like it is for Ivy. Again, dual core, hyper threaded i5 is the most likely candidate.

(i find it incredibly fishy that there's no mention of -Y parts for Haswell)

So there you go. Still no Broadwell for desktops in a wholesome sense. Whoever wants to buy an i7 can, and let's be honest, they're the only ones likely to upgrade every year for 10% performance gains, since they're more likely to be lose money over lost time, and of course they'd likely be the only ones with the kind of money to spend on a $300+ CPU every year. I said "likely", so don't kill me if you can do that but don't because it doesn't make sense for you

The sub $200 market (more like, the sub $260 market) doesn't get Broadwell at all, they'll get Skylake.

I rest my case.

p.s. The slide still feels like it's not giving the full picture, and no, i don't mean the "?" next to "GT" for Broadwell SKUs.

p.p.s. I just realised that the 8MB cache size doesn't indicate much, they'd obviously be binning them. So yeah, you're more accurate with the $200+ price range. Though i'd suppose it should now be $180+, since i5s occupy that price bracket as well.

p.p.p.s. I remember Anand (AnandTech) saying that Broadwell will be a bigger change from Sandy/Ivy than Haswell is, so i'm not sure if motherboard manufacturers can expect 8-series boards to work with Broadwell chips. So i'm not sure if they're not telling Tom's Hardware everything and pretending there won't be any LGA chips, or whether that slide is inaccurate, or whether Broadwell LGA chips will work fine with 8-series boards. Ok, this is getting a bit nuts now. :lol:

 

[blazorthon]

p.p.p.s. I remember Anand (AnandTech) saying that Broadwell will be a bigger change from Sandy/Ivy than Haswell is, so i'm not sure if motherboard manufacturers can expect 8-series boards to work with Broadwell chips. So i'm not sure if they're not telling Tom's Hardware everything and pretending there won't be any LGA chips, or whether that slide is inaccurate, or whether Broadwell LGA chips will work fine with 8-series boards. Ok, this is getting a bit nuts now. :lol:

FX was a huge change in architecture coming from Phenom II, Athlon II, and Sempron along with the Opteron variants, yet they managed to keep the same socket. What the type of change rather than how different they are seems more important for socket changing. For example, if Intel integrates any more of the motherboard's components into their CPUs such as the rest of the VRM and such, then sure, a socket change may be necessary. However, if Broadwell is merely a performance changer through a more differentiated architecture, then it could be no less compatible with Haswell's platforms than Ivy is with Sandy Bridge's platforms.

 

[Junoh315]

I just got the 3770k and I haven't been able to get it past 30% using video editing software and gaming. I don't think I'll need the 4770k any time soon. Also, my motherboard is 1155 and I love my motherboard.

 

[InvalidError]

[citation][nom]ojas[/nom]You observe yourself that it'll be a pain for motherboard manufacturers,[/citation]
There is a simple solution for motherboard manufacturers who decide to offer desktop boards with BGA Broadwell: quit making minor variations of every model and only offer the most popular and cost-effective subset of available CPU models only in the most popular form factor for each particular performance tier... so a motherboard manufacturer's Broadwell lineup could end up looking like:
- ITX: Pentium 4120, i3-5120
- micro-ATX: i3-5225, i5-5570
- ATX: i5-5670

Even if Intel did not make standard-power Broadwells specifically for desktop, I doubt motherboard manufacturers would skip a whole product cycle only due to lack of socket. They would certainly release far fewer models though, effectively dictating what you can and cannot buy and in which form factors you are allowed to do so.

Since so many motherboard models have only minute differences between them, I personally would welcome BGA CPUs forcing motherboard manufacturers to quit splitting hairs. No need to worry about supporting power delivery from Celeron all the way to overclocked i7k on every board model either since the CPU is predetermined at assembly time and non-overclockable. No need to support multiple chipset-CPU mixes either since the IO Hub will be on the CPU package for BGA parts.

While losing the socket may be scary at least for some, it might actually do a fair bit of good at the mainstream level. For the average person, this is no worse than choosing a smartphone, tablet or laptop where picking a model already sets just about every hardware spec in stone.