Haswell Benchmarks?

[BrightCandle]

Those are some disappointing numbers if true. We were all quietly hoping for a general 10% IPC increase along with our AVX2 and other extensions, as well as better overclocking. If there genuinely is almost nothing in terms of general IPC and only the new instructions and iGPU its looking like a lemon rather than being this godlike CPU that everyone will want. Might be wonderful in a laptop but as a desktop gamer I want higher performance from my CPUs more than I want a power reduction.

 

[Idontcare]

The numbers don't jive with everything Intel has officially said about Haswell's performance, nor does it jive with what Intel employees claim to see on their internal testbeds.

Performance on pre-release hardware cannot be expected to be indicative (to any degree) of performance on release hardware. There is a reason the hardware is not released to market, if it was production-ready then we'd see it for sale at Newegg.

 

[lehtv]

I only do light gaming 1080p on my PC and most of the time I just use it to watch youtube or chatting online. The few games I play right now are diablo 3, starcraft 2 (buying sc2 expansion when release in march) and maybe little bit of GTA IV. I was thinking about getting haswell and radeon 8850 (sea island) when release later in June this year. Guess I'll wait.

Your CPU can handle 7850 crossfire with zero bottlenecking. Hell it'd handle 7950 crossfire in most games as well. Most likely the 8850 or whatever it'll be called will be slower than 7850 CF.

 

[piesquared]

wow, so not good then. Performance is actually worse in SPi! As history has shown time and time again, graphics performance is going to by abysmal when coupled with intel's 'drivers'. This leak sure looks like intel lowering expectations.

 

[MisterMac]

wow, so not good then. Performance is actually worse in SPi! As history has shown time and time again, graphics performance is going to by abysmal when coupled with intel's 'drivers'. This leak sure looks like intel lowering expectations.

Of course.

Just like you said - when OBR leaked BD benchmarks.

Better lower those expectations for

 

[pauldun170]

The charts I have show a 11%-25% improvement across the board with Haswell sample clocked at 3.0 and Ivy Bridge 3770K at stock.
The big improvement is in power draw.
However I have to say that my chart is made up and I created it in mspaint. Only reason why I created it is so that I could put the link to my website on it and distribute it to all the tech sites so impatient internets could up my visitor counts looking for the haswells.

 

[frozentundra123456]

The numbers don't jive with everything Intel has officially said about Haswell's performance, nor does it jive with what Intel employees claim to see on their internal testbeds.

Performance on pre-release hardware cannot be expected to be indicative (to any degree) of performance on release hardware. There is a reason the hardware is not released to market, if it was production-ready then we'd see it for sale at Newegg.

While I agree with you in principle, the leak is disturbing. Reminds me of BD early leaks when people kept saying it could not really be that bad, but it was. So we have to wait and see, but the more I see about Haswell, the more I lower my expectations, for the desktop at least.

 

[BenchPress]

It's utterly impossible for Haswell to be any slower than Ivy Bridge. From Core 2 up to Ivy Bridge, Intel has used 6 execution ports. With Haswell, they're adding not one, but two more execution ports!

They wouldn't do that unless they're convinced they can achieve up to 33% increase in peak instruction throughput. In fact for some workloads which are bottlenecked by instructions that require execution port 0, performance could in theory double since the new execution port 6 can take most of the same instructions. It will specifically be a big boost for software that takes advantage of Hyper-Threading. So the 10% number that is floating around, is quite conservative.

Also they added AVX2 support, which doubles both the integer and floating-point vector throughput (for multimedia and game physics and such). And on top of that Haswell features TSX for much improved multi-core efficiency, and the internal bandwidth has doubled to keep this number cruncher well fed. All of this has been achieved without compromising anything.

Anyone spreading rumors about Haswell not being substantially faster, is making a fool of himself (either out of ignorance or to get rid of Ivy Bridge overstock).

 

[piesquared]

Of course.

Just like you said - when OBR leaked BD benchmarks.

Better lower those expectations for

Yep, the damage is done for haswell. Expectations are 2-3-4-5x graphics performance, 20% cpu performance increase, 2x lower power consumption. Anything worse than that and it will ultimately be a failure. Looks like intel lost control of the hype machine.

 

[piesquared]

It's utterly impossible for Haswell to be any slower than Ivy Bridge. From Core 2 up to Ivy Bridge, Intel has used 6 execution ports. With Haswell, they're adding not one, but two more execution ports!

They wouldn't do that unless they're convinced they can achieve up to 33% increase in peak instruction throughput. In fact for some workloads which are bottlenecked by instructions that require execution port 0, performance could in theory double since the new execution port 6 can take most of the same instructions. It will specifically be a big boost for software that takes advantage of Hyper-Threading. So the 10% number that is floating around, is quite conservative.

Also they added AVX2 support, which doubles both the integer and floating-point vector throughput (for multimedia and game physics and such). And on top of that Haswell features TSX for much improved multi-core efficiency, and the internal bandwidth has doubled to keep this number cruncher well fed. All of this has been achieved without compromising anything.

Anyone spreading rumors about Haswell not being substantially faster, is making a fool of himself (either out of ignorance or to get rid of Ivy Bridge overstock).

Exactly, I expect haswell to be at least 2-3x faster than ivy bridge across the board! Graphics performance is going to utterly destroy everything else on the market for the next 1-2 years.

 

[inf64]

While I agree with you in principle, the leak is disturbing. Reminds me of BD early leaks when people kept saying it could not really be that bad, but it was. So we have to wait and see, but the more I see about Haswell, the more I lower my expectations, for the desktop at least.

Well for one heck of an optimized core such as SB,it's not easy to get gains in legacy code that are going to wow anyone. Haswell may look better when compared to SB than when compared to IB. In the former case it will probably end up more than 10% faster at around the same clock. Power draw gains will probably allow it to run at higher bin Turbo too ,which may allow it to more often than not beat 3770K by more than 10% (combined IPC+clock). So ~10% more x86 performance with less power draw is not bad at all IMO

PS Also for AVX2 optimized code the speed up will probably exceed 50%.

 

[Blandge]

*Sigh*

Must we go through this before every launch? Leaked benchmark scores are not a good indicator of production performance primarily because of BIOS, FIRMWARE AND DIVERS. And that is completely ignoring any bugs that may exist in the silicon or reference boards. Many of these ESs have to be tuned each time they are powered on, so some review website that gets their hands on ES silicon isn't going to be able to be able to glean much of any useful information with simple benchmarks.

The delta in performance from ES with early software to QS with produciton softeware is massive. The same goes for every piece of silicon produced by any company.

 

[tweakboy]

Bring on the 6 and 8 core Ivy E enthusiast desktops. I think 10 and 12 core are for Xeon Ivy E server chips. Tho Im not shure of this. gl

 

[Puppies04]

Exactly, I expect haswell to be at least 2-3x faster than ivy bridge across the board! Graphics performance is going to utterly destroy everything else on the market for the next 1-2 years.

Can I just confirm you are expecting a 200-300% speed increase across the board and for it to have faster onboard graphics than AMDs APUs? I haven't seen a single piece of information that points to anything like those sorts of numbers.

 

[inf64]

Can I just confirm you are expecting a 200-300% speed increase across the board and for it to have faster onboard graphics than AMDs APUs? I haven't seen a single piece of information that points to anything like those sorts of numbers.

I think he is obviously being sarcastic there and he is clearly referring to impossible gains(~2-3x) in x86 part of the chip and not GPU part . GPU part is going to be much faster than HD4000,especially the mobile version of it. There is no doubt about it.

 

[moonbogg]

Wake me up in 6 months when its ready with real benchmarks.

 

[Ferzerp]

Can I just confirm you are expecting a 200-300% speed increase across the board and for it to have faster onboard graphics than AMDs APUs? I haven't seen a single piece of information that points to anything like those sorts of numbers.

You can tune anything out that that poster says if it involves AMD, Intel, or Nvidia.

If AMD : It will be spun so much that the truth (if it is there), will be impossible to find, to make AMD look good (except at this point, he fools no one because he's so blatant).

If Intel or Nvidia: It will be hyperbole, or outright falsehood designed to make Intel or Nvidia look bad, even if it some something inconsequential, or actually positive for either of those companies.

All in all, even our open representatives of Intel and Nvidia on this board display less bias. (Ok, the Intel one is a class act, he just avoids participating in any conversation where mudslinging could happen. To me, the best way.)

 

[Jacky60]

This looks crap I need an incentive to replace my 4 year old cpu, these tiny incremental increases are really annoying, I wish AMD were properly in the CPU game as Intel is sitting on it's hands or twiddling thumbs with no PC CPU competition.

 

[mikk]

Totally unrealistic these scores. Ivy Bridge improved IPC by 3% to Sandy Bridge. And look what Intel changed, only a few minor changes were made for Ivy Bridge. Haswell on the other side has much bigger changes included from the infos we got from IDF last year. So the IPC gain must be much bigger than from Sandy to Ivy.

 

[Torn Mind]

Exactly, I expect haswell to be at least 2-3x faster than ivy bridge across the board! Graphics performance is going to utterly destroy everything else on the market for the next 1-2 years.

Conroe happened a long time ago. Doubt it will happen again, unless there is a huge litter of very very very very healthy cats somewhere.

 

[tweakboy]

Your going to get similar results running other benchmarks. Ivy and Haswell are about same speed clock for clock 2.8Ghz both ,,,,,, No change in CPU performance really.. but Im sure the GPU inside CPU is much much faster and lower TDP. thx gl

 

[Puppies04]

Totally unrealistic these scores. Ivy Bridge improved IPC by 3% to Sandy Bridge. And look what Intel changed, only a few minor changes were made for Ivy Bridge. Haswell on the other side has much bigger changes included from the infos we got from IDF last year. So the IPC gain must be much bigger than from Sandy to Ivy.

It really depends where intel is spending its R+D budget, we really aren't seeing IPC being given the front seat performance/watt is a lot more important at the moment. In an ideal world intels desktop and mobile chip departments would be a lot more autonomous but by all accounts they keep moving closer and closer together.

I'm sure we will see performance gains but the only numbers I expect to be mind blowing is performance/watt and to a lesser extent IGP improvements.

 

[Exophase]

True or not, it doesn't really matter because the benchmarks are garbage. No point even mentioning what you expected from AVX2 benefits from software that wouldn't be using it.

Engineering sample chips will often have some things disabled because they aren't working, which can hurt performance. It's also remotely possible (although not especially likely) that performance went down in some non-critical stuff like x87 because of ucode changes.

 

[MisterMac]

Well - some of us are hoping to upgrade on the Tocks.

I'd rather not wait for the 14nm redesign (Skylake?).



Hoping for atleast 15% IPC on desktop vs SandyBridge + perhaps some more oc room than average sandy for a total 20%.


I don't get why people measure against IB?
It gave something unusual... a IPC increase on a shrink.

Should be tock vs tock.

Also hoping the AVX2\decoder pipeline will be a gift that keeps on giving as more software takes use of it.


Altho if leaks continue in this direction - doesn't look too good.

 

[inf64]

Penryn was a tick and had IPC improvement Vs Merom/Conroe . Westmere was kind of a tick but had "IPC" improvements(if you can call it that since it had more cache vs Nehalem). Both of these examples featured ~3-5% IPC improvements with minor tweaks to the cores.

I'd say it's realistic to compare two tocks : SB and Haswell. In this comparison ~10-15% IPC improvement is not unreasonable at all,it's very probable. In AVX2 code it will be much more.