Quote:Ivy Bridge was only about a 4-5% increase at best from Sandy Bridge chips
Leaks have shown Haswell to be about 3-4% faster
Where did you read that ?
If it was true , AMD would laugh at Intel right now . Seriously , People won't upgrade their CPU for only being 5% faster
To be correct : 3.8 Haswell ~ 4.2 Ivy ~ 4.5 or 4.6 Sandy
Quote:Ivy Bridge was only about a 4-5% increase at best from Sandy Bridge chips
Yeah... No.
5% is around the minimum increase seen in some tasks and that sounds very, very conservative if you tried applying it as the maximum for all tasks (and 5% isn't the maximum for all tasks either). We've seen decent increases from Ivy Bridge on Dolphin at any rate. Haswell is expected to increase performance around 10%.
(02-27-2013, 12:29 AM)admin89 Wrote: [ -> ]Quote:Ivy Bridge was only about a 4-5% increase at best from Sandy Bridge chips
Leaks have shown Haswell to be about 3-4% faster
Where did you read that ?
If it was true , AMD would laugh at Intel right now . Seriously , People won't upgrade their CPU for only being 5% faster
To be correct : 3.8 Haswell ~ 4.2 Ivy ~ 4.5 or 4.6 Sandy
http://www.guru3d.com/news_story/haswell_versus_ivy_bridge_clock_for_clock_performance.html
http://www.engadget.com/2012/12/12/haswell-leak/
Obviously the first site could be false, but with the focus on the Intel 4600 (and with it a raise of 7W TDP), I really doubt much focus was given on improving the IPC. Add to that the lack of competition and what we saw from Ivy Bridge (lack of solder under the heat spreader, the very small IPC increase, greatest boon being the addition of PCI-E 3.0) and we'll most likely see a small upgrade from Intel. I guess it depends on the scale of improvement AVX2 will bring.
I honestly hope it's a 10-15% increase to be able to smoothly play Mario Galaxy 1/2 without a 4.7GHz+ overclock, but I can't help but feel that's unrealistic considering their focus on improving mobile GPUs.
Haswell isn't a tock for nothing you know...
Sarasvati Wrote:Ivy Bridge was only about a 4-5% increase at best from Sandy Bridge chips
Where did you get this from? The benchmarks on tomshardware, anandtech, guru3d, and every other benchmarking site that I frequent have all shown an 8-12% increase averaging around 9%.
Sarasvati Wrote:Haswell is looking to be almost exclusively an iGPU update with most of the increase in power going to that.
Sarasvati Wrote:I really doubt much focus was given on improving the IPC.
Oh good god I don't even know where to start on this. Literally every article I have read on haswell contradicts this so I don't know where you are getting these ideas from. Intel has gone into great details about the new changes to the pipeline and cache. You clearly have no idea what you're talking about. But hey, don't take my word for it:
http://www.anandtech.com/show/6355/intels-haswell-architecture/6
http://www.realworldtech.com/haswell-cpu/
Those are the haswell microarchitecture analysis articles from the first two sites to pop into my head. I particularly like the realworldtech article because as usual they go far more in-depth than the other sites.
Sarasvati Wrote:I honestly hope it's a 10-15% increase to be able to smoothly play Mario Galaxy 1/2 without a 4.7GHz+ overclock, but I can't help but feel that's unrealistic considering their focus on improving mobile GPUs.
There is little emphasis being placed on mobile IGPs. With the GT3 being the only IGP that is supposed to be significantly faster than the HD 4000. Seriously, where are you getting this stuff from?
Sarasvati Wrote:Leaks have shown Haswell to be about 3-4% faster, so I imagine it would be clocked at 4.0-4.1GHz to match a 4.4GHz Sandy Bridge.
Sarasvati Wrote:http://www.guru3d.com/news_story/haswell...mance.html
http://www.engadget.com/2012/12/12/haswell-leak/
So your "evidence" is an unverified x87 test? Forgive me if I remain unconvinced for obvious reasons. An unverified SSE2 or integer test would have at least offered some relevance if it were true.
Sarasvati Wrote:Obviously the first site could be false, but with the focus on the Intel 4600 (and with it a raise of 7W TDP)
The maximum TDP has indeed been slightly raised due to the wider pipeline. But the average TDP should be lower and the pipeline should be faster so there is really no need to be concerned.
Sarasvati Wrote:Add to that the lack of competition and what we saw from Ivy Bridge (lack of solder under the heat spreader, the very small IPC increase, greatest boon being the addition of PCI-E 3.0) and we'll most likely see a small upgrade from Intel.
Do you know nothing of the tick-tock cycle? Ivy bridge was a tick, haswell is a tock. The goal of ticks is to shrink the die and lower power consumption and manufacturing costs (which they did). We usually get small or no performance improvements during ticks. In fact the improvement from sandy bridge to ivy bridge was greater than the previous 3 ticks. Haswell on the other hand is a tock. Which is a drastic overhaul of the microarchitecture. These have historically always delivering significant increases in performance but do not lower the TDP or increase clock rate. This is consistent with the improvements that they are making in haswell.
Sarasvati Wrote:I guess it depends on the scale of improvement AVX2 will bring.
AVX2 has little to do with the improvements that are being made to the pipeline.
(02-27-2013, 06:51 AM)NaturalViolence Wrote: [ -> ]Oh good god I don't even know where to start on this. Literally every article I have read on haswell contradicts this so I don't know where you are getting these ideas from. Intel has gone into great details about the new changes to the pipeline and cache. You clearly have no idea what you're talking about. But hey, don't take my word for it:
http://www.anandtech.com/show/6355/intels-haswell-architecture/6
http://www.realworldtech.com/haswell-cpu/
Those are the haswell microarchitecture analysis articles from the first two sites to pop into my head. I particularly like the realworldtech article because as usual they go far more in-depth than the other sites.
Even in the realworldtech article they cite the improvement as "Haswell core will offer around 10% greater performance for existing software, compared to Sandy Bridge" for existing software and with newer AVX2 instructions (in theory) being able to make considerable improvements.
(02-27-2013, 06:51 AM)NaturalViolence Wrote: [ -> ]There is little emphasis being placed on mobile IGPs. With the GT3 being the only IGP that is supposed to be significantly faster than the HD 4000. Seriously, where are you getting this stuff from?
They definitely focused on reducing power and improving the GPU architecture over Ivy Bridge quite a bit. These are expected to have massive gains in laptops and tablets, so I really don't see how that wasn't a major part behind haswell. The iGPU gains are quite significant and shows they had a large focus on improving it.
(02-27-2013, 06:51 AM)NaturalViolence Wrote: [ -> ]Do you know nothing of the tick-tock cycle? Ivy bridge was a tick, haswell is a tock. The goal of ticks is to shrink the die and lower power consumption and manufacturing costs (which they did). We usually get small or no performance improvements during ticks. In fact the improvement from sandy bridge to ivy bridge was greater than the previous 3 ticks. Haswell on the other hand is a tock. Which is a drastic overhaul of the microarchitecture. These have historically always delivering significant increases in performance but do not lower the TDP or increase clock rate. This is consistent with the improvements that they are making in haswell.
I don't appreciate the condesending attitude nor the assumption I have no idea what I'm talking about. Yes I know exactly how tick-tock cycles work and I really don't need a lecture from such an overly critical person.
Unless you have real world experience of the chip you shouldn't treat your knowledge as fact either. From what I can see it doesn't look like it'll be a massive improvement some people think (15%+) even if I hope it would be for a decent upgrade from my 2500k.
(02-27-2013, 06:51 AM)NaturalViolence Wrote: [ -> ]AVX2 has little to do with the improvements that are being made to the pipeline.
Again your cited source even lists the greatest improvements being to software that can utilize that instruction set. When it isn't utilized it's a small increase in IPC. If dolphin were to use that instruction set I can imagine it being much better, but without it it's not that great an increase.
Next time act a bit less snide and assume less; you'll come off as a nicer person.
Edit: Sorry if I derailed a topic any further, I probably shouldn't have continued responding.
Sarasvati Wrote:Even in the realworldtech article they cite the improvement as "Haswell core will offer around 10% greater performance for existing software, compared to Sandy Bridge" for existing software and with newer AVX2 instructions (in theory) being able to make considerable improvements.
And? Both statements that you made that I quoted are still wrong.
Sarasvati Wrote:They definitely focused on reducing power and improving the GPU architecture over Ivy Bridge quite a bit. These are expected to have massive gains in laptops and tablets, so I really don't see how that wasn't a major part behind haswell. The iGPU gains are quite significant and shows they had a large focus on improving it.
They made minor improvements to the actual EU architecture. For most models the number of EUs is the same and the total expected performance is barely higher. The die shrink will cause a significant reduction in power consumption but I would hardly say that the emphasis of the architecture was on improving IGP performance or to go so far as you and say that it is "almost exclusively an iGPU update". That's just flat out wrong.
Sarasvati Wrote:I don't appreciate the condesending attitude nor the assumption I have no idea what I'm talking about.
The assumption that you have no idea what you're talking about comes from you posting things that are wrong. You made a number of sweeping statements about how haswell is just IGP improvements and AVX2 support. Posted statements about ivy/sandy bridge performance that can easily be confirmed as wrong. And to top if off when someone pointed this out your respond was an unverified x87 benchmark. So yeah, my opinion of you is not real high right now. I think you would agree that it's the logical conclusion to draw from the content that you have produced thus far.
Sarasvati Wrote:Yes I know exactly how tick-tock cycles work and I really don't need a lecture from such an overly critical person.
Then stop talking about it like it's a tick!
Sarasvati Wrote:Unless you have real world experience of the chip you shouldn't treat your knowledge as fact either. From what I can see it doesn't look like it'll be a massive improvement some people think (15%+) even if I hope it would be for a decent upgrade from my 2500k.
My knowledge and assumptions are based on the information available to us from Intel and from analysts. I make no claims that have not already been made by others who are far more reputable than me.
Sarasvati Wrote:Again your cited source even lists the greatest improvements being to software that can utilize that instruction set. When it isn't utilized it's a small increase in IPC.
This doesn't invalid my statement at all. AVX2 has little to do with the improvements being made to the pipeline. Average application performance will increase by at least 10% with the same number of cores. And that is a significant increase in IPC. Though I still have my doubts that it will only be 10% considering how much they are revamping the pipeline.
I don’t know if this is known already but Super Mario Galaxy
2 can be played full speed (most of the time) on Dolphin-win-x86-r7719 using
the ivy bridges turbo boost and idle skipping unchecked with LLE and LLE on thread used maybe
somehow this info could help the developers idk.
But at least now I know its playable
(02-27-2013, 06:50 PM)natnint4000 Wrote: [ -> ]full speed (most of the time) on Dolphin-win-x86-r7719
Old builds are usually faster but they're no more supported on the forum (and by devs I suppose
).
(02-27-2013, 06:50 PM)natnint4000 Wrote: [ -> ]with LLE and LLE on thread used
You're likely to get freezes in this game if you use this option though
(02-27-2013, 10:11 PM)LordVador Wrote: [ -> ] (02-27-2013, 06:50 PM)natnint4000 Wrote: [ -> ]full speed (most of the time) on Dolphin-win-x86-r7719
Old builds are usually faster but they're no more supported on the forum (and by devs I suppose ).
(02-27-2013, 06:50 PM)natnint4000 Wrote: [ -> ]with LLE and LLE on thread used
You're likely to get freezes in this game if you use this option though
Yeah thats right i guess i will have to wait before i can play Super Mario Galaxy 2 on the PC at least Sonic Colours works