AMD CPU speculation... and expert conjecture

[Cataclysm_ZA]

And here I thought hasbeen was all about lowering power usage ...

Also looking at AT's GT3e review ... it gets stomped by Intel's proposed "650M performance" claims. only looks ok when its Vsyinc limited.

To be fair, Haswell does what it set out to do - improve power consumption at a set performance level. I saw that in most tests with a i7-3770K and the i7-4770K at the same frequencies, the Haswell chip is faster thanks to IPC improvements and improvements to Turbo Boost. I think that the chips would perform similarly with Ivy at 3.5GHz and Haswell at 3.2GHz.

And additionally, Intel never said that Iris had better performance than the GT650M - they just said that was their benchmark. Iris/Iris Pro is good, but its not $600+ good.

 

[noob2222]

@richland: just for lulz, i took a peek at tpu's haswell gt2 testing and noticed something funny (but not funny for the r.a.l.f.):
http://www.techpowerup.com/reviews/Intel/Core_i7_4770K_Haswell_GPU/
gt2 seems to meet and sometimes beat richland's 8670D (i.e. not the mobile sku). also check out the test settings' igpu(amd) driver version. apparently, gt2 played(!) crysis 3(!!!) and borderlands 2(!!) better than richland.

I like how richland is ~0.4fps faster than trinity

The TPU article isn't bad. Haswell manages to win 1 game, tie 2 (4.8 to 5.1pfs are both fails on c3), and lose 14. Total loss - 30% roughly
http://www.techpowerup.com/reviews/Intel/Core_i7_4770K_Haswell_GPU/20.html

Aside from that, not sure where they got their richland numbers. http://lensfire.in/25501/news/amd-richland-apu-a10-6800k-benchmark-review/
has richland 11% faster than 5800k on 3dmark, TPU has 2% over 5800k. Looks like tpu attempted richland #s by overclocking trinity to richland's stock numbers.

 

[noob2222]

And here I thought hasbeen was all about lowering power usage ...

Also looking at AT's GT3e review ... it gets stomped by Intel's proposed "650M performance" claims. only looks ok when its Vsyinc limited.

To be fair, Haswell does what it set out to do - improve power consumption at a set performance level. I saw that in most tests with a i7-3770K and the i7-4770K at the same frequencies, the Haswell chip is faster thanks to IPC improvements and improvements to Turbo Boost. I think that the chips would perform similarly with Ivy at 3.5GHz and Haswell at 3.2GHz.

And additionally, Intel never said that Iris had better performance than the GT650M - they just said that was their benchmark. Iris/Iris Pro is good, but its not $600+ good.

http://www.techspot.com/news/51307-intel-haswell-integrated-graphics-are-on-par-with-geforce-gt-650m.html

on par with? hardly.

25% slower, sure.

 

[Cataclysm_ZA]

http://www.techspot.com/news/51307-intel-haswell-integrated-graphics-are-on-par-with-geforce-gt-650m.html

on par with? hardly.

25% slower, sure.

Read the TechSpot article again. Did Intel say anything about being on par? NO. TechSpot did, and that's the problem that propagated through to other news sites once that article went up. Not even Tech Report said "on par" they actually said:

Both systems are running DiRT 3 with all the details cranked at 1080p resolution. Scott had difficulty detecting differences in performance and image quality between the two in person, and they look very similar on video. I believe the Haswell system is the one on the left.

The fact that Haswell's IGP can run DiRT 3 at these settings is certainly impressive. So is keeping up with the GeForce GT 650M, which is a mid-range part with 384 ALUs, a 128-bit path to dedicated memory, and GPU clock speeds as high as 900MHz.

I never expected Intel to be on par with the GT650M, it would have been nearly impossible, given their current driver and hardware team assigned to GPUs, to match that chip. If they had, GT3e would only be on a chip costing north of $800 and it would have consumed more power and die space.

But the fact that they're on par with a desktop GT640M is a great achievement.

 

[noob2222]

Not many posted temps, but hasbeen runs hotter than Ivy.

Speaking of ambient cooling, you’re going to want a solid cooling setup for Haswell. They don’t run much hotter than Ivy Bridge, but you can expect a 5-10 °C increase due to the integrated voltage regulator.http://www.overclockers.com/intel-i7-4770k-haswell-cpu-review

also 6c hotter load here at stock and 14c hotter at idle? and a 96c 4.6ghz overclock: http://www.tweaktown.com/reviews/5522/intel-core-i7-4770k-haswell-4th-gen-cpu-and-z87-express-chipset-review/index10.html

 

[noob2222]

on another note, here comes the GPU's worst nightmare.

http://www.tomshardware.com/news/Asus-4K-Gaming-Monitor-PQ321,22851.html

good luck feeding 60-120 fps @ 3840x2160

I wonder what this is going to do to offloading to the gpu vs HSA programming.

 

[truegenius]

Anandtech, funny enough, actually shows how far Intel still has to go. With dual-channel DDR3-2133 on Trinity and dual-channel DDR3-2400 on Haswell, the HD7660D is still appreciably faster than the HD4600 and only Iris and Iris Pro graphics takes the win.

and intel's cpu get higher bandwidth vs trinity using same speed ram

intel and amd discussions running simultaneously in a thread and no flame war is happening its must be a dream

 

[Cataclysm_ZA]

and intel's cpu get higher bandwidth vs trinity using same speed ram

Indeed. It would be great if Richland brought in some improvements to the memory controller because that's one of AMD's biggest weak spots now. It was a boon for performance with the Athlon 64, but they should have kept on improving it.

intel and amd discussions running simultaneously in a thread and no flame war is happening its must be a dream

I don't get the flame wars at all. Maybe its buyer's guilt? I mean these are just computer components, for crying out loud. However anyone spends their money is not my business. If it was something more utilitarian, like a car, then I'd understand. Smart buying is encouraged, but I don't get the fanboyism.

 

[sarinaide]

Looking at the richland pricing R1700 or $149 for the 6800K, I dont think I will be chucking my 5800K this time round.

 

[anxiousinfusion]

If flame wars pop up, I just whip out my VIA Nano and preach to the world how superior it is to any other consumer processor. Flamers promptly bring their argument elsewhere.

 

[Cataclysm_ZA]

Looking at the richland pricing R1700 or $149 for the 6800K, I dont think I will be chucking my 5800K this time round.

R1700? Are you looking at a South African website?

 

[The Q6660 Inside]

Looking at the richland pricing R1700 or $149 for the 6800K, I dont think I will be chucking my 5800K this time round.

After buying the 6800K, get some LN2 and push it to DDR3-3200.

 

[hartacus]

Isnt this AMD Steamroller speculation? Why does every AMD thread end up with tons of intel trolls. I think intel must pay these people.

 

[juanrga]

@richland: just for lulz, i took a peek at tpu's haswell gt2 testing and noticed something funny (but not funny for the r.a.l.f.):
http://www.techpowerup.com/reviews/Intel/Core_i7_4770K_Haswell_GPU/
gt2 seems to meet and sometimes beat richland's 8670D (i.e. not the mobile sku). also check out the test settings' igpu(amd) driver version. apparently, gt2 played(!) crysis 3(!!!) and borderlands 2(!!) better than richland.

I like how richland is ~0.4fps faster than trinity

The TPU article isn't bad. Haswell manages to win 1 game, tie 2 (4.8 to 5.1pfs are both fails on c3), and lose 14. Total loss - 30% roughly
http://www.techpowerup.com/reviews/Intel/Core_i7_4770K_Haswell_GPU/20.html

Aside from that, not sure where they got their richland numbers. http://lensfire.in/25501/news/amd-richland-apu-a10-6800k-benchmark-review/
has richland 11% faster than 5800k on 3dmark, TPU has 2% over 5800k. Looks like tpu attempted richland #s by overclocking trinity to richland's stock numbers.

A very weird 'review'.

 

[8350rocks]

Isnt this AMD Steamroller speculation? Why does every AMD thread end up with tons of intel trolls. I think intel must pay these people.

LOL...I was just thinking that earlier...

 

[hcl123]

I thought Haswell was the Trinity killer, if anything a A10 5800k is looking better than it ever did. Iris better be amazing because Richland is on its way.

At least when Bulldozer sucks i admit it these Intel fanboys won't admit that Haswell was one of the biggest disappoints in years at that company, then they justify it by saying oh its more efficient, funny seems like to me Amd is improving performance with lowering power consumption.

What are you saying! ?...

Intel never fails, even when it does lol ... and this is not only rhetoric, Failwell has hardware lock elision and some other minimal Hardware Transactional Memory features, and even if this features takes *YEARS* to be here in common software specially on the client/desktop side, you can bet that many benchmarks even of apps (which are synthetic derivatives more than anything else) will present them, and Haswell in very short time can growth to double of performance... but telling the general audience that benchmarks have HLE and HTM, that *IF* SR doesn't have it (which have no reasons to have since there is NONE software out there that has this outside of very specialized solutions) will only run well on Intel hardware... informing about this is going to slip the mind of general reviewers even if they are alerted to the fact(always happened since 3Dnow vs SSE2 and SSE2 vs SSE4 )

So DON'T say Failwell... at least AT that some here seems to take issues, has the right tools to make ppl eat their words, but if then its NOT representative (for not saying FRAUD) or meaningful of any software and systems out there ( already is, only representative of a particular bench soft), that isn't going to make a difference or stop nothing, specially when the gran public is none the wiser (never did).

Bulldozer(wrong designation for the first Orochi) sucks so to speack but for current "common" desktop jobs (for future if soft becomes MT is going to grow in perf). For servers was very good for the kind of entry level systems, specially the MCM socket (G34) deals where performance was(is) quite a good boost compared with previous, that not only were considerable cheaper in the hardware(below the average perf/$), but also meant a cut in software licensing for many deployments compared to intel(1 socket for ~the same perf of number of cores of 2 sockets, but at half the licenses cost).

Yet it never catched up, the contrary AMD server is almost gone. Making out of this server die a desktop chip variant was another mistake, usually intel used to do the inverse... but amd doesn't have the resources to support 2 design implementations based on the same uarch for shrinking markets(which if desktop oriented, could had been much smaller, much less power hungry, clock much higher... and cheaper... than the FX zambezi was).

So all comparative assumptions are always *RELATIVE* (there isn't this "sith" absolute dealing lol like this "sucks" parroted forever, as if is everywhere, every solution and no solution or improvements on anything is possible), and tremendous short lived, and serves any side including the fact that after Orochi/Zambezi, intel server solutions also improved, WHICH DOESN'T MAKE UNTRUTH THAT FOR *A SHORT WHILE* amd server solutions was quite better than intel.

Also when intel subsidizes its sales, and the propaganda is tremendous, its obvious everything "feels" from number of dealers to support to good builds.

Nevertheless about the "bulldozer sucks" it was visible because clocks where not even close to possible targets, and because "multithreading" never exploded as amd seems to have been counting in the "client market", actually it has been a ferocious campaign against this -multithreading- feature/factor, which is the only one factor that can mean *high performance* in current trends, single thread is NOT about high performance, its an obsolete metric...

Just know that Piledriver , Steamroller to Excavator ARE BULLDOZER... in truth they are the same *uarch*, AFAIK bulldozer was and is the designation of a uarch (including CMT, vertical MT etc), not of any particular implementation... the rest is propaganda, that in this case felt very short of the intents i suspect, which was about to "bad mouth" and put a negative connotation with all those possible lines, but ppl only parrot it about the first Orochi implementation.

I think that SR *IF* proved to be 30% ops per clock better is enough to prove the point, but then *if* the "excavator" first reveal is truth (The Beast http://www.tomshardware.co.uk/forum/352312-28-steamroller-speculation-expert-conjecture/page-79#10888698 ) it clearly proves that Bulldozer is light years away from any intel design and is the best and most scalable uarch ever design for x86.( if ever it does shows like this much on benches is another issue altogether. P4 sold much more than athlon 64, though this last one was quite a lot ahead to).

 

[palladin9479]

If flame wars pop up, I just whip out my VIA Nano and preach to the world how superior it is to any other consumer processor. Flamers promptly bring their argument elsewhere.

Pfft
http://www.viaembedded.com/en/products/boards/1893/1/EPIA-M920.html

Via Nano QuadCore (they got Arm offerings too)

You gotta give them credit, they've pretty much cornered the market on x86 for industrial / integrated devices. They saw the war between AMD and Intel then said "f*ck that" and have completely ignored the desktop / mobile segments.

 

[hcl123]

I just want to say: Why is this such a huge thread?

Simple answer: Ranting about off topic things.

I'll sum up for everyone what they are complaining about:

In a performance aspect with prices out of the question, Nvidia and Intel are dominating the market.

With Prices in hand, they are about hand in hand, as wholes.

In conclusion, if you are on a tighter budget or want to use OpenCL, Bit Mine, or Video Edit, then hit up the AMD market.

If you have the money and you want kick-ass gaming for the performance, go for Intel and/or Nvidia.

Enough Said.

UPDATE: By the way, I encountered a problem with an AMD fan on a GAMING topic and he brought up Operating Systems and how AMD works better with these, his main point was Linux. Keep in mind, most of the World Population isn't a hardware geek and/or an intense gamer, so really the focus will be mostly Windows. Also that Linux is terrible with games.

^ Probably doesn't know that the Radeon 8000HD series hasn't came out yet to defeat his beloved titan. But hey if most people use it, it must be better yet just like Windows and Capitalization,

Also check your Tock haswell yet? Probably justifies it in some way.

Also i see nothing wrong with going off topic when their is no info it educates people and does not hurt anything.

In a pure performance POV, comparing card to card doesn't the 7990 already does that ?

Its different ? ... only on your mind (lol)... Matter of fact(edit for typo) the *smaller* dies makes a lot of sense, i always like it. What i don't like is why use PCIe, so having to be 2 sockets for those top SKUs, instead of a MCM solution on Hypertransport like the server solutions MgC/Interlagos/Abu Dabhi.

Meaning in the future AMD could pass with only 1 or 2 dies for GPGPUs and the rest could be Triple, Quadruple or more MCM (multi chip module) of those single "smaller" dies. At least "a priori" it can mean a tremendous boost in yields, making those cards more cheaper than otherwise... also it would go in favor of 2.5D interposed DRAM that is the future trend.

Also i think Nvidia went in a "regression" way with Kepler, the Fermi design was quite better, and could had been proved to be quite more scalable... much included here the double pumping of shaders... comparing to previous a 1024sp Fermi style could beat hands down Titan (the big keller), and be substantially smaller... only everything like a "processor" might not be the best for Intel platforms, but $1.5B (billion with a B) that Nv received from Intel might had an almost magic influence on design.

The major problem of Nv for Fermi like, like for intel for P4, was and is FAB process, TSMC is not much about clock, ~2Ghz shaders might mean terrible power... but i just imagine something like a Tahiti with its 2048 shaders at 2Ghz made at other Fab than not TSMC... it could be a relative quite smaller die at something around the 20nm node, but with stratospheric performance...

 

[Cazalan]

Says FM2+ and will feature GDDR5/DDR3 controllers, dual channel controllers and HUMA, this is more or less what is expected.

And now you see why AMD wanted to delay Kaveri news as long as possible. Socket changes always put people off. Just look how many down talk Haswell because of the socket change.

 

[hcl123]

If flame wars pop up, I just whip out my VIA Nano and preach to the world how superior it is to any other consumer processor. Flamers promptly bring their argument elsewhere.

Pfft
http://www.viaembedded.com/en/products/boards/1893/1/EPIA-M920.html

Via Nano QuadCore (they got Arm offerings too)

You gotta give them credit, they've pretty much cornered the market on x86 for industrial / integrated devices. They saw the war between AMD and Intel then said "f*ck that" and have completely ignored the desktop / mobile segments.

C'mon VIA Nano (old Cyrix) is a niche player on x86 that has no significant presence in any market wahtsoever.

the 4 core took i thing, 5 or 6 years to come out after the double core( when others might go 16 cores / chip). If AMD did that it had been dead by now. AMD problem is PR and marketing, and i pick your post because its very illustrative, everything serves to "bash", to speculate reasons of failure, what it can have been done better... most of the things written have some truth, but all companies do a lot of mistakes that only are so, when only visible, and to prevent that "visibility", massive whitewash and propaganda campaigns ensue, which amd never was able to do even when it was in much better shape...

At the end of the day amd problem is not technical but commercial, and you are right, if AMD could had been like VIA, going unnoticed nobody bashes it nobody bias reviews it, then probably it would be better, only it would sell less...

Perhaps downsizing and go ARM is not a bad idea for amd, goes directly according to that, go unnoticed for a while... just wonder then what Intel to sell will commend reviews of its x86 against what !?.. ARM ? ..which one ? ... and a bench of ARM to x86 is not comparable in any case... not really, complete different uarch...

Perhaps then VIA Nano (cyrix) would have to be financed heavily, even by Intel (hey! they already did that to Nvidia) and promoted for a while, even technically advised and helped, so that intel could have a partner in the short future to continue this "FREAK SHOW" that goes on today... the show just sells a lot and gives them a lot of promotion by negative endorsement, how stupid it might sound it seems to work...

 

[hcl123]

Says FM2+ and will feature GDDR5/DDR3 controllers, dual channel controllers and HUMA, this is more or less what is expected.

And now you see why AMD wanted to delay Kaveri news as long as possible. Socket changes always put people off. Just look how many down talk Haswell because of the socket change.

Its inevitable and perfectly justifiable in this case with any player even future ARM IDMs. Putting eDRAM/stackedDRAM in/on the socket, has complete different electric requirements.

 

[The Q6660 Inside]

I thought Haswell was the Trinity killer, if anything a A10 5800k is looking better than it ever did. Iris better be amazing because Richland is on its way.

At least when Bulldozer sucks i admit it these Intel fanboys won't admit that Haswell was one of the biggest disappoints in years at that company, then they justify it by saying oh its more efficient, funny seems like to me Amd is improving performance with lowering power consumption.

What are you saying! ?...

Intel never fails, even when it does lol ... and this is not only rhetoric, Failwell has hardware lock elision and some other minimal Hardware Transactional Memory features, and even if this features takes *YEARS* to be here in common software specially on the client/desktop side, you can bet that many benchmarks even of apps (which are synthetic derivatives more than anything else) will present them, and Haswell in very short time can growth to double of performance... but telling the general audience that benchmarks have HLE and HTM, that *IF* SR doesn't have it (which have no reasons to have since there is NONE software out there that has this outside of very specialized solutions) will only run well on Intel hardware... informing about this is going to slip the mind of general reviewers even if they are alerted to the fact(always happened since 3Dnow vs SSE2 and SSE2 vs SSE4 )

So DON'T say Failwell... at least AT that some here seems to take issues, has the right tools to make ppl eat their words, but if then its NOT representative (for not saying FRAUD) or meaningful of any software and systems out there ( already is, only representative of a particular bench soft), that isn't going to make a difference or stop nothing, specially when the gran public is none the wiser (never did).

Bulldozer(wrong designation for the first Orochi) sucks so to speack but for current "common" desktop jobs (for future if soft becomes MT is going to grow in perf). For servers was very good for the kind of entry level systems, specially the MCM socket (G34) deals where performance was(is) quite a good boost compared with previous, that not only were considerable cheaper in the hardware(below the average perf/$), but also meant a cut in software licensing for many deployments compared to intel(1 socket for ~the same perf of number of cores of 2 sockets, but at half the licenses cost).

Yet it never catched up, the contrary AMD server is almost gone. Making out of this server die a desktop chip variant was another mistake, usually intel used to do the inverse... but amd doesn't have the resources to support 2 design implementations based on the same uarch for shrinking markets(which if desktop oriented, could had been much smaller, much less power hungry, clock much higher... and cheaper... than the FX zambezi was).

So all comparative assumptions are always *RELATIVE* (there isn't this "sith" absolute dealing lol like this "sucks" parroted forever, as if is everywhere, every solution and no solution or improvements on anything is possible), and tremendous short lived, and serves any side including the fact that after Orochi/Zambezi, intel server solutions also improved, WHICH DOESN'T MAKE UNTRUTH THAT FOR *A SHORT WHILE* amd server solutions was quite better than intel.

Also when intel subsidizes its sales, and the propaganda is tremendous, its obvious everything "feels" from number of dealers to support to good builds.

Nevertheless about the "bulldozer sucks" it was visible because clocks where not even close to possible targets, and because "multithreading" never exploded as amd seems to have been counting in the "client market", actually it has been a ferocious campaign against this -multithreading- feature/factor, which is the only one factor that can mean *high performance* in current trends, single thread is NOT about high performance, its an obsolete metric...

Just know that Piledriver , Steamroller to Excavator ARE BULLDOZER... in truth they are the same *uarch*, AFAIK bulldozer was and is the designation of a uarch (including CMT, vertical MT etc), not of any particular implementation... the rest is propaganda, that in this case felt very short of the intents i suspect, which was about to "bad mouth" and put a negative connotation with all those possible lines, but ppl only parrot it about the first Orochi implementation.

I think that SR *IF* proved to be 30% ops per clock better is enough to prove the point, but then *if* the "excavator" first reveal is truth (The Beast http://www.tomshardware.co.uk/forum/352312-28-steamroller-speculation-expert-conjecture/page-79#10888698 ) it clearly proves that Bulldozer is light years away from any intel design and is the best and most scalable uarch ever design for x86.( if ever it does shows like this much on benches is another issue altogether. P4 sold much more than Athlon 64, though this last one was quite a lot ahead to).

Even the 4770K is deep down inside an ancient P6 Pentium Pro, even the 1100T is an K7/K8 Athlon/Athlon 64. Bulldozer was an attempt to break free from the old Athlon uarch, of which it has mostly replaced in its products.

However, only time will tell if it was the right choice...

 

[palladin9479]

Spoiler

If flame wars pop up, I just whip out my VIA Nano and preach to the world how superior it is to any other consumer processor. Flamers promptly bring their argument elsewhere.

Pfft
http://www.viaembedded.com/en/products/boards/1893/1/EPIA-M920.html

Via Nano QuadCore (they got Arm offerings too)

You gotta give them credit, they've pretty much cornered the market on x86 for industrial / integrated devices. They saw the war between AMD and Intel then said "f*ck that" and have completely ignored the desktop / mobile segments.

C'mon VIA Nano (old Cyrix) is a niche player on x86 that has no significant presence in any market wahtsoever.

the 4 core took i thing, 5 or 6 years to come out after the double core( when others might go 16 cores / chip). If AMD did that it had been dead by now. AMD problem is PR and marketing, and i pick your post because its very illustrative, everything serves to "bash", to speculate reasons of failure, what it can have been done better... most of the things written have some truth, but all companies do a lot of mistakes that only are so, when only visible, and to prevent that "visibility", massive whitewash and propaganda campaigns ensue, which amd never was able to do even when it was in much better shape...

At the end of the day amd problem is not technical but commercial, and you are right, if AMD could had been like VIA, going unnoticed nobody bashes it nobody bias reviews it, then probably it would be better, only it would sell less...

Perhaps downsizing and go ARM is not a bad idea for amd, goes directly according to that, go unnoticed for a while... just wonder then what Intel to sell will commend reviews of its x86 against what !?.. ARM ? ..which one ? ... and a bench of ARM to x86 is not comparable in any case... not really, complete different uarch...

Perhaps then VIA Nano (cyrix) would have to be financed heavily, even by Intel (hey! they already did that to Nvidia) and promoted for a while, even technically advised and helped, so that intel could have a partner in the short future to continue this "FREAK SHOW" that goes on today... the show just sells a lot and gives them a lot of promotion by negative endorsement, how stupid it might sound it seems to work...

Ahh you might not be versed in what those guys do. Via does lots of business in the integrated / industrial segments. Think medical machines, KIOSKs, digital ad bulletin boards, ATMs, cash registers, airport flight boards, robotic factory control systems, sensor boards and home automation (if your into that).

If you actually took the time to look you'd realize their boards are horrifically overpriced, if your goal was to run Crysis 3 or do iTunes. Otherwise they tend to come equipped with 3~4 serial ports that run at standard RS-232 speeds as well as RS-422/423. They also tend to have either 1~2 GPIO headers along with another 1~2 LVDS headers. Basically what their lacking in raw CPU power they more then make up for in IO capability and thermal tolerance. So you wouldn't ever buy one of these for your home PC, their graphics units really sucks. They make semi-decent low-power home servers though AMD finally caught up with them (Atom is still far behind). Their padlock technology is amazing, fastest AES encryption and SHA digest processing you can get though it's rather niche and requires it to be supported at the ASM level. Intel's AES-NI was a blatant copy of padlock btw.

Anyhow if you take the time to look around you'd notice many of the integrated systems surrounding you have a Via processor inside. Look for things that you would normally ignore like ATM's, billboards or the cash register (running Windows CE) at your local store.

 

[JAYDEEJOHN]

2 things
If youre the only game in town, you can charge a pretty penny.
Low volume/higher costs.

Its the markets they settled in, and has kept them going

 

[hcl123]

Spoiler

I thought Haswell was the Trinity killer, if anything a A10 5800k is looking better than it ever did. Iris better be amazing because Richland is on its way.

At least when Bulldozer sucks i admit it these Intel fanboys won't admit that Haswell was one of the biggest disappoints in years at that company, then they justify it by saying oh its more efficient, funny seems like to me Amd is improving performance with lowering power consumption.

What are you saying! ?...

Intel never fails, even when it does lol ... and this is not only rhetoric, Failwell has hardware lock elision and some other minimal Hardware Transactional Memory features, and even if this features takes *YEARS* to be here in common software specially on the client/desktop side, you can bet that many benchmarks even of apps (which are synthetic derivatives more than anything else) will present them, and Haswell in very short time can growth to double of performance... but telling the general audience that benchmarks have HLE and HTM, that *IF* SR doesn't have it (which have no reasons to have since there is NONE software out there that has this outside of very specialized solutions) will only run well on Intel hardware... informing about this is going to slip the mind of general reviewers even if they are alerted to the fact(always happened since 3Dnow vs SSE2 and SSE2 vs SSE4 )

So DON'T say Failwell... at least AT that some here seems to take issues, has the right tools to make ppl eat their words, but if then its NOT representative (for not saying FRAUD) or meaningful of any software and systems out there ( already is, only representative of a particular bench soft), that isn't going to make a difference or stop nothing, specially when the gran public is none the wiser (never did).

Bulldozer(wrong designation for the first Orochi) sucks so to speack but for current "common" desktop jobs (for future if soft becomes MT is going to grow in perf). For servers was very good for the kind of entry level systems, specially the MCM socket (G34) deals where performance was(is) quite a good boost compared with previous, that not only were considerable cheaper in the hardware(below the average perf/$), but also meant a cut in software licensing for many deployments compared to intel(1 socket for ~the same perf of number of cores of 2 sockets, but at half the licenses cost).

Yet it never catched up, the contrary AMD server is almost gone. Making out of this server die a desktop chip variant was another mistake, usually intel used to do the inverse... but amd doesn't have the resources to support 2 design implementations based on the same uarch for shrinking markets(which if desktop oriented, could had been much smaller, much less power hungry, clock much higher... and cheaper... than the FX zambezi was).

So all comparative assumptions are always *RELATIVE* (there isn't this "sith" absolute dealing lol like this "sucks" parroted forever, as if is everywhere, every solution and no solution or improvements on anything is possible), and tremendous short lived, and serves any side including the fact that after Orochi/Zambezi, intel server solutions also improved, WHICH DOESN'T MAKE UNTRUTH THAT FOR *A SHORT WHILE* amd server solutions was quite better than intel.

Also when intel subsidizes its sales, and the propaganda is tremendous, its obvious everything "feels" from number of dealers to support to good builds.

Nevertheless about the "bulldozer sucks" it was visible because clocks where not even close to possible targets, and because "multithreading" never exploded as amd seems to have been counting in the "client market", actually it has been a ferocious campaign against this -multithreading- feature/factor, which is the only one factor that can mean *high performance* in current trends, single thread is NOT about high performance, its an obsolete metric...

Just know that Piledriver , Steamroller to Excavator ARE BULLDOZER... in truth they are the same *uarch*, AFAIK bulldozer was and is the designation of a uarch (including CMT, vertical MT etc), not of any particular implementation... the rest is propaganda, that in this case felt very short of the intents i suspect, which was about to "bad mouth" and put a negative connotation with all those possible lines, but ppl only parrot it about the first Orochi implementation.

I think that SR *IF* proved to be 30% ops per clock better is enough to prove the point, but then *if* the "excavator" first reveal is truth (The Beast http://www.tomshardware.co.uk/forum/352312-28-steamroller-speculation-expert-conjecture/page-79#10888698 ) it clearly proves that Bulldozer is light years away from any intel design and is the best and most scalable uarch ever design for x86.( if ever it does shows like this much on benches is another issue altogether. P4 sold much more than Athlon 64, though this last one was quite a lot ahead to).

Even the 4770K is deep down inside an ancient P6 Pentium Pro, even the 1100T is an K7/K8 Athlon/Athlon 64. Bulldozer was an attempt to break free from the old Athlon uarch, of which it has mostly replaced in its products.

However, only time will tell if it was the right choice...

Well if for one thing, AMD most probably have found a way to have the most powerful "commercial" x86 CPU on the market with Piledriver http://www.sweclockers.com/nyhet/17101-amd-forbereder-fx-9000-pa-upp-till-50-ghz

Lets see if intel responds, which seems it cannot with at least with Haswell, because more than IB, HB is a truly Hotwell.
http://imagescdn.tweaktown.com/content/5/5/5522_43_intel_core_i7_4770k_haswell_4th_gen_cpu_and_z87_express_chipset_review.png

All comparatives are *RELATIVE*, but if the TOP performance at a commercial speed is of any value, then AMD might got that position comparing oranges to oranges, that is, or number of threads with number of threads... which shows that BD uarch is very scalable indeed.

Only i don't expect any reviews of this, only bashing about power... showing in a top position on a chart, a product from "others" is something that can break the status quo of "sponsoring". They never minded of comparing a 12thread i7 with FX, for general reviewers is fair, this might not be LOL

Falls back to a past posts discussion...

http://www.techspot.com/news/51307-intel-haswell-integrated-graphics-are-on-par-with-geforce-gt-650m.html

on par with? hardly.

25% slower, sure.

Read the TechSpot article again. Did Intel say anything about being on par? NO. TechSpot did, and that's the problem that propagated through to other news sites once that article went up. Not even Tech Report said "on par" they actually said:

Both systems are running DiRT 3 with all the details cranked at 1080p resolution. Scott had difficulty detecting differences in performance and image quality between the two in person, and they look very similar on video. I believe the Haswell system is the one on the left.

The fact that Haswell's IGP can run DiRT 3 at these settings is certainly impressive. So is keeping up with the GeForce GT 650M, which is a mid-range part with 384 ALUs, a 128-bit path to dedicated memory, and GPU clock speeds as high as 900MHz.

I never expected Intel to be on par with the GT650M, it would have been nearly impossible, given their current driver and hardware team assigned to GPUs, to match that chip. If they had, GT3e would only be on a chip costing north of $800 and it would have consumed more power and die space.

But the fact that they're on par with a desktop GT640M is a great achievement.

where there is a shred of evidence of any of those assertions ??

The Techspot article link, references in the article a Techreport article, but none of them has a review, only the first poster showed a chart of AT.

Without wanting to be rude, or entering in polemic, isn't there anyone else review that might be much more representative ? (AT has proved throughout time of not being prove of nothing... sorry...)

At least a past posts showed a review in TH, and there seems Trinity still wins in most games though it loses in a lot of synthetics, which nevertheless Trinity might be less than a discrete 640 or 50 GPU of Nvidia... never expect the exact truth, but sorry there might be something a little more representative, a clue not a "close" bias...