First Sandy Bridge Numbers?

[grimpr]

This isn't a AMD BD thread . Stop baiting. Start a BD topic keep it out of this topic. Were not talking BD. I heard enough with the PhI and how great it would be.

AMD doesn't have AVX VEX prefix Which I assume to be the jit compiler . That is why AMD doesn't have it. Intel doesn't have to share jit compilers with AMD.

Vexprefix has everthing to do with sse2 instructions. Which to me is a hugh deal. 20% IPC improvement not 5% Bob has one so we know. The App running in this video runs all 4 cores with 8 threads . If intel can get all 4 cores to do as your saying with turbo mode I be very surprised because were not seeing that at all.

Thanks for the incentive to read more detailed info about Ct,AVX,the Vex prefix and SSE2, i think i'm beginning to understand where you're getting at. Now as for the SB sample your friend Bob has and the 20% ipc across the board claims you make, if thats from a ES and on very premature bios/microcode support, its very impressive, to say the least.

 

[Nemesis 1]

Yes it is. I haven't a clue whats coming . Bob did recieve an updated M/B tho. The 20% is not across the board but an average. The 20% is I don't want to stick foot in mouth number . But the review sites will all say 20% +

 

[grimpr]

Intel Math Kernel libraries beta are out supporting AVX, isnt Cinebench on MKL code?, if so a simple recompile will double the throughput just fine.

http://software.intel.com/en-us/articles/intel-math-kernel-library-103-beta/
http://www.agner.org/optimize/blog/read.php?i=49#107

 

[Nemesis 1]

Ya a read those already . I hope it takes off faster than sse because that was a hugh boost in performance . When developers seen this they jumped all over the sse instruction set ,SSE 2 was fairly fast develoment because of the performance improvements . The AVX offers 4x the improvement over SSE 2 offered so developers will be all over this. PLUS all SSE2 apps need only a simple recompile using the vex prefix . How sweet these Cpus will be.

 

[grimpr]

Ya a read those already . I hope it takes off faster than sse because that was a hugh boost in performance . When developers seen this they jumped all over the sse instruction set ,SSE 2 was fairly fast develoment because of the performance improvements . The AVX offers 4x the improvement over SSE 2 offered so developers will be all over this. PLUS all SSE2 apps need only a simple recompile using the vex prefix . How sweet these Cpus will be.

SSE2 is the most used extension in apps out there, about 80% roughly speaking, if it only needs a prefix in the code used and a recompile in Intels compiler to run code from 2001 till now passing through Sandys 256bit crunchers and pipelines then things change, for the huge.

 

[Nemesis 1]

Exactly!

 

[grimpr]

Intel News Release.

SANTA CLARA, Calif., June 21, 2010

Intel, FTC File Motion to Suspend Administrative Trial Proceedings.

http://www.intel.com/pressroom/archive/releases/20100621corp.htm

...

Brilliant tactics Intel, now i finally get it. Now, lets move on on to benchmarks....

 

[Nemesis 1]

Ya probably another Billion intel will hand out for No good reason . After all Intel invented the X86 cpu. It was IBM that insisted upon a second source . But this to is off topic.

 

[ModestGamer]

Ya probably another Billion intel will hand out for No good reason . After all Intel invented the X86 cpu. It was IBM that insisted upon a second source . But this to is off topic.

You sure about that ?

 

[IntelUser2000]

The "high-end" Sandy Bridge doesn't do anything special. The 3-channel(or 4 if you believe that arriving on PC) memory controller won't help most users and so will the extra cores.

Increased L3 caches? Nothing special either. There's diminishing returns beyond certain point and bigger caches might have resulted in higher latency.

Expecting more than 5% increase over the "crappy 115x" Sandy Bridge is ridiculous. If Intel is gonna deliver for Sandy Bridge, the very first versions will tell us how its doing.

 

[extra]

The "high-end" Sandy Bridge doesn't do anything special. The 3-channel(or 4 if you believe that arriving on PC) memory controller won't help most users and so will the extra cores.

Increased L3 caches? Nothing special either. There's diminishing returns beyond certain point and bigger caches might have resulted in higher latency.

Expecting more than 5% increase over the "crappy 115x" Sandy Bridge is ridiculous. If Intel is gonna deliver for Sandy Bridge, the very first versions will tell us how its doing.

You're prolly right, but it all adds up. Like gearing up in an mmo...few percent here, few percent there, and wow what a difference. But yeah, I think it's unlikely SB will be more than 10-20% faster per clock...which is a lot imho. Question is how will they clock. And come on Nem, we know you love Intel but even from you, saying they didn't engage in some horribly anti-competitive practices is over the top (but that doesn't have anything to do with this thread)..

Anyway, if Intel can pull off some huge gains from AVX, great. If not, even if its only 10-20% that's pretty friggin nice, Nehalem is already amazing. Apps definitely need to catch up.

The huge question, much bigger imho...is how is the GPU? I love PC gaming and I'm really, really, hoping that it's good enough to play modern titles even if they are at low or medium settings. What's holding PC gaming back from more blockbuster titles is intel gma (imho). One of the reasons WoW is so successful is that it plays okay on GMA. Playing decently on Intel integrated graphics like quadruples the size of your target audience. Please deliver on this Intel!

 

[IntelUser2000]

I'm not saying any numbers in general, but just that people that are saying "wait for the high end Sandy Bridge coming later next year" are going to be disappointed. AVX is going to need recompiling to take advantage so I'm totally discounting AVX.

Intel Math Kernel libraries beta are out supporting AVX, isnt Cinebench on MKL code?, if so a simple recompile will double the throughput just fine.

That may be true, but when I looked up Phenom vs. Athlon X2 comparisons, the addition of faster SSE didn't do much on Cinebench, at least nowhere near what it can do theoretically.

 

[ydnas7]

looking a coolaler's benchmarks the sandybridge varies from same IPC as nehalem (super PI - single threaded application) to about 50% faster (photoworxx - multhithreaded and somewhat memory bound graphical application)
i doubt either of these are AVX aware or using GPU functions
intel was showcasing Picassa as a suitable application for Sandybridge, i guess that is typical application for which they tuned sandybridge. (Ie picasa was not changed, sandybridge is optimized for picassa type applications)

i would also add that my impression of Llanos and Sandybridge is that both companies are seriously going after (creating) the notebook gaming market. If Intel can provide the equivalent of console game on a notebook then they should be happy, likewise ditto for AMD.

 

[ydnas7]

if sandybridge can achieve 30fps on a modern game with full settings on a laptop screen at 1600 x 900 with good battery life, then its a win

at the end of the day both chips cannot go beyond its memory access limits which is 2 channel DDR3 1600 - although cache may change that a bit for sandybridge.

 

[ydnas7]

just going through coolaler's results and worked out a clock per clock equivalent versus nehalem 4C8T (generally triple channel i96x except the superPI being dual channel i750)

everest test 1 33%
everest test 2 55%
superPi 3%
Cinebench 11.5 3%
everest v5.5 5%
everest v5.5 35%
superPi 7%
everest photoworxx 34%
everest AES 18x ie 1,800%
everest 33%
everest 2.25x ie 225%

as i can't read chinese, i can't tell which everest test is which parameter, but it does look like multithreaded performance is significantly increased but single thread performance is not. it also looks like memory bound tests are improved by 1/3.

apart from language issues, there is the turboboost profile uncertainty, as Sandybridge is almost certain to have significantly more core speed variability than current nehalem, these preliminary results are even less useful.

 

[scooterlibby]

I'm not saying any numbers in general, but just that people that are saying "wait for the high end Sandy Bridge coming later next year" are going to be disappointed.

I think you are correct. However, the only thing that makes me want to wait for the 'higher end' SB is the fact that the lower end doesn't do x16 x 16. I like to run multi-GPU setups and this would not work for me.

 

[aigomorla]

if your an overclocker...

aim at a 970X and kiss sandy bridge goodbye.

2 cents... until they refresh SB.

if your a power green freak.. SB is golden.. like cherry pie.

 

[extra]

KK Aigo...too bad but not really unexpected...however...

This is only talking about socket 1155 and laptop stuff, right? We don't really know anything about how the socket 2011 will overclock, right? Or is that bad news as well?

 

[khon]

if your an overclocker...

aim at a 970X and kiss sandy bridge goodbye.

2 cents... until they refresh SB.

if your a power green freak.. SB is golden.. like cherry pie.

The hexacore i7-970 (for some reason there's no X in the name) will cost $800-900. Why would you pay that even if you're an overclocker, when you could get a quadcore SB that will beat it in most applications for something like $300-400 ?

The i7-980X makes sense if you absolutely have to have the best and don't care about performance/price. The i7-970 doesn't seem to make much sense at all to me. Who would buy it ?

 

[aigomorla]

The hexacore i7-970 (for some reason there's no X in the name) will cost $800-900. Why would you pay that even if you're an overclocker, when you could get a quadcore SB that will beat it in most applications for something like $300-400 ?

cuz a 970X overclocked will slaughter it.

 

[khon]

cuz a 970X overclocked will slaughter it.

Only if you're doing applications that are fully multithreaded. And even then the performance gap won't be anywhere near as wide as the price gap.

 

[aigomorla]

Only if you're doing applications that are fully multithreaded. And even then the performance gap won't be anywhere near as wide as the price gap.

excuse me are we looking at overclocking or not?

Because if we ignore overclocking yes, your point is valid.

But i said in regards to overclocking.

And a intel hexcore with at least 200 in its Bclk will just about rape anything out there.

I was showing the forum exactly how bad it did rape everything out there.

AMD 1090T's would need about a 5.8ghz overclock to even stand next to a 4.3ghz Intel Hexcore on Wprime.

Score Board **UPDATED** 07/14 Post #3375


Six Core with HTT

1. rickss69 - Intel Core i7 980X 5319 MHz - 3.003
2. DOM - Intel Core i7 980X 5076 MHz - 3.401
3. fatguy1992 - Intel Core 980X 5279 MHz - 3.447
4. MetalRacer - Intel Core i7 980X 4682 MHz - 3.522
5. Asylum - Intel Core i7 980X 4714 MHz - 3.697
6. Naekuh - Intel Q3QP Gulftown 4381 MHz - 4.058


Six Core without HTT

1. sno.lcn - Phenom II X6 1055T 5320 MHz - 4.797
2. rickss69 - Phenom II X6 1090T 5350 MHz - 4.836
3. crunchie - Phenom II X6 1090T 4408 MHz - 5.578
4. FlanK3r - Phenom II X6 1090T 4500 MHz - 5.609

http://forums.techpowerup.com/showthread.php?t=77369

 

[khon]

First of all WPrime is fully multithreaded, so yes the i7-970 will almost certainly beat a quadcore SB in that benchmark. It is not however an accurate reflection of most applications.

Secondly I have to point out that its a bit premature to talk about how well SB will overclock, as there is no way either of us can accurately predict that. The only thing we can know at this point is that it will be faster than a quadcore nehalem, which is already fairly close to a hexacore nehalem for most purposes (http://www.anandtech.com/bench/Product/142?vs=99)

 

[aigomorla]

Secondly I have to point out that its a bit premature to talk about how well SB will overclock, as there is no way either of us can accurately predict that.

:\

You need to spend more time on this forum.
Then you will realize who i exactly am.

We don't really know anything about how the socket 2011 will overclock, right? Or is that bad news as well?

Correct.. no one is telling me about 2011.
From what i heard, its a bigger mess over there.

 

[Edrick]

The "high-end" Sandy Bridge doesn't do anything special. The 3-channel(or 4 if you believe that arriving on PC) memory controller won't help most users and so will the extra cores.

Increased L3 caches? Nothing special either. There's diminishing returns beyond certain point and bigger caches might have resulted in higher latency.

Expecting more than 5% increase over the "crappy 115x" Sandy Bridge is ridiculous. If Intel is gonna deliver for Sandy Bridge, the very first versions will tell us how its doing.

I certainly consider the 40 lanes of PCIe (3.0) compared to the 20 lanes of PCIe (2.0) something special. Mostly for SLI users, but isnt that what the X chipsets are targeted for anyways? Granted this is not going to impact core performance at all. Just saying that there is more to CPU design than just raw performance numbers.