http://wccftech.com/intel-14nm-broa...ially-launching-2nd-june-benchmarks-revealed/
Two weeks from Monday if accurate...
Two weeks from Monday if accurate...
![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
Broadwell Launch Date June 2nd
- Thread starter Ultima99
- Start date
drescherjm
[H]F Junkie
- Joined
- Nov 19, 2008
- Messages
- 14,941
I would hope expect it to be priced lower than the i7 4790k since the CPU portion performs lower at stock. However it will be more energy efficient and have a better GPU so I am not sure.
This Haswell with similar clock speed, power consumption and Iris Pro graphics (esp. 128MB eDRAM die) cost ~$360 at launch, and might give a good guesstimate of how much Intel values CPUs in that segment.
Or maybe the newness will put it in the ridiculous $500+ list price range at launch.
Or maybe the newness will put it in the ridiculous $500+ list price range at launch.
MacLeod
[H]F Junkie
- Joined
- Jul 28, 2009
- Messages
- 8,280
Barely any gain over Devils Canyon? Man Zen really needs to be something great. We really need some competition in the desktop CPU market.
People need to stop dumping money at Intel and it wouldn't be an issue.
drescherjm
[H]F Junkie
- Joined
- Nov 19, 2008
- Messages
- 14,941
Kind of hard when the alternative is not a really better value when you consider all factors. Although for me the owner of an i7 970 the alternative does not yet make a CPU that is as fast as the CPU that I currently have.
People are going crazy over the little information AMD has released. If the SMT* rumors are true, and that is responsible for much of the 40% performance increase, many people are going to be disappointed. That means single threaded performance still sucks, and most of the theoretical performance gain is simply from HyperThreading-like multi-threading, which varies widely by load.
People are typically going to see most of that "40%" performance gain the same way everyone else sees the "30%" on Intel processors with HT: they don't, but only see around 10-15% at best in most highly parallel applications.
And that's not a negative view: I'll take AMD's word that it significantly improved IPC/throughput in the core, but it's unlikely to suddenly become competitive with such a ghetto design team that's largely been limping along for a decade. SMT is a good bet for improving performance in some ways, and if AMD can finally fix it's cache subsystem limitations (bandwidth and latency at all levels), Zen is going to be a better product than what it currently offers. But as far as challenging Intel at the high end? lol, no.
MacLeod
[H]F Junkie
- Joined
- Jul 28, 2009
- Messages
- 8,280
Right, that's the problem. Why is Intel putting out new CPU's with ~3-4% improvements for the third time straight? Because they can because they don't have any competition. Don't like their 3% improvement? What else are you gonna buy?
WPrime and SuperPI aren't very good CPU benchmarks. It looks like Broadwell is close to release and a wider range of benchmarks should be coming. Broadwell is a die shrink and not a new uarch, and if it's only a few % faster as with other shrink/tweak cycles, that's expected. I don't get the drama over it. Wait for Skylake if you want a new uarch.
Intel could give a one time nice boost by removing the relatively low TDP caps and cranking up the clock speed, but what happens next time? It's back to single digit IPC improvements on legacy software.
I don't think you understand the difficulty in extracting higher IPC on legacy software, or that Broadwell is a shrink of Haswell. It's not magical or throttled just to be mean. It's a hard problem and it's why x86 extensions are important.
Intel could give a one time nice boost by removing the relatively low TDP caps and cranking up the clock speed, but what happens next time? It's back to single digit IPC improvements on legacy software.
drescherjm
[H]F Junkie
- Joined
- Nov 19, 2008
- Messages
- 14,941
I am hoping the 40% is IPC as a result of adding additional APUs ... to support HT.
It doesn't work like that. SMT/HT count on using under-utilized execution resources in the CPU to run additional thread(s). That can happen while one thread is waiting on main memory access, or in more advanced methods, while instruction dependencies create bubbles in the execution pipeline. GPU execution units aren't general purpose enough to be plugged in like that and have different design goals anyways which make them undesirable for that usage (see CPU/GPGPU latency vs throughput design consideration discussions).
drescherjm
[H]F Junkie
- Joined
- Nov 19, 2008
- Messages
- 14,941
Don't they also add additional execution units to create more bubbles? And this over provisioning helps some single threaded code as well as HT.
MacLeod
[H]F Junkie
- Joined
- Jul 28, 2009
- Messages
- 8,280
It's not that I'm going crazy over a rumor, I'm just hoping Zen is good enough to be a viable alternative to Intel so we can get some competition in our segment.
So the fact that Intel stopped introducing CPU lines with anything more than a marginal performance increase happened the same time AMD fell on their face with Bulldozer and became non-competitive is just a coincidence? I guess it is easier to just say I'm too dumb to understand. That could be true because you're right, I definitely don't understand why the i7-5700 is only going to be marginally better than the i7-5700 and somebody that has owned a 2700 for the last 5 years still doesn't have much of a reason to upgrade.
I just think that if AMD was nipping at Intel's heels like the Phenom II line was we'd see more than 3% increases in performance over the last 4 versions of Intel chips.
Considering the length of time CPU development cycles run on, that Bulldozer flopped on release had no relation to what Intel was developing in the pipeline.
The large leaps Intel experienced in CPU performance in the semi-recent past came from 1) Conroe widening and reworking the Core uarch (Core is based on a Pentium 3-like design) and 2) Nehalem adding an IMC. That's it. Haswell widening integer execution resources subsequently gave more modest improvements than Core 2 did compared to Core. Skylake will do this again and the benefits may be even more modest.
Intel deciding in the Conroe era that it wanted to lower desktop power consumption and principles derived from that (improvements must give at least 2% performance gain for 1% increase in power) do have an effect, but that's not due to Bulldozer.
x86 extensions are the next frontier on improving performance, but that doesn't help legacy software. Believe it or not, there's only so much optimization possible on general legacy code. The instructions are already being split into uops, those uops reordered optimally, executed in and out of order, execution path prediction running way above 99% and also prefetching code and data to minimize stalls. Not all problems can be infinitely scaled with more threads, so splitting it between cores wouldn't help. Still, some applications can benefit from more modern compilers which can more effectively target new instruction set extensions, optimize for particular newer chips and the developers can make the code more parallel/scalable when possible.
Upping TDP ranges could give a nice boost to clock speeds, but it does nothing for long term performance improvements. The "K" processors are a good compromise for people who don't care about power consumption.
There's no conspiracy. It's just a difficult problem. To exploit the improvements in new processors, code has to take advantage of it. You can no longer expect fantastic improvements on legacy code performance because the largest bottlenecks have long been mostly fixed (lowering memory latency via IMC, code reordering and minimizing stalls/misprediction). There is very little left to improve on those former bottlenecks.
If AMD were anywhere near competitive, it would most likely go back to higher TDPs to eke out a few hundred more MHz. If you know something the top chip architects don't know about extracting more performance, I'm sure you would be hired in a heartbeat by your choice of company.
Broadwell is a limited release and will likely never see a refresh.
If these prices are remotely accurate...
http://wccftech.com/intel-broadwell...ake-core-i76700-core-i56500-listed-preorders/
http://wccftech.com/intel-broadwell...ake-core-i76700-core-i56500-listed-preorders/
Zangmonkey
Supreme [H]ardness
- Joined
- Jul 6, 2005
- Messages
- 5,062
What sort of performance change/gain would you expect?
Intel is exceeding the performance requirements of the significant majority of their buyers. They're working to reduce the TDP, power requirements and onboard solutions... because those are the hotly demanded industry features.
MacLeod
[H]F Junkie
- Joined
- Jul 28, 2009
- Messages
- 8,280
Im not expecting 40% increases with each launch but from Sandy Bridge to Broadwell we're talking what, a 12, I maybe 15% increase? I mean if you're running a 2700K that's nearly 5 years old there is no reason really to upgrade despite Intel about to release the 3rd line of CPU's since SB. Am I really the only one here that thinks that's weak sauce?
No one is forcing you to upgrade. 
The performance increases depend on the application. Some have large increases, others less. http://www.anandtech.com/show/7003/the-haswell-review-intel-core-i74770k-i54560k-tested/6
SB -> Haswell (3.5GHz)
VS2012 20% increase
TrueCrypt 25% increase
x264 20% increase
PCMark 25% increase
Kraken 15% increase
7-zip 7% (1T), 9% (MT)
POV-Ray 25%
Cinebench 15% (1T/MT)
gaming (seems to be GPU limited, so can't really tell from that link... not that it matters much in practice)
Most of those are not microbenchmarks, so seeing 20% is actually pretty impressive.
Broadwell may have 0 - low single digit percentage increases, but it's just a shrink/tweak of Haswell so that's expected.
The performance increases depend on the application. Some have large increases, others less. http://www.anandtech.com/show/7003/the-haswell-review-intel-core-i74770k-i54560k-tested/6SB -> Haswell (3.5GHz)
VS2012 20% increase
TrueCrypt 25% increase
x264 20% increase
PCMark 25% increase
Kraken 15% increase
7-zip 7% (1T), 9% (MT)
POV-Ray 25%
Cinebench 15% (1T/MT)
gaming (seems to be GPU limited, so can't really tell from that link... not that it matters much in practice)
Most of those are not microbenchmarks, so seeing 20% is actually pretty impressive.
Broadwell may have 0 - low single digit percentage increases, but it's just a shrink/tweak of Haswell so that's expected.
Zangmonkey
Supreme [H]ardness
- Joined
- Jul 6, 2005
- Messages
- 5,062
What is the performance per watt improvement?
MrGuvernment
Fully [H]
- Joined
- Aug 3, 2004
- Messages
- 21,812
Nothing, i think that is the point people need to stop replacing their perfectly fine CPU's for a CPU with such little gain all so they can have a bigger e-peen to brag about..
M
mls1995
Guest
Yeah I'm in the skylake boat. Minimum of 1 build, maybe 2
Skylake was created by the Israeli team, right? Generally, good things come out of that research facility.
From what I've heard, Skylake uses about half the power that Broadwell does at a 5-10% IPC increase. I'm starting to think Skylake make actually be a worthy upgrade step by Intel. It will probably be the chip to get the Sandy Bridge crowd to finally update.
MacLeod
[H]F Junkie
- Joined
- Jul 28, 2009
- Messages
- 8,280
Im sure this matters to some people but to me I couldnt care less. I mean Im running a FX 8350 at 4.8 GHz along with an overclocked 290x. Obviously I dont care about power usage.
Zangmonkey
Supreme [H]ardness
- Joined
- Jul 6, 2005
- Messages
- 5,062
Im sure this matters to some people but to me I couldnt care less. I mean Im running a FX 8350 at 4.8 GHz along with an overclocked 290x. Obviously I dont care about power usage.
Right, I understand.
The point is that Intel is showing real improvements every release with impressive shows of engineering.
It just happens that what they're accomplishing doesn't resonate with you, and you're not the average consumer.
Michaelius
Supreme [H]ardness
- Joined
- Sep 8, 2003
- Messages
- 4,684
Im sure this matters to some people but to me I couldnt care less. I mean Im running a FX 8350 at 4.8 GHz along with an overclocked 290x. Obviously I dont care about power usage.
Well you also seem to not care about performance when you have FX
MacLeod
[H]F Junkie
- Joined
- Jul 28, 2009
- Messages
- 8,280
