Page 1 of 2

Why I won't recommend a Haswell system until August+

Posted: Fri Apr 05, 2013 8:30 am
by CA_Steve
For all of you early adopters: Intel released a Product Change Notification for all of the Haswell chipsets (found this at Hardwareinfo). There's a bug with the USB 3 implementation causing problems when waking from sleep. Intel's fix is the C2 stepping, which won't be qualified until July 1 and in mobo's end July. The PCN can be found here.

Note this is a chipset issue and not a CPU problem.

Re: Why I won't recommend a Haswell system until August+

Posted: Mon Apr 08, 2013 8:27 am
by Scrooge
That... looks extremely minor. I don't like using sleep anyway.

Re: Why I won't recommend a Haswell system until August+

Posted: Mon Apr 08, 2013 8:46 am
by Abula
Im going to wait till August mostly to see more mature BIOS, and hopping the GENE VI gets a C2 version... but will see, i dont use sleep at all on my desktop, so i dont care much.

Re: Why I won't recommend a Haswell system until August+

Posted: Mon Apr 08, 2013 9:23 am
by CA_Steve
Supposedly, it just affects some USB 3 devices like thumb drives and not keyboard/mouse. That said, I use sleep mode all the time during the day. Before I walk away from my PC, I put it to sleep which dismounts my Truecrypt drive along the way.

But yeah, in general, I stay away from newly released chipsets/mobos. I don't like being the beta tester. Too many BIOS spins happening in the first 2-3 months of release...let alone what hardware bugs there may be.

Re: Why I won't recommend a Haswell system until August+

Posted: Sun May 12, 2013 6:41 pm
by DeltaForce
It's been a long time since my last build, and I have a new case, and was almost done researching parts for an Ivy Bridge build, but am now wondering if waiting for Haswell makes more sense? This is for a medium-high performance PC.

The lower power consumption of Haswell seems like a plus, but does that necessarily translate into less heat? And possibly less fan noise (in a comparison where the heat required higher than min fan speed in Ivy Bridge)?

Reliability is important to me, and is the August date the earliest where there is a good probability of a stable chip and chipset without any real problems?

One thing that appealed to me about Haswell was the improved integrated graphics. I've had problems with GPU cards in the past and liked the option of being able to run off the motherboard temporarily if there is any problem with a GPU card.

On the other hand, would like to get a new machine put together, and this would be another delay.

Any thoughts would be appreciated.

Thanks.

Re: Why I won't recommend a Haswell system until August+

Posted: Sun May 12, 2013 6:59 pm
by CA_Steve
I don't think there will be a crash and burn level problem at intro, but there will be the return-from-sleep issue with USB 3 connected devices for all mobos purchased before ~August. And you never know if your particular build will be stable with early release BIOS.

There's always risk when you are an early adopter. It seems like every CPU/chipset released has some kind of bug the necessitates a silicon spin. If you take a look through any motherboard mfgr's BIOS downloads, the revision history tells a similar story - lots of revisions in the first few months and then things settle down.

As for Haswell power use, there should be significant idle power savings. At heavy load, the CPU side will probably be similar to Ivy Bridge. At light load, the CPU power use should be less (with the added states for better dynamic control). The GPU is bigger, hence the 84W TDP vs 77W for the i5/i7 desktops. Figure under similiar 2D loads, the power use will be similar..or perhaps less for Haswell. For heavy GPU loads, the power use will be higher.

My big hope is that Intel improved the die/package interface, so Haswell will run cooler than IVB at similar loads.

Re: Why I won't recommend a Haswell system until August+

Posted: Sun May 12, 2013 9:01 pm
by xan_user
CA_Steve wrote: I don't like being the beta tester. Too many BIOS spins happening in the first 2-3 months of release...let alone what hardware bugs there may be.
Gotta love how many people line up to pay top dollar to be beta testers tho.... they help keep the economy afloat.

Re: Why I won't recommend a Haswell system until August+

Posted: Sun May 12, 2013 10:15 pm
by QUIET!
If you haven't bought yet, you can try to put together a budget.

If you want to keep looking forward, Haswell is going to miss ddr4 and Intel might be on the wrong side of the HSA thing but at some point you're going to want to stop waiting.

I think ivy bridge has some life left in it and you can get it now.

My prediction is that for desktop and server usage Haswell won't be much more of an upgrade to ivy bridge than ivy bridge was an upgrade to sandy bridge.

If you are concerned about power usage, I think you'll get more bang for the buck spending on a good power supply than a Haswell upgrade.

Re: Why I won't recommend a Haswell system until August+

Posted: Mon May 13, 2013 5:35 am
by Abula
QUIET! wrote:My prediction is that for desktop and server usage Haswell won't be much more of an upgrade to ivy bridge than ivy bridge was an upgrade to sandy bridge.
The problems is that Broadwell (tock of Haswell) dont seem much better aside from a die shrink, probably will consume slightly less power, the only thing we can look forward is sata express maybe ddr4 (i kinda doubt this). So probably the worthy platform to change for people with sandy/ivy will be until skylake.... two years from now. Im still going to upgrade my gaming PC, out of wanting some things like FanXpert2, PCIe 3.0, trim on raid arrays, hyperthreading.

The problem that i see, is i feel Intel could have delivered a better CPU much powerful as Sandy Bridge was in its time, but the lack of competition is creating like a slack on intels part where they are focusing on things like iGPU and lower power mobile CPUs, and broadwell and skylake can be very similar if AMD doesn't push their cpu line.

Re: Why I won't recommend a Haswell system until August+

Posted: Mon May 13, 2013 7:14 am
by CA_Steve
Typically, we see 10-15% improvement in CPU performance year to year. Even Sandy Bridge over Lynnfield. The big improvements have been in the other features like integrated SATA 3, USB 3, GPU performance going from utter crap to something you can do low end gaming on, etc. But, even lowly 10-15% CPU performance jumps add up over time. I wouldn't toss out an IVB build for Haswell. But, my old e8400 looks like a Model T compared to either. :) So, Haswell is in my future...but not until Aug/Sept.

I think the days of expecting a 40-50% jump in PC CPU performance are over until the next significant breakthrough in chip design/materials.

Re: Why I won't recommend a Haswell system until August+

Posted: Mon May 13, 2013 7:34 am
by nutball
Abula wrote:The problem that i see, is i feel Intel could have delivered a better CPU much powerful as Sandy Bridge was in its time, but the lack of competition is creating like a slack on intels part where they are focusing on things like iGPU and lower power mobile CPUs, and broadwell and skylake can be very similar if AMD doesn't push their cpu line.
I think it's a bit simplistic to blame lack of competition for the slow increase in single-threaded performance. More likely it's that improving single-thread performance is just plain difficult these days. All the easy stuff has been done, a long time ago.

For a lot of users raw single-threaded CPU performance isn't as important as the other things you list, neither are more cores. For many people what we've got already is more than enough. The bottlenecks in many desktop PCs lie elsewhere (ie, a 2-core PC with SSD will feel much faster than a 4-core PC with an HDD).

Even in application such as large data-centres, more cores and lower power can be seen as higher priority than single-threaded performance. Obviously the HPC crowd care, they want more cores and more performance per core, but they're not the bulk of the market.

Re: Why I won't recommend a Haswell system until August+

Posted: Mon May 13, 2013 11:04 am
by yakuman
I've been trying to keep up with Haswell, but there are two areas that are unclear regarding the R models with Iris Pro 5200 graphics, their flagship iGPU.

1) There are reports that these will only be available to AiO systems and small form factor motherboards (have to buy entire pre-assembled systems). While there others who claim they'll also be available for desktop forms (can be purchased as parts as you would on traditional custom PC builds). Which is true?

2) Since these will only come in BGA packaging (i.e. soldered to motherboard instead of LGA 1150 socket), will there be a need for any special heatsinks compatibility-wise or will existing 3rd party heatsinks work?

Re: Why I won't recommend a Haswell system until August+

Posted: Mon May 13, 2013 12:56 pm
by CA_Steve
It's all wild assed guessing for now. Mine is mobo mfgrs will sell a few versions of soldered CPU + mobo. Whether or not the stock Intel cooler is forced upon them or if they are free to pursue their own or go bare is tbd. Also don't know if the BGA layout will conform to the same cooler exclusion area / mounting points as Socket 1150/55/56.

Re: Why I won't recommend a Haswell system until August+

Posted: Mon May 13, 2013 3:56 pm
by QUIET!
If you believe the road maps (and I see no reason not to), the good graphics edram versions are all going to be mobile parts. They will probably show up in a NUC or something but they won't be typical PC parts.

It makes sense, people who want good graphics for gaming will still buy discrete GPUs and people who just want to play free cell and stream YouTube will do just fine on the graphics that are in the desktop parts.

In terms of upcoming die shrinks, fin FETs are going to bring scaling back all the way down to 10nm or less because leakage is pretty much solved. I think the thing that is running out of steam is Intel's ability to gain 10% between die shrinks which is why I think Haswell will be underwhelming.

I like their direction of going more toward an SOC by integrating the voltage regulation, if they do it right it will make board layout a lot easier for the motherboard companies and they can get the low power states right while they are at it.

I hope they went back to the Sandy Bridge style IHS attachment, it would help get rid of the heat for over clocking and might be required with those voltage regulators.

Re: Why I won't recommend a Haswell system until August+

Posted: Mon Jun 03, 2013 11:09 am
by CA_Steve
Looks like Hardware.info was able to reproduce the USB 3 return-from-S3 bug in 14 of 22 USB drives. Not a killer, just an annoyance.

Also from Hardware.info, a nice roundup of 33 mobos. Same ballpark in idle power used when compared to IVB but not as wide of a variance from the best to worst.

Same horrible TIM method used as IVB. Haven't seen a lot of thermal results. Came across this at Hardware.fr(in googlish). Looks like the i5/i7 run a few degrees warmer at stock vs IVB under Prime 95. The upside is stable underclock at 0.9-0.95V at stock freq will reduce CPU power by 20-25W which should really help with temps.

My plan was to get an i5-4670K...I'm now thinking about getting the lowest price quad core instead and undervolting it at stock freq. Keeping it for a year and then selling it on ebay/replacing it with the Broadwell equivalent to the i5-4670K. :)

Re: Why I won't recommend a Haswell system until August+

Posted: Mon Jun 03, 2013 11:18 am
by Abula
CA_Steve wrote:My plan was to get an i5-4670K...I'm now thinking about getting the lowest price quad core instead and undervolting it at stock freq. Keeping it for a year and then selling it on ebay/replacing it with the Broadwell equivalent to the i5-4670K. :)
Im going for 4770K and just skipping broadwell, well i might refresh my HTPC for broadwell with better consumption, but my main pc will stay with haswell, unless there is sata express or ddr4 support on it, else skylake will be my next upgrade.

Re: Why I won't recommend a Haswell system until August+

Posted: Mon Jun 03, 2013 12:34 pm
by CA_Steve
Nothing I do needs 8 threads. :)

Re: Why I won't recommend a Haswell system until August+

Posted: Mon Jun 03, 2013 11:58 pm
by QUIET!
Abula wrote:
CA_Steve wrote:My plan was to get an i5-4670K...I'm now thinking about getting the lowest price quad core instead and undervolting it at stock freq. Keeping it for a year and then selling it on ebay/replacing it with the Broadwell equivalent to the i5-4670K. :)
Im going for 4770K and just skipping broadwell, well i might refresh my HTPC for broadwell with better consumption, but my main pc will stay with haswell, unless there is sata express or ddr4 support on it, else skylake will be my next upgrade.
Don't forget to consider AMD, they should come come out with some compelling products because they have to.

I have a feeling the PS4 might replace a lot of computers...

14nm is going to be a big deal if anyone can come up with an illumination system that can support good wafer throughput, GF/IBM or TSMC might beat Intel to the punch on production.

AMD probably should beat Intel in adopting DDR4 because APUs can use all the memory bandwidth they can get, while Intel tries embedded dram to increase IGP performance.

The next few years should be interesting.

Re: Why I won't recommend a Haswell system until August+

Posted: Tue Jun 04, 2013 12:04 am
by Abula
QUIET! wrote:
Abula wrote:
CA_Steve wrote:My plan was to get an i5-4670K...I'm now thinking about getting the lowest price quad core instead and undervolting it at stock freq. Keeping it for a year and then selling it on ebay/replacing it with the Broadwell equivalent to the i5-4670K. :)
Im going for 4770K and just skipping broadwell, well i might refresh my HTPC for broadwell with better consumption, but my main pc will stay with haswell, unless there is sata express or ddr4 support on it, else skylake will be my next upgrade.
Don't forget to consider AMD, they should come come out with some compelling products because they have to.

I have a feeling the PS4 might replace a lot of computers...

14nm is going to be a big deal if anyone can come up with an illumination system that can support good wafer throughput, GF/IBM or TSMC might beat Intel to the punch on production.

AMD probably should beat Intel in adopting DDR4 because APUs can use all the memory bandwidth they can get, while Intel tries embedded dram to increase IGP performance.

The next few years should be interesting.
I really hope you are right, i do want amd to come back to the scene and compete, if they do release something competitive ill give a shot, but i somewhat doubt it, seems like AMD already gave up competing with intel on the high end cpus. AMD Issues Roadmap Update: Nothing really new in the pipeline ?

Maybe i spoke too soon, even though TDP is way to high, seems AMD is releasing new FX to try to compete with haswell, Centurion is the AMD FX-9000 processor, does 5.00 GHz

Re: Why I won't recommend a Haswell system until August+

Posted: Tue Jun 04, 2013 6:20 am
by QUIET!
Compelling does not necessarily mean competitive in a raw performance sense.

On the other hand, AMD has been hiring back a lot of the people who left while they were on top.

A friend of mine worked for AMD back in the day and he was upset with the company for letting a few of their rock star design engineers get away. Evidently those departures mark the point where performance started to stagnate.

We are in a brave new world where mobile performance is more important for volume CPU sales and AMD APUs still have better integrated graphics. So the question is will Intel catch up on graphics or will AMD catch up on CPU first?

Re: Why I won't recommend a Haswell system until August+

Posted: Tue Jun 04, 2013 8:16 am
by HFat
It has come to my attention that some SPCR posters do not realize how great is the discrepancy between Intel's and AMD's financial situations.
Intel has the best technology which is why Intel's CPUs are so expensive. That will not change for the forseeable future. If threatend, Intel may simply drop its prices and cut down on the artifical market segmentation. So the only question is, how high will Intel's profits be? No one likes to pay the Intel tax and the better AMD does, the lower it'll be. Go AMD!

Re: Why I won't recommend a Haswell system until August+

Posted: Tue Jun 04, 2013 11:18 am
by CA_Steve
I have a hard time believing AMD will be able to catch up to Intel on the desktop. Their single threaded performance/Watt is abysmal. The big threat to Intel is on the low power side, specifically ARM.

Re: Why I won't recommend a Haswell system until August+

Posted: Tue Jun 04, 2013 11:18 am
by Matthew9226
Well I'm an Intel fan. Quite a big Intel fan. But I care for perf per watt and mainly just perf..
I seriously DO hope AMD to get back in the game.
At least with iGPU's they rock. I still don't like their APU's power consumption though.
My updated HTPC pulls 95w with IBT+Furmark going on full blast. HKEPC's review of the A10-6800K shows it pulling 180w+.....
Of course the CPU perf of that thing is higher, especially multithreaded perf but I haven't seen any A6 numbers yet.
Can't wait to see what Kaveri will be like.
/OT

Re: Why I won't recommend a Haswell system until August+

Posted: Wed Jun 05, 2013 3:11 pm
by smilingcrow
CA_Steve wrote:I'm now thinking about getting the lowest price quad core instead and undervolting it at stock freq. Keeping it for a year and then selling it on ebay/replacing it with the Broadwell equivalent to the i5-4670K. :)
There have been strong rumours that Haswell’s replacement will be BGA only for the desktop but the platform after that will still be socketed.
QUIET! wrote:Don't forget to consider AMD, they should come out with some compelling products because they have to.
On that basis they should have come out with compelling products a long time ago which hasn’t happened outside of recent APUs. It’s called wishful thinking.
QUIET! wrote:I have a feeling the PS4 might replace a lot of computers.
In what way does that make sense?
QUIET! wrote:14nm is going to be a big deal if anyone can come up with an illumination system that can support good wafer throughput, GF/IBM or TSMC might beat Intel to the punch on production.
That made me think of the cover of the Pink Floyd album Animals.
QUIET! wrote:AMD probably should beat Intel in adopting DDR4 because APUs can use all the memory bandwidth they can get, while Intel tries embedded dram to increase IGP performance.
The problem is new memory technologies tend to be very expensive especially if there is a slow ramp which would be the case if AMD adopted DDR4 before Intel.
So AMD would gain performance but at the cost of higher prices. That trade-off would make more sense on mobile as on the desktop better to save money with DDR3 and buy a dGPU with the money saved.
QUIET! wrote:So the question is will Intel catch up on graphics or will AMD catch up on CPU first?
Intel already seem to have the GPU so the question is more a matter of how far they want to push it next generation? I say that because Intel still aren’t allocating much TDP to graphics but if they decided to do so at 14nm will AMD be able to compete in hardware terms?
A 100W 14nm Haswell upgrade which can allocate the TDP very dynamically should be quite a monster.
If I was AMD I’d be sweating because you know that your competitor is holding back when they could be sitting on a royal flush.
So yeah, interesting times. :)
HFat wrote:It has come to my attention that some SPCR posters do not realize how great is the discrepancy between Intel's and AMD's financial situations.
It has come to my attention that some SPCR posters wear tin foil hats.

Re: Why I won't recommend a Haswell system until August+

Posted: Thu Jun 06, 2013 11:26 am
by Tzupy
It has been reported, but not confirmed yet, that the retail Haswells run hotter than the review samples.
It is unclear if they use more power or the IHS to die space is larger than in the review samples. Buyers beware.

Re: Why I won't recommend a Haswell system until August+

Posted: Thu Jun 06, 2013 11:38 am
by Abula
Tzupy wrote:It has been reported, but not confirmed yet, that the retail Haswells run hotter than the review samples.
It is unclear if they use more power or the IHS to die space is larger than in the review samples. Buyers beware.
Im intersted on this, into reading more, maybe you have a link to threads of users that are reporting this.

Re: Why I won't recommend a Haswell system until August+

Posted: Thu Jun 06, 2013 12:03 pm
by yakuman

Re: Why I won't recommend a Haswell system until August+

Posted: Thu Jun 06, 2013 2:31 pm
by CA_Steve
yakuman wrote:I think the main reason is due to on-die VRM.
+1.

The larger GPU probably impacts it a bit, too. That said, for stock voltage, it's only a couple of degrees higher when running Prime 95. Nonetheless, tt's unfortunate that Intel opted not to fix the crappy TIM/IHS issue. As mentioned earlier, it looks like you can do some decent undervolting at stock freq. A 0.1V drop on 1.1V base is ~17.5% drop in power. That should drop a 40C rise over ambient to ~33C rise.

Re: Why I won't recommend a Haswell system until August+

Posted: Thu Jun 06, 2013 3:04 pm
by yakuman
Really disappointing. Haswell was my planned build because of Ivy Bridge's heat problem. It's ironic that die shrinks are supposed to reduce heat, but each preceding generation has been increasing heat :x. On the other hand I'm kind of tired of waiting for the next generation.

Re: Why I won't recommend a Haswell system until August+

Posted: Thu Jun 06, 2013 6:45 pm
by QUIET!
smilingcrow wrote:
CA_Steve wrote:I'm now thinking about getting the lowest price quad core instead and undervolting it at stock freq. Keeping it for a year and then selling it on ebay/replacing it with the Broadwell equivalent to the i5-4670K. :)
There have been strong rumours that Haswell’s replacement will be BGA only for the desktop but the platform after that will still be socketed.
QUIET! wrote:Don't forget to consider AMD, they should come out with some compelling products because they have to.
On that basis they should have come out with compelling products a long time ago which hasn’t happened outside of recent APUs. It’s called wishful thinking.
QUIET! wrote:I have a feeling the PS4 might replace a lot of computers.
In what way does that make sense?
QUIET! wrote:14nm is going to be a big deal if anyone can come up with an illumination system that can support good wafer throughput, GF/IBM or TSMC might beat Intel to the punch on production.
That made me think of the cover of the Pink Floyd album Animals.
QUIET! wrote:AMD probably should beat Intel in adopting DDR4 because APUs can use all the memory bandwidth they can get, while Intel tries embedded dram to increase IGP performance.
The problem is new memory technologies tend to be very expensive especially if there is a slow ramp which would be the case if AMD adopted DDR4 before Intel.
So AMD would gain performance but at the cost of higher prices. That trade-off would make more sense on mobile as on the desktop better to save money with DDR3 and buy a dGPU with the money saved.
QUIET! wrote:So the question is will Intel catch up on graphics or will AMD catch up on CPU first?
Intel already seem to have the GPU so the question is more a matter of how far they want to push it next generation? I say that because Intel still aren’t allocating much TDP to graphics but if they decided to do so at 14nm will AMD be able to compete in hardware terms?
A 100W 14nm Haswell upgrade which can allocate the TDP very dynamically should be quite a monster.
If I was AMD I’d be sweating because you know that your competitor is holding back when they could be sitting on a royal flush.
So yeah, interesting times. :)
HFat wrote:It has come to my attention that some SPCR posters do not realize how great is the discrepancy between Intel's and AMD's financial situations.
It has come to my attention that some SPCR posters wear tin foil hats.
On the PS4 front, it might replace a lot of computers because its essentially an 8 core CPU with mid level discrete class graphics that will be in a smallish form factor, hopefully quiet and probably with a high spousal approval factor.

DDR4 is a no brainer. DDR3 and GDDR5 that is based upon DDR3 are running out of steam. DDR4 is in production now, it is only expensive because the volume is low.

Right now Intel needs embedded dram to match AMD APU graphics and you'll most likely never see that in a desktop chip. Once AMD gives the APU some serious memory bandwidth Intel will be playing catch up again because you can only squeeze so much embedded dram in to the package.

Concerning a monster 100 watt 14nm Intel processor, I really doubt that we will see that. Intel always tries to slice the market in to segments where they can squeeze the most money out of their customers (why no k series i3?). On top of that, when you make a fin fet as skinny as 14nm will be, you start having issues with removing heat from the fins so to make a 100 watt part, the die would be huge. I expect to see more 30-40-50 watt parts at 14nm that perform very well and are easy to manufacture, not an engineering tour de force that yeilds 50% and costs a couple grand.

Any way, that's my educated guess, not wild speculation or wishful thinking and my "education" is very much on topic...