AMD CPU speculation... and expert conjecture

[de5_Roy]

I would like to point out, I was talking about i3s and Pentiums...Intel's bargain bin junk is just that...junk. The FX6300 is vastly superior to any Intel CPU within $20 of it one way or the other.

you actually made a blanket statement.
anyway, intel's lga11xx lineup is overpriced from top to bottom. intel doesn't change price points, only releases cpus with minor clockrate bump and mid-single to low double digit performance improvement every generation.
as for the fx6300, you oughta thank amd's own stagnation, market condition, and lastly - intel. if intel leaves out a price and performance gap, amd sics centurion cpus on that.

 

[jdwii]

Overpriced garbage, really? :lol: I don't spend my money on inferior products when I have the option to buy better. I don't need fluff like modular PSU, or and SSD, or even a fancy case. They are luxury items and nothing more. I care about performance for my dollar. None of those things will increase my FPS, at all. I don't desire a feature filled motherboard. All I care is if I am getting the performance I desire. I would be happier with a 1230v3 and an ASRock H81M-DGS R2.0, than I would an overpriced space heater of an FX 9xxx series, or even an FX 8350 with a 970a-ud3p. So would my power bill in the summer trying to keep the room cool. I don't hold allegiance to AMD or Intel. I go where the performance/$$, and longevity is. For me, that is an Intel CPU, with an AMD GPU. Granted I want to replace my HD 5850's with GTX 750ti's, but that is only because their power consumption is so low, while still be decently faster. If AMD responds with a similar card, I would be more than happy to go with one. Power consumption matters to me as well, at least for my file server. It has to stay Intel due to the Raid 5, for now, anyway. I have had my share of AMD rigs. I still own a couple now. But I really cannot recommend FX anymore, unless you happen to live near a Microcenter. Their combo deals are impossible to ignore.

I would like to point out, I was talking about i3s and Pentiums...Intel's bargain bin junk is just that...junk. The FX6300 is vastly superior to any Intel CPU within $20 of it one way or the other.

No not really when it comes to the Pentiums. The I3 i agree a 6350-6300 any day but the Pentium is a good choice and Intel boards are REAL cheap now and come with a lot. Lets take a look at watch dogs

Spoiler

Here the Pentium dual core is outperforming even my old 1100T and is even with a 110$ Kaveri Quad core.

Now lets look at other "Next gen" games
BF4
This guy on Youtube(the lowest measure of evidence) maxed this game out with his Pentium
I won't dare talk much about games that are made with only using 2 cores heavily which most still are for now and as shown on watch dogs 2 Intel cores equal 4 Amd ones in those cases but when a pentium doesn't equal a 100$ Amd quad core it equals a I5 when games use 2 cores at 1080P. For example

Spoiler

Spoiler

Spoiler

Spoiler

Spoiler

However its not always so black and gray

Spoiler

Its not easy to even recommend anything for cheap everything has very huge bottlenecks. I have to say a 6300 is what i would personally recommend since its fast enough for older games and is beating intel in newer ones or is even per dollar and of course its application performance is better.
Based on this evidence we can conclude that a Pentium equals a Quad core fx in newer games(most of the time) and a I5 in older ones using less cores. I would never say that is bad value and as stated before those same boards can support a I7 in the future if someone wants so upgrade.

 

[jdwii]

It is about equal in normal use, but the 6300 is cheaper allowing extra money to go towards gpu.

http://static.techspot.com/articles-info/827/bench/CPU_...

They cost the same plus Intel has better boards for less money?

 

[logainofhades]

Been debating on helping my sister out with a system upgrade for Sims 4. She has a core 2 E8190 with a GT 240 right now. As it sits now, it kinda struggles to play WoW @ 1280 x 1024. It pegs the CPU @ 100% usage quite easily. I am thinking a trip to microcenter is in order, in the near future. An FX 6300, GA-78lmt-usb3 paired with my old GTS 450 should make for a decent enough upgrade on the cheap for her.

 

[jdwii]

It is about equal in normal use, but the 6300 is cheaper allowing extra money to go towards gpu.

http://static.techspot.com/articles-info/827/bench/CPU_...

They cost the same plus Intel has better boards for less money?

I am seeing a 10 dollar minimum difference in price. to someone in this budget range it makes a difference.

In the U.S?
6300
http://www.amazon.com/AMD-FD6300WMHKBOX-FX-6300-Processor-Edition/dp/B009O7YORK
I3 haswell
http://www.amazon.com/Intel-i3-4130-FCLGA-Processor-BX80646I34130/dp/B00EUUKVXM

???????

Plus looking i have to spend 60$ on a AM3+ board instead of 45$ on a intel board just to get a 970 board. The Intel board i am seeing is a 51$ now instead of 45$
http://www.newegg.com/Product/Product.aspx?Item=N82E16813130731

This has 2 USB 3.0's, PCI-E and so on i used this board for the I5 build my friend got its a very good board for so cheap. Amd needs a complete new refresh on their higher-end boards. I think the main reason why Intel boards are getting cheaper on the low-end is over power management since Intel uses about half the power. Not to mention intel has more integrated on their CPU
Personally i wouldn't get anything lower then a 6300 for gaming even the quad core PD without L3 for 80$ wouldn't cut it i would rather have a Pentium then since older games will be as fast as a I5 and newer ones would be about equal to a 750K and you can get a Pentium for just 60$ vs 80$(750K) and 20$ on 400$ builds is a huge difference

 

[jdwii]

Been debating on helping my sister out with a system upgrade for Sims 4. She has a core 2 E8190 with a GT 240 right now. As it sits now, it kinda struggles to play WoW @ 1280 x 1024. It pegs the CPU @ 100% usage quite easily. I am thinking a trip to microcenter is in order, in the near future. An FX 6300, GA-78lmt-usb3 paired with my old GTS 450 should make for a decent enough upgrade on the cheap for her.

Yeah even Sims 3 uses 3 cores under my tests i'm guessing 4 will probably use the same sloppy engine and the graphics are meh and probably a 260X(750 TI) is more then enough. 8GB of ram and you are good. Sims 3 skips on even my PC which is lame
Also since Windows uses the 6 core fx as a 3 core with 6 threads its best to get that at least to allow for maximum performance for 3 threads.
One thing i LOVE from Amd is their Athlon 5350 build bet my butt i will be building one soon very soon and the total cost for everything is 250$ and that includes a 1TB hard drive, i'll be using this as my future HTPC builds. Amd great work.

 

[juanrga]

@juan: we will see who is right when the time comes. Though I do not know if AVX3 will be in or not.

@szatkus: JK says they will be better by skylake. My source says even if they fall short of predictions they will be "on par".

As for limited resources, they had less when JK designed K7/K8, and HTX...so...what?

You all are so caught up in the R&D budget you are forgetting the PEOPLE. These are not your average engineers. Every project leader at team green right now is the equivalent to a Michael Jordan caliber player at what they do. If the bulls had paid MJ less money, would he have been any less proficient at playing basketball? No. I understand your concerns, but these guys are BRILLIANT! You can spend less on R&D when you know what works to begin with...

Brilliant people as Keller can do better things than average people with more budget, but Keller cannot do miracles.

The comparison with K7/K8 doesn't hold. Even assuming that now AMD has the same budget than then (which is an unfounded assumption), the context is different: modern designs are one order of magnitude more complex, AMD relies heavily on third parties, and AMD is no longer competing with Intel alone. Moreover, K7/K8 were not designed from scratch, the K7 was heavily borrowed from a former DEC/alpha design, and the K8 was a minor modification to the K7. The same about Hypertransport; Meyer and Keller based it in Alpha's Lightning Data Transport. Now Keller doesn't have DEC advanced designs to copy... Finally, the K7 was released in an epoch when Intel did make several strategic and execution errors, which increased the success of AMD designs.

Keller return will almost guarantee AMD CPUs will be very competitive (at least I hope so), but the old times of AMD humiliating Intel in raw performance are gone. In fact, during last conferences AMD claimed

At AMD, we intended to be the leader in ARM CPUs

No similar comment for x86 was done.

I am pretty sure that AMD next arch. will not outperform Skylake.

 

[juanrga]

@juan: who said the dies were no longer monolithic? If that is your speculation, you are using a minnow bucket to hold the water.

That the new architecture will be highly modular instead monolithic is not my speculation.

 

[juanrga]

Skylake doesn't have that much to improve performance on. And, the big elephant in the room regarding Intel is that every generation since 32nm, they lose maximum overclocking potential. Before 32nm, they were pretty consistent at 4ghz. But, it's not really maximum overclocking potential, it's a chip built on a process that simply doesn't clock as high.

Meaning that if this trend continues, Intel will continue to lose clock speed, and eventually it could get bad enough for them to have to lower clockspeed on stock products on the high end.

People don't want to admit this. When Haswell and IB (both 22nm products) didn't clock so well, every Intel guy jumped up and blamed the TIM. A few delidded and had good results (but not all of them), so everyone just assumed it was only the TIM.

And then Intel released Devil's Canyon and everyone was disappointed.

Unless my memory is wrong, 32nm SB was good for 5ghz, 22nm IB was good for about 4.5ghz, and 22nm Haswell is good for about 4.5ghz.

Are my numbers about right? I want to do some math on them and make some guesses as to where 14nm clocks would end up but I want to make sure we're all in consensus here.

source for SB needed!
Haswell: http://www.overclock.net/t/1411077/haswell-overclocking... 4.53ghz average
IB: http://www.xbitlabs.com/articles/cpu/display/core-i7-37...

If Intel loses 10% maximum clockspeed potential with each die shrink, we are looking at 14nm Intel parts that don't make it past 4.05ghz on average. if AMD could keep clocks how they are now, they could end up with a 10% to 20% clock speed advantage over Intel.

I don't have high hopes for 14nm Intel as far as HEDT goes. I do have high hopes to see what Intel fanatics blame the poor clock scaling on this time. Perhaps we are going to go back to a time when Intel breaking 4ghz is a rare treat?

In the first place I expect Skylake will bring about 40--50% improvements in floating point. I also expect also big upgrade on the graphics side. No idea about integers.

Average OC frequencies achieved with air cooling from Lynnfield to Devils Canyon are

i7-880: 4.3GHz (45nm planar)
i7-2600k: 5.1GHz (32nm planar)
i7-3770k: 4.7GHz (22nm finfet)
i7-4770k: 4.5GHz (22nm finfet)
i7-4790k: 4.7GHz (22nm finfet)

Haswell 200MHz reduction is explained by the introduction of FIVR on die. However, DevilsCanyon get back at IB levels (I predicted DC would hit 4.8GHz). The reason why Ivy overclocks poor than Sandy was that the introduction of finfets modified the shape of the voltage curve as shown in next figure

Around ~1.32V the 22nm FF curve crosses the 32nm PLN curve. This is the point where Ivy, Haswell, and devil Canyon hit a voltage wall. Neither 14nm FF nor subsequent nodes have crossing point, which means that Broadwell-K will overclock better. If everything else goes the same, Broadwell-K would hit ~5.5Ghz on air.

22m FF was an intermediate process for the transition [planar --> 3D]. 14nm FF will introduce again classical scaling. Moreover Intel 14nm will use second gen FINFETs., whereas Globalfoundries/Samsung and TSMC will use first gen.

AMD will not have a clock advantage over Intel.

 

[juanrga]

Which is the whole problem with Faildozer. Clock for clock it was worse than K10. To be honest, piledriver wasn't exactly a big leap from K10 either. Its ability to hit higher clocks and more cores are what was able to make it push ahead. A 4.2ghz FX 4350 is less than 10% faster than a Phenom II X4 965 @ 4.0 in games, where as a 3.9ghz FX 6350 is nearly 20% faster. Most likely, the gap would be similar with an FX 6350, vs a Phenom II X6 1100t, at similar clock rates as they are for the FX 4350 vs the PhII X4 965.

Faildozer and its successors are AMD's P4. Pentium 3 was superior to Pentium 4 as well at similar clock speeds. Hell, even a 1.2ghz AMD Duron was faster than, I believe, a 1.6ghz P4. :lol: Intel learned from their mistake, and made core 2, which shared more with P3 and its mobile successors, than it did P4. It sounds like AMD is also learning from its mistakes and starting from scratch.

I honestly hope they pull off a winner of a chip like core 2 was. I do not hate AMD, but have been horribly disappointed in them lately. I want to see true competition again, like we had before Core 2 spanked K8. The K7/K8, PIII/P4 days were exciting competition wise. AMD got cocky and were caught with their pants down, when core 2 came out, and have never really recovered. Intel was just as arrogant before P4 got spanked by K8.

As I mentioned before, the whole team responsible for the Bulldozer fiasco was fired: the engineers, the CEO, the vice president of engineering... AMD did learn the lesson years ago.

In fact, the posterior AMD's Jaguar/Puma cores are not based in CMT modules and outperform Intel similar cores. It is Jaguar as well the core used in the consoles that saved AMD finances. Piledriver continues generating loses to the CPU division.

 

[vmN]

In the first place I expect Skylake will bring about 40--50% improvements in floating point. I also expect also big upgrade on the graphics side. No idea about integers.

Average OC frequencies achieved with air cooling from Lynnfield to Devils Canyon are

i7-880: 4.3GHz (45nm planar)
i7-2600k: 5.1GHz (32nm planar)
i7-3770k: 4.7GHz (22nm finfet)
i7-4770k: 4.5GHz (22nm finfet)
i7-4790k: 4.7GHz (22nm finfet)

Haswell 200MHz reduction is explained by the introduction of FIVR on die. However, DevilsCanyon get back at IB levels (I predicted DC would hit 4.8GHz). The reason why Ivy overclocks poor than Sandy was that the introduction of finfets modified the shape of the voltage curve as shown in next figure

Around ~1.32V the 22nm FF curve crosses the 32nm PLN curve. This is the point where Ivy, Haswell, and devil Canyon hit a voltage wall. Neither 14nm FF nor subsequent nodes have crossing point, which means that Broadwell-K will overclock better. If everything else goes the same, Broadwell-K would hit ~5.5Ghz on air.

22m FF was an intermediate process for the transition [planar --> 3D]. 14nm FF will introduce again classical scaling. Moreover Intel 14nm will use second gen FINFETs., whereas Globalfoundries/Samsung and TSMC will use first gen.

AMD will not have a clock advantage over Intel.

Skylake will support AVX512. Basically we can expect the same difference from AVX to AVX2.
Intel had big plans on their iGP, so we could suspect atleast 60% improvement.

I would suspect skylake to introduce another ALU/AGU per ALU cluster.

Basically broadwell to skylake will be like ivy to haswell. (In my predictions)

 

[colinp]

Juanrga, you mean to say the Bulldozer team were "fired," not "hired." Seen the same mistake several times now!

 

[colinp]

You're welcome.

 

[logainofhades]

Been debating on helping my sister out with a system upgrade for Sims 4. She has a core 2 E8190 with a GT 240 right now. As it sits now, it kinda struggles to play WoW @ 1280 x 1024. It pegs the CPU @ 100% usage quite easily. I am thinking a trip to microcenter is in order, in the near future. An FX 6300, GA-78lmt-usb3 paired with my old GTS 450 should make for a decent enough upgrade on the cheap for her.

Yeah even Sims 3 uses 3 cores under my tests i'm guessing 4 will probably use the same sloppy engine and the graphics are meh and probably a 260X(750 TI) is more then enough. 8GB of ram and you are good. Sims 3 skips on even my PC which is lame
Also since Windows uses the 6 core fx as a 3 core with 6 threads its best to get that at least to allow for maximum performance for 3 threads.
One thing i LOVE from Amd is their Athlon 5350 build bet my butt i will be building one soon very soon and the total cost for everything is 250$ and that includes a 1TB hard drive, i'll be using this as my future HTPC builds. Amd great work.

EA is claiming that Sims 4 will work on low end hardware far better than 3 did. Count me in on the skeptical group on that one. Funds will be a bit limited, so the GTS 450 is all I can do GPU wise, for now. I can probably give her one of my HD 5850's once I replace those with GTX 750ti's, unless AMD can answer with a faster card with similar power consumption. The HD 5850 is roughly between an HD 7770 and 7790 in performance.

 

[szatkus]

Juanrga, you mean to say the Bulldozer team were "fired," not "hired." Seen the same mistake several times now!

Now expect 200+ reposts by Juan that "hire" actually means "fire".

 

[8350rocks]

Jim Keller has had tons of time in the industry working on projects since K7...if he says it will be better, I have no reason to doubt. Though, like those I know at AMD, I consider it a win if they are within 5%.

 

[etayorius]

Jim Keller has had tons of time in the industry working on projects since K7...if he says it will be better, I have no reason to doubt. Though, like those I know at AMD, I consider it a win if they are within 5%.

Current Intel arch is about 50% faster than current AMD offers in regards to IPC, if next x86 AMD arch gains 50% it will put them right where Intel is sitting NOW, and i bet Intel will also bring a good bunch of extra performance on Skylake, therefor i do not expect AMD to even match Intel with their new Arch, but 50% gains in IPC is quite good none the less.

 

[logainofhades]

I have to agree that the GPU division has been carrying much of the weight financially for AMD lately. The console wins helped big time as well. Desktop sales are down, and have been for awhile. Even laptops are probably not doing as well as tablets are currently. I remember AMD was wanting to get more into the mobile sector, with regards to tablets, but I haven't kept up with them enough to see where they are with that ambition.

 

[szatkus]

Jim Keller has had tons of time in the industry working on projects since K7...if he says it will be better, I have no reason to doubt. Though, like those I know at AMD, I consider it a win if they are within 5%.

Current Intel arch is about 50% faster than current AMD offers in regards to IPC, if next x86 AMD arch gains 50% it will put them right where Intel is sitting NOW, and i bet Intel will also bring a good bunch of extra performance on Skylake, therefor i do not expect AMD to even match Intel with their new Arch, but 50% gains in IPC is quite good none the less.

They have a lot of building blocks from K10, Bulldozers, Cats and now A-57. Hard to say what you can get using them in good design.

JK worked on A7 which is close to Sandy Bridge (but bigger, so overall it isn't that good).

 

[Rum]

Juan did you even watch the AMD Core Innovation Summit? Lisa Su and Jim Keller said that x86 is just as important as ARM or did you chose to tune that part out? Keller also said that the build process will be fast, he and his team are taking things from other builds/ arch that AMD perfected in the past and implement them in the new Arch. Your assumptions that the new x86 Arch won't outperform the K12 are asinine and unfounded.
One last thing I am sure that Keller has plenty of funds to build his new x86 toy, all your nonsense just sounds like sour grapes cuz it's not all about ARM!

 

[logainofhades]

Juan did you even watch the AMD Core Innovation Summit? Lisa Su and Jim Keller said that x86 is just as important as ARM or did you chose to tune that part out? Keller also said that the build process will be fast, he and his team are taking things from other builds/ arch that AMD perfected in the past and implement them in the new Arch. Your assumptions that the new x86 Arch won't outperform the K12 are asinine and unfounded.
One last thing I am sure that Keller has plenty of funds to build his new x86 toy, all your nonsense just sounds like sour grapes cuz it's not all about ARM!

Sounds like they are doing what Intel did when they scrapped crapburst and made Core 2. I really hope this new arch is a winner. Phenom II was a very solid chip, and still are pretty capable. More capable than faildozer, and not really all that far behind piledriver at similar clock speeds and core counts. Had they been able to take the memory controller improvements of faildozer/piledriver, and tweaked PhII in a way that it could handle the higher clock speeds, and core counts, I think we would have had a far better chip than what we were given.

 

[8350rocks]

I would like to point one thing out...as at the time we did not know...

When Jim Keller said, "AMD are on track to catch Intel in high performance cores"

We expected that would be excavator likely...knowing what I know now...that would be the next uarch.

As for 50% better IPC with Intel, no...they have roughly 10-15% better IPC with better hardware prediction. I put this forth...

AMD is on the rise...we will see what comes...

 

[szatkus]

I would like to point one thing out...as at the time we did not know...

When Jim Keller said, "AMD are on track to catch Intel in high performance cores"

We expected that would be excavator likely...knowing what I know now...that would be the next uarch.

As for 50% better IPC with Intel, no...they have roughly 10-15% better IPC with better hardware prediction. I put this forth...

AMD is on the rise...we will see what comes...

But branch prediction is one of factors which influence IPC. Who cares how good the rest is if you can't utilize it?

So every software that relies on prediction (i.e. almost everything) is AMD-biased?

 

[colinp]

Luckily, we now have 2 roughly comparable chips in terms of core count and frequency: The 8350 and 4790k.

8350rocks: are you saying that the 8350 is within 10 to 15% of the 4790k on average in cpu tests (I.e. excluding gpu limited tests)? If so, then I'll be asking myself why I've waited so long to upgrade the amd way instead of being tempted to go intel for the first time in 18 years.

 

[de5_Roy]

But branch prediction is one of factors which influence IPC. Who cares how good the rest is if you can't utilize it?

So every software that relies on prediction (i.e. almost everything) is AMD-biased?

he can't be called wrong. as with all p.r., if he turns out to be right, he'll gloat. but if he turns out to be wrong, he'll claim "but...the data we had at that time showed that." he's kept enough gap for face-saving. as of now, take everything he says about future stuff with a steam-excavated pile of NaCl (notice the puns!! the puns!! i worked hard on those!!). it's better than nothing after amd's gone back to silent mode after the 7800 apus announcements. no, the 9590 trickery doesn't count.
i'm looking forward to the tonga gpus.
@8350rocks: do you have anything on the new tonga gpus? i thought those were gonna be called r9 275x. now i read about r9 280/280x... is amd messing with the model numbering again? :S