AMD CPU speculation... and expert conjecture

[Socket Invalid]

I expect Zen core to have ~40% better IPC (integer) than Piledriver and twice bigger FPU. I think this is realistic according to the little info I have about the arch. However, certain AMD engineer did claim that the goal is "100% higher IPC than Piledriver". This is the same kind of claims made during Bulldozer.

If they can pull off a 40% IPC improvement, it would be a huge win. It close the gap with intel considerably. But I would be surprised if they hit 40 on the first generation.

40% more IPC compared to Piledriver sounds realistic, anything more is pure vapor and wishful thinking.

I think AMD is heading in the right direction with their A - series. Powerful processor with a magnificent graphics card all in one? Lovin' it. (I know it's off topic lel)
😉

 

[de5_Roy]

When he presented Vulkan to everyone he is talking as president of the Khronos Group, not as Nvidia executive. You said that downplay was coming from Nvidia, when in reality it is just Khronos Group saying the truth.

yes, the downplaying is coming from nvidia, not just from tevrett per se. however, his statement and affiliation (and prior actions of said affiliated vendor) present a strong connection that i cannot ignore.

What he said coincides with what other companies of the Group were saying the presentation day. This is from Imagination:

Spoiler

Imagination is proud to have been a major contributor to the new Vulkan standard. With this new API alongside OpenGL ES, we feel that Khronos now has the right set of APIs in place to address the extreme performance and mainstream ends of the graphics programming market,” said Peter McGuinness, director of technology marketing, Imagination Technologies.

The bold font is from mine.

you know what that color coding in your posts mean.

i wanted to ignore the rest of the post but did for completeness's sake.

Spoiler

Nvidia was originally interested

http://www.eteknix.com/nvidia-asked-amd-mantle-montreal-event/

Spoiler

then latter lose any interest because DX12 would render "Mantle obsolete and unnecessary"
which is true.
http://www.lazygamer.net/general-news/nvidia-no-benefit-to-using-amds-mantle/

from your own source:

Spoiler

even if it is filled with far too much Nvidia evangelism.

see, that's what i mean when i describe nvidia's actions incl. them downplaying mantle's contribution.

in the end, no matter how bitter you're with amd or how many quote you throw at people or other stuff, you cannot force your own conclusion by attempting to win in this kind of argument. that's how strongly amd's efforts have affected everyone concerned. you may deny it, that's your prerogative (i am attempting to terminate this argument here because the facts won't change).

 

[Cazalan]

see, that's what i mean when i describe nvidia's actions incl. them downplaying mantle's contribution.

in the end, no matter how bitter you're with amd or how many quote you throw at people or other stuff, you cannot force your own conclusion by attempting to win in this kind of argument. that's how strongly amd's efforts have affected everyone concerned. you may deny it, that's your prerogative (i am attempting to terminate this argument here because the facts won't change).

I think (hope) most people can see through the corporate politics. Mantle went as far as it could being an AMD creation. Once they passed on the core of it to Khronos and give it a new name, swap in/out some features, other companies can get behind it. Squabbling over the size or percentage of the donation is just petty. We all benefit from it in the long run if you use any nix systems.

 

[juanrga]

When he presented Vulkan to everyone he is talking as president of the Khronos Group, not as Nvidia executive. You said that downplay was coming from Nvidia, when in reality it is just Khronos Group saying the truth.

yes, the downplaying is coming from nvidia, not just from tevrett per se. however, his statement and affiliation (and prior actions of said affiliated vendor) present a strong connection that i cannot ignore.

You want reduce everything to a Nvidia vs AMD fight, but the words you quoted are coming from the president of the Khronos Group, are found in the news archived in the Kronos group website and are supported by other members of the group also quoted in the new release.

Call me when you have a real argument.

 

[de5_Roy]

AMD to introduce its Radeon R9 300-series lineup at Computex
http://www.kitguru.net/components/graphic-cards/anton-shilov/amd-to-introduce-its-radeon-r9-300-series-lineup-at-computex/
meanwhile, nvidia let maximum pc take pics of it's new titan x gfx card. pc gpu war is slowly heating up.
kitguru doesn't mention the recent claim that amd has excess stock of current gen cards. if amd doesn't clear the excess cards they risk in-competition with the new cards. how the heck did amd come up with an entirely new lineup. i've been expecting only one two new flagship and the rest being rebrands ever since the news broke.

I think (hope) most people can see through the corporate politics. Mantle went as far as it could being an AMD creation. Once they passed on the core of it to Khronos and give it a new name, swap in/out some features, other companies can get behind it. Squabbling over the size or percentage of the donation is just petty. We all benefit from it in the long run if you use any nix systems.

+1. this is corporate politics.
amusingly, the people singing mantle's death seem to be mostly tech journalists or i.s.v. people. trevett was the only one close to an authority figure who downplayed mantle's contribution and he incidentally happens to be an nvidia v.p. i could go full conspiracy theorist and twist the events into a subtle smear campaign against amd's achievements . i wonder if it's too far fetched with the new gfx card launches so near. 😗

 

[jdwii]

Spoiler

Dude steamroller isn't anywhere close to a 20% upgrade in IPC over piledriver not sure why people think this when the majority of benchmarks claim otherwise.

And if I had said steamroller, that would mean something. I didn't. I said Steamroller AND Excavator. And they should combine to hit a 15 to 20 percent IPC increase over Vishera.

A 4ghz, 95 watt, Excavator FX would perform at the level of a 5ghz+ Zambezi, while using a whole lot less power. Bulldozer may be a crap arch, but it has come a long way in the last 3 years. I think that's chip worth making.

I expect Zen core to have ~40% better IPC (integer) than Piledriver and twice bigger FPU. I think this is realistic according to the little info I have about the arch. However, certain AMD engineer did claim that the goal is "100% higher IPC than Piledriver". This is the same kind of claims made during Bulldozer.

If they can pull off a 40% IPC improvement, it would be a huge win. It close the gap with intel considerably. But I would be surprised if they hit 40 on the first generation.

40% more IPC compared to Piledriver sounds realistic, anything more is pure vapor and wishful thinking.

I think AMD is heading in the right direction with their A - series. Powerful processor with a magnificent graphics card all in one? Lovin' it. (I know it's off topic lel)
😉

Not off topic at all, APU's are still great if you want a computer that can game without a GPU and with a small itx board and such, BUT on a budget if not who cares about them.
If not on a budget a small profile 285-970 and I5(+ ITX) will use 250 watts on load and be fast and quiet as well.
Actually going to be building a 300 PC for my dad its going to have a haswell pentium with 8GB of ram and a SSD. Will probably use 20-25 watts most of the time.
I could probably drop to 4GB of ram and include a unlocked pentium and have a 750Ti for 400$ for budget gaming machines. I dropped my I7 down to a dual core and a lot of games still played fine but some did require a bit work like far cry 4.

A CPU for gaming really needs 4 threads even if its done using HT.

 

[jdwii]

AMD to introduce its Radeon R9 300-series lineup at Computex
http://www.kitguru.net/components/graphic-cards/anton-shilov/amd-to-introduce-its-radeon-r9-300-series-lineup-at-computex/
meanwhile, nvidia let maximum pc take pics of it's new titan x gfx card. pc gpu war is slowly heating up.
kitguru doesn't mention the recent claim that amd has excess stock of current gen cards. if amd doesn't clear the excess cards they risk in-competition with the new cards. how the heck did amd come up with an entirely new lineup. i've been expecting only one two new flagship and the rest being rebrands ever since the news broke.

I think (hope) most people can see through the corporate politics. Mantle went as far as it could being an AMD creation. Once they passed on the core of it to Khronos and give it a new name, swap in/out some features, other companies can get behind it. Squabbling over the size or percentage of the donation is just petty. We all benefit from it in the long run if you use any nix systems.

+1. this is corporate politics.
amusingly, the people singing mantle's death seem to be mostly tech journalists or i.s.v. people. trevett was the only one close to an authority figure who downplayed mantle's contribution and he incidentally happens to be an nvidia v.p. i could go full conspiracy theorist and twist the events into a subtle smear campaign against amd's achievements . i wonder if it's too far fetched with the new gfx card launches so near. 😗

Great links but man that first one was depressing, and yes i think Mantle is dead but it was reborn for the better its not like Amd lost they got what they wanted but people can use the API on everything not just a company with limited market share. One thing is for sure developers care more about these new API's then 10,11 ever seen.

The challenge for Advanced Micro Devices is that it needs to achieve two rather different (and sometimes opposing) tasks in the first half of this year. It must significantly reduce inventory in the channel and it has to try and boost its dropping market share. If AMD does not ship anything new, its own sales and market share will decline. However, if AMD ships brand-new offerings, older hardware stalls in the channel, impacts prices and makes it difficult to sell new products.

 

[con635]

Not off topic at all, APU's are still great if you want a computer that can game without a GPU and with a small itx board and such, BUT on a budget if not who cares about them.
If not on a budget a small profile 285-970 and I5(+ ITX) will use 250 watts on load and be fast and quiet as well.
Actually going to be building a 300 PC for my dad its going to have a haswell pentium with 8GB of ram and a SSD. Will probably use 20-25 watts most of the time.
I could probably drop to 4GB of ram and include a unlocked pentium and have a 750Ti for 400$ for budget gaming machines. I dropped my I7 down to a dual core and a lot of games still played fine but some did require a bit work like far cry 4.

Would you not consider used market for a budget build? Last new budget build I done was built around fm2+, athlon and 260x, for same-ish money I've got my brother-in-laws used build nearly finished, b85, antec psu, haswell i5 and maybe a 280 or 280x not bought gpu yet. My own sig rig is used, bought from local 'buy and sell' and ebay, cheaper than i3, 750ti, even psu was cheaper than cx430. Of course its not for everybody but I think its insanity to buy new atm if on a tight budget, eg 280x's are 100-140£ here at the moment.

 

[jdwii]

Not off topic at all, APU's are still great if you want a computer that can game without a GPU and with a small itx board and such, BUT on a budget if not who cares about them.
If not on a budget a small profile 285-970 and I5(+ ITX) will use 250 watts on load and be fast and quiet as well.
Actually going to be building a 300 PC for my dad its going to have a haswell pentium with 8GB of ram and a SSD. Will probably use 20-25 watts most of the time.
I could probably drop to 4GB of ram and include a unlocked pentium and have a 750Ti for 400$ for budget gaming machines. I dropped my I7 down to a dual core and a lot of games still played fine but some did require a bit work like far cry 4.

Would you not consider used market for a budget build? Last new budget build I done was built around fm2+, athlon and 260x, for same-ish money I've got my brother-in-laws used build nearly finished, b85, antec psu, haswell i5 and maybe a 280 or 280x not bought gpu yet. My own sig rig is used, bought from local 'buy and sell' and ebay, cheaper than i3, 750ti, even psu was cheaper than cx430. Of course its not for everybody but I think its insanity to buy new atm if on a tight budget, eg 280x's are 100-140£ here at the moment.

Oh yeah i easily would consider the used market. Linus made some good videos about it(favorite from him) and he bought parts from the used market for gaming. 400$ is easily enough if you are willing to wait and get the best deals.

 

[de5_Roy]

AMD to release FreeSync-capable drivers on March 19
http://www.tweaktown.com/news/43955/amd-release-freesync-capable-drivers-march-19/index.html
very slow news day... yawn....

 

[gamerk316]

Great links but man that first one was depressing, and yes i think Mantle is dead but it was reborn for the better its not like Amd lost they got what they wanted but people can use the API on everything not just a company with limited market share. One thing is for sure developers care more about these new API's then 10,11 ever seen.

DX10 really didn't add anything *new* to the API. It was more of a re-design to make the API work with the new driver model, which BTW was a VERY good thing, when you consider how piss poor GPU drivers were back on XP. 11 added a little (Tesselation), and put in the first attempt at multithreaded rendering, limiting though it was. 12 is adding the ability to code at lower levels to improve performance.

Notice the lack of new graphical features? That's because there's very little left to do that isn't too computationally expensive. Any additional features basically requires Ray Tracing. There's really nothing in DX12 that DX9.0c couldn't do, just with a lot lower performance. That's where DX has been going for a few years now.

 

[8350rocks]

This will be worse than normal. Phenom II was still competitive, price/performance wise, vs Sandy Bridge. Piledriver FX did well in that regard, and even with ivy bridge, to an extent. The gap widened a bit with Haswell. With Broadwell and Skylake on the way, that gap is even worse. AMD needs that new arch now, not a year from now. 🙁

Piledriver is/was far from competitive and the reason why AMD did lost server/desktop/laptop market share and money. AMD needs the new arch for yesterday, but will be not ready before end 2016 early 2017.

AMD unofficially confirms Radeon Flagship – R9 390X - Launches at Computex:
http://www.guru3d.com/news-story/amd-unofficially-confirms-radeon-flagship-%E2%80%93-r9-390x-launches-at-computex.html

Uhm... The chart says ~8 TFlops; that's roughly equivalent to the Titan Z and the Titan X is just around the corner (me thinks). I wonder how much time the 390X will have as king. If nVidia didn't increase the flops for the X and focused only in efficiency, then they could have the upper hand in compute per watt (pro segment digs that, haha).

Nvidia presented the Titan X as the fastest GPU in the world. According to preliminary benchmarks Titan X will be faster than 390X at 1080p, and both will be neck to neck at 4K.

Both those cards are not meant for 1080p. Who cares which will be faster at that resolution. Totally looking at 4k with these cards. Do you play computer games?

By giving performance at two resolutions, 1080p and 4K, you can extrapolate to obtain an approx. idea of how the cards will behave at some other resolution. So simple like that.

Actually...if you are talking about extrapolating numbers toward a bigger resolution, then 1080p has little to do with anything. 8K would be far more accurately extrapolated by performance on 4K. If your frame buffer and processing is not good at 4K, but excellent at 1080p, then you have a poor design that has issues with efficiency of frame buffer utilization, and also likely bandwidth problems.

Perfect example:

R9-295X2 can play many games reliably at 4K between 30-60 FPS

R9-270X can play many games reliably at 1080p between 30-60 FPS (or more)

By your logic, the 270X should be just as valid at 8K as the 295X2 because we are looking at 1080p and 4K.

Do better "research" next time, your non-engineer is showing.

Am I on the right thread, or does the title need to be changed?
Seriously, I think I may need to stop following this thread.
Too much trouble to weed through all the hate, doom and gloom and self-back patting to find news about AMD.
I don't give a hoot about who's predictions were right or wrong, or comparing three year old cpu's to present.
I am using an FX, and it works great for all my needs.

I am writing this from AMD hardware. Does it mean that I must close my eyes and distort reality or just select news are only good to AMD? No, thanks.

Truthfully, you are not contributing anything useful to the discussion besides venom...and your alter ego jdwii is not helping much either.

There is an Intel thread if you want to bash AMD.

For all of you who own AMD hardware and complain...aside from benchmarks being slightly less, are you honestly unhappy with what your hardware can do? If so, why?

AM3+ steamroller/excavator don't really make sense any more because they are both mobile focused designs. It would mean releasing new 95W cpus that perform like the 125W piledriver ones but having them not even OC as high and it would end up pointless for the desktop space for most people.

Steamroller improved IPC over Vishera as much as Visher did over Zambezi. It also used less power and still could clock to about 4.4 GHZ, which is good enough for 99 percent of users. . Excavator improves the situation even further.

So yes, it would definitely be a worthwhile upgrade. It wouldn't overtake intel, but a 15-20 percent IPC improvement and lower power would make it a much more compelling product.

Dude steamroller isn't anywhere close to a 20% upgrade in IPC over piledriver not sure why people think this when the majority of benchmarks claim otherwise.

Actually, it is about 10-12% on average, depending on the task, 15-20% is not out of the ballpark. The issue is, you lose some of that progress with lower clocks...

juanrga wrote:

Any bet that when Zen comes and it is not a Skylake killer, don't give AMD lots of money and don't give AMD the 80% of the market, I will be hearing something like:

"Hey, Zen was a first attempt, wait for post-Zen it will fix everything and will deliver the promises."

Depends on whether Zen does enough to deserve the benefit of the doubt. Its unrealistic to expect a huge jump in one generation. It IS realistic to expect a significant improvement. We will see.

However, it is the first project from this team. I'm not so quick to throw them to the wolves. They have done a lot with someone else's trash. lets see what they can do with a project of their own.

I expect Zen core to have ~40% better IPC (integer) than Piledriver and twice bigger FPU. I think this is realistic according to the little info I have about the arch. However, certain AMD engineer did claim that the goal is "100% higher IPC than Piledriver". This is the same kind of claims made during Bulldozer.

Not a chance. The size of the market would purchase an AM3+ version of steamroller/excavator is 0.01% or so. The cancellation of servers based in Steamroller/Excavator implied cancellation of AM3+, because HEDT alone is not economically viable.

Zen will be released in servers, and thus a HEDT version will come to 'AM4' (the real name of the socket is other).

Not a chance what? I didn't say they were going to release one. I said I would have. Or at least on FM2+. The FX didn't have to be AM3+, but it would have been a nice gesture to the people who purchased those boards.

The market would have been bigger than 0.01. Your just making up numbers.

And I said that the chance of AMD following your advice/desire was exactly zero because was economically inviable. The "0.01% or so" comes from AMD sales of FX (AM3+) chips.

If they can pull 40% more out of Zen, they will literally be on par/better than haswell, and not far behind skylake.

However, you are doom and gloom over it?

Think about what you said 40%. Considering that since SB, Intel has not gained more than 30% improvement in 3 generations...you are talking about a bigger increase than that in one leap.

That would be an engineering marvel above and beyond the likes of which even Intel could claim to have pulled off...ever.

Yes, they are shooting for 100% more IPC. If you are going to aim big, shoot for the stars, at least you will hit the moon. In reality, I think you are honestly pretty close on this prediction; however, I suspect it is more pulling numbers from a body orifice than actually having anything concrete. Time will tell, and we shall see...

 

[de5_Roy]

MediaTek to license AMD graphics
http://www.fudzilla.com/news/graphics/37209-mediatek-to-license-amd-graphics

The fact that Samsung 14nm SoC works surprised the industry
http://www.fudzilla.com/news/processors/37210-the-fact-that-samsung-14nm-soc-works-surprised-the-industry

@8350rocks: aren't you trying to distract us from the upcoming gpu launch? :3 you need to dish on this "all new gpus" stuff i keep reading about. are they 20nm or not?

 

[juanrga]

By giving performance at two resolutions, 1080p and 4K, you can extrapolate to obtain an approx. idea of how the cards will behave at some other resolution. So simple like that.

Actually...if you are talking about extrapolating numbers toward a bigger resolution, then 1080p has little to do with anything. 8K would be far more accurately extrapolated by performance on 4K. If your frame buffer and processing is not good at 4K, but excellent at 1080p, then you have a poor design that has issues with efficiency of frame buffer utilization, and also likely bandwidth problems.

I didn't mention 8K, you did. Most gamer will play at sub-4K resolutions not at 8K.

Also an extrapolation requires at least two points, in this case 1080p and 4K. You cannot extrapolate using only 4K as you pretend. Check some basic math

http://en.wikipedia.org/wiki/Extrapolation#Linear_extrapolation

I expect Zen core to have ~40% better IPC (integer) than Piledriver and twice bigger FPU. I think this is realistic according to the little info I have about the arch. However, certain AMD engineer did claim that the goal is "100% higher IPC than Piledriver". This is the same kind of claims made during Bulldozer.

If they can pull 40% more out of Zen, they will literally be on par/better than haswell, and not far behind skylake.

Nope. 40% implies it will be behind Haswell and far from Skylake. Note this 40% is only for integer. The FP gap will be huge.

Think about what you said 40%. Considering that since SB, Intel has not gained more than 30% improvement in 3 generations...you are talking about a bigger increase than that in one leap.

That would be an engineering marvel above and beyond the likes of which even Intel could claim to have pulled off...ever.

Yes, they are shooting for 100% more IPC. If you are going to aim big, shoot for the stars, at least you will hit the moon. In reality, I think you are honestly pretty close on this prediction; however, I suspect it is more pulling numbers from a body orifice than actually having anything concrete. Time will tell, and we shall see...

This is all wrong. First, increasing 40% IPC over Piledriver, starting from scratch, is feasible (but hard) because the Piledriver arch is a complete disaster and because a blank piece of paper means engineers have no backward compatibility compromises. You confound that with Increasing IPC by evolving/tweaking an existent architecture as SB.

In fact whereas improving 40% over Piledriver starting from scratch is a feasible target. AMD is only improving 5--15% per gen by evolving/tweaking Piledriver. Check Steamroller and Excavator. Excavator brings a mere 5% over Steamroller.

The same comments about Intel, improving 5--10% per gen by evolving evolving/tweaking SB is just as expected, but with one remark. SB is actually one of the best archs in the world (only IBM can compete) and adding 5% on top of Haswell is much more complex and difficult than adding 5% on top of Steamroller.

100% over Piledriver is pure nonsense even starting a new design from scratch. Some day I will say the name of the engineer behind the hype. I guess you already know whom he is.

 

[etayorius]

Well i don't know, Haswell is already about 60% ahead of Piledriver in IPC... if we take into account 5% more from Broadwell and 5% (minimum) more from Skylake that is a good 70% ahead of Piledriver, i honestly don't think AMD can manage a 100% more in one generation, i seriously don't buy it.

40% is good and reasonable, i would definitely buy a Zen system if it were out right now considering AMD prices seem to be cheaper most of the time compared to damn Intel, sadly 2016 is waay to late (at least for me and i have been waiting since 2011).

Every little gains in IPC helps AMD, and i really hope they manage at least 40% because anything less would just be a another fail and i doubt AMD can survive another Faildozer.

I really want to see AMD succeed, i definitely do not want to see them go or transform into VIA but the state in which their current CPUs is honestly shameful to be honest, i can understand people who need more than 2 cores using them, but as a gamer i see little need of their piledriver CPUs. Also games are barely starting to use 4 cores, not even MANTLE can take advantage of more than 4 as we seen in benchmarks with the i7 with 8 cores, i was expecting waaay more performance from that CPU, but all the gains in FPS was due to the super fast IPC instead of the 8 cores with HT.

 

[de5_Roy]

BIOSTAR Reveals the TA970 Plus AMD Mainboard with ACC
http://www.techpowerup.com/210508/biostar-reveals-the-ta970-plus-amd-mainboard-with-acc.html
finally, a budget am3+ mobo that doesn't look like 2011.

 

[8350rocks]

Well i don't know, Haswell is already about 60% ahead of Piledriver in IPC... if we take into account 5% more from Broadwell and 5% (minimum) more from Skylake that is a good 70% ahead of Piledriver, i honestly don't think AMD can manage a 100% more in one generation, i seriously don't buy it.

40% is good and reasonable, i would definitely buy a Zen system if it were out right now considering AMD prices seem to be cheaper most of the time compared to damn Intel, sadly 2016 is waay to late (at least for me and i have been waiting since 2011).

Every little gains in IPC helps AMD, and i really hope they manage at least 40% because anything less would just be a another fail and i doubt AMD can survive another Faildozer.

I really want to see AMD succeed, i definitely do not want to see them go or transform into VIA but the state in which their current CPUs is honestly shameful to be honest, i can understand people who need more than 2 cores using them, but as a gamer i see little need of their piledriver CPUs. Also games are barely starting to use 4 cores, not even MANTLE can take advantage of more than 4 as we seen in benchmarks with the i7 with 8 cores, i was expecting waaay more performance from that CPU, but all the gains in FPS was due to the super fast IPC instead of the 8 cores with HT.

What is everyone smoking to claim 60% lead in IPC from intel? Seriously...are you all from Amsterdam, Colorado, or Washington?

http://www.tomshardware.com/charts/cpu-charts-2013/compare,3143.html?prod%5B6234%5D=on&prod%5B5877%5D=on&prod%5B5755%5D=on

There are a few edge cases where the 4770k is about 20-25% faster than the 8350 at stock clocks for both CPUs.

However, that is not 60%, nor is it anywhere near that.

Clock for clock is irrelevant as well...because intel puts a processor out where it is comfortable being compared, and so does AMD. If you want to talk about overclocking something, compare both overclocked, or both stock. Otherwise, you are comparing apples and oranges. As it sits...we are looking at a 25% difference between haswell and FX at best.

Now, if we assume 15% for broadwell and skylake combined...

25+15 = 40%

Zen improvements = 40%

Hmm...interesting...who said they would be close when zen launched?

 

[blackkstar]

By giving performance at two resolutions, 1080p and 4K, you can extrapolate to obtain an approx. idea of how the cards will behave at some other resolution. So simple like that.

Actually...if you are talking about extrapolating numbers toward a bigger resolution, then 1080p has little to do with anything. 8K would be far more accurately extrapolated by performance on 4K. If your frame buffer and processing is not good at 4K, but excellent at 1080p, then you have a poor design that has issues with efficiency of frame buffer utilization, and also likely bandwidth problems.

I didn't mention 8K, you did. Most gamer will play at sub-4K resolutions not at 8K.

Also an extrapolation requires at least two points, in this case 1080p and 4K. You cannot extrapolate using only 4K as you pretend. Check some basic math

http://en.wikipedia.org/wiki/Extrapolation#Linear_extrapolation

I expect Zen core to have ~40% better IPC (integer) than Piledriver and twice bigger FPU. I think this is realistic according to the little info I have about the arch. However, certain AMD engineer did claim that the goal is "100% higher IPC than Piledriver". This is the same kind of claims made during Bulldozer.

If they can pull 40% more out of Zen, they will literally be on par/better than haswell, and not far behind skylake.

Nope. 40% implies it will be behind Haswell and far from Skylake. Note this 40% is only for integer. The FP gap will be huge.

Think about what you said 40%. Considering that since SB, Intel has not gained more than 30% improvement in 3 generations...you are talking about a bigger increase than that in one leap.

That would be an engineering marvel above and beyond the likes of which even Intel could claim to have pulled off...ever.

Yes, they are shooting for 100% more IPC. If you are going to aim big, shoot for the stars, at least you will hit the moon. In reality, I think you are honestly pretty close on this prediction; however, I suspect it is more pulling numbers from a body orifice than actually having anything concrete. Time will tell, and we shall see...

This is all wrong. First, increasing 40% IPC over Piledriver, starting from scratch, is feasible (but hard) because the Piledriver arch is a complete disaster and because a blank piece of paper means engineers have no backward compatibility compromises. You confound that with Increasing IPC by evolving/tweaking an existent architecture as SB.

In fact whereas improving 40% over Piledriver starting from scratch is a feasible target. AMD is only improving 5--15% per gen by evolving/tweaking Piledriver. Check Steamroller and Excavator. Excavator brings a mere 5% over Steamroller.

The same comments about Intel, improving 5--10% per gen by evolving evolving/tweaking SB is just as expected, but with one remark. SB is actually one of the best archs in the world (only IBM can compete) and adding 5% on top of Haswell is much more complex and difficult than adding 5% on top of Steamroller.

100% over Piledriver is pure nonsense even starting a new design from scratch. Some day I will say the name of the engineer behind the hype. I guess you already know whom he is.

I think you need to check your math, friend. 10% increase from BD -> PD -> SR -> EX is a 33.1% increase in performance. 15% increase is a 52% IPC increase from BD to EX.

If you assume EX has 15% IPC over SR and SR has 15% over PD, then EX has 32% better IPC than PD. So really, I don't think 40% over PD with Zen is anything exciting. In fact, it wouldn't be much better than EX.

You did it again, Juan, you disproved another AMD product with your marvellous mathematical abilities. Someone give this man a Nobel for his brilliant math already!

 

[COLGeek]

Civility, please. No need to get snippy with one another. Some of you need to take a deep breath and relax a bit before continuing to engage other members in the thread.

Attack positions, not each other. Thank you.

 

[logainofhades]

BIOSTAR Reveals the TA970 Plus AMD Mainboard with ACC
http://www.techpowerup.com/210508/biostar-reveals-the-ta970-plus-amd-mainboard-with-acc.html
finally, a budget am3+ mobo that doesn't look like 2011.

Still would rather have the MSI 970 Gaming.

http://us.msi.com/product/mb/970-GAMING.html#hero-overview

 

[cdrkf]

Well i don't know, Haswell is already about 60% ahead of Piledriver in IPC... if we take into account 5% more from Broadwell and 5% (minimum) more from Skylake that is a good 70% ahead of Piledriver, i honestly don't think AMD can manage a 100% more in one generation, i seriously don't buy it.

40% is good and reasonable, i would definitely buy a Zen system if it were out right now considering AMD prices seem to be cheaper most of the time compared to damn Intel, sadly 2016 is waay to late (at least for me and i have been waiting since 2011).

Every little gains in IPC helps AMD, and i really hope they manage at least 40% because anything less would just be a another fail and i doubt AMD can survive another Faildozer.

I really want to see AMD succeed, i definitely do not want to see them go or transform into VIA but the state in which their current CPUs is honestly shameful to be honest, i can understand people who need more than 2 cores using them, but as a gamer i see little need of their piledriver CPUs. Also games are barely starting to use 4 cores, not even MANTLE can take advantage of more than 4 as we seen in benchmarks with the i7 with 8 cores, i was expecting waaay more performance from that CPU, but all the gains in FPS was due to the super fast IPC instead of the 8 cores with HT.

What is everyone smoking to claim 60% lead in IPC from intel? Seriously...are you all from Amsterdam, Colorado, or Washington?

http://www.tomshardware.com/charts/cpu-charts-2013/compare,3143.html?prod%5B6234%5D=on&prod%5B5877%5D=on&prod%5B5755%5D=on

There are a few edge cases where the 4770k is about 20-25% faster than the 8350 at stock clocks for both CPUs.

However, that is not 60%, nor is it anywhere near that.

Clock for clock is irrelevant as well...because intel puts a processor out where it is comfortable being compared, and so does AMD. If you want to talk about overclocking something, compare both overclocked, or both stock. Otherwise, you are comparing apples and oranges. As it sits...we are looking at a 25% difference between haswell and FX at best.

Now, if we assume 15% for broadwell and skylake combined...

25+15 = 40%

Zen improvements = 40%

Hmm...interesting...who said they would be close when zen launched?

I think it depends on what type of application. In well threaded apps the difference isn't big (actually I think teh FX chips are great value in that context). The issue is single thread, and in those situations the gap can be pretty big (60% isn't out of the realms, I'd have to look at some benches, although I'll admit a gap that large *is worst case*).

You mentioned about anyone actually having issues with AMD kit, well I'm an AMD fan (have lots of AMD based machines both at home at at work- AMD's cheap multi core procs are excellent for cpu based ray trace rendering which is something I do lots). I do have an issue with my home rig however, which is an FX 8320 + R9 280 setup. The cpu really holds that graphics card back in one game in particular (planetary annihilation) compared to an Intel based rig, to the point where I've hardly noticed an improvement from the Phenom II X3 710 I upgraded from (which is only clocked at 2.6ghz). The game in question actually appears to perform comparatively better on Phenom II which leads me to believe there is something in the coding that doesn't suit FX at all. Plenty of videos of casts and such have the game running totally smooth (these are games *I've been playing in*) where I'm getting 12fps and the gpu usage is really low (~40%). I've checked everything and there's no specific problems with the machine (can complete stability tests, 3D mark and runs other games fine). I've even tried the game under Linux as I was concerned windows thread management was causing an issue however the performance is pretty much the same.

The sad thing is that during development, the developers reached out to AMD, Intel and nVidia for help as this was a new engine developed from scratch. Out of the 3 companies, only nVidia actually helped- to the level of sending out a software engineer to help with the rendering engine. That said they've got the game running pretty well on Radeon cards, although I'm sure AMD could have advised them how to eek a bit more oomph out of their kit had they responded. I think to some extent AMD are kinda their own worst enemies. Now I'm left with in the position of contemplating getting a Haswell i3 as it will provide a more consistent basis for gaming. I don't care about benchmark results, however I would have expected the fx cpu to be able to keep up well enough to not totally stall out like this.

This is the problem, a difference of 10 fps when a game is already running at 60+ fps means nothing. A difference of 10 fps when your running at 12fps for a large scale RTS like this means the difference between winning and loosing. Essentially I can't compete in large team games on this hardware whereas if I had an Intel rig I would be able to :/

I got the FX due to the fact it was a drop in upgrade (maintaining the platform like this is something I think AMD has got very right btw), so it's not such a bad thing. It's just frustrating as I really don't have a better option from AMD so my only choices now are a: Live with it or b: Go Intel.

AMD can't get that new uarch out soon enough imo.

 

[8350rocks]

Have you tried overclocking? I have a friend running an old FX6100, and over the weekend he complained about how it was not competitive. He was still running stock clocks at 3.3 GHz. I set him up on aftermarket cooler with 2 aftermarket fans, and gave him a simple multiplier bump to an OC @ 4.0 GHz.

He literally gained 20 FPS in a game based on CryEngine 3 after spending $40 for hardware and $10 for shipping.

Good luck.

 

[etayorius]

Well i don't know, Haswell is already about 60% ahead of Piledriver in IPC... if we take into account 5% more from Broadwell and 5% (minimum) more from Skylake that is a good 70% ahead of Piledriver, i honestly don't think AMD can manage a 100% more in one generation, i seriously don't buy it.

40% is good and reasonable, i would definitely buy a Zen system if it were out right now considering AMD prices seem to be cheaper most of the time compared to damn Intel, sadly 2016 is waay to late (at least for me and i have been waiting since 2011).

Every little gains in IPC helps AMD, and i really hope they manage at least 40% because anything less would just be a another fail and i doubt AMD can survive another Faildozer.

I really want to see AMD succeed, i definitely do not want to see them go or transform into VIA but the state in which their current CPUs is honestly shameful to be honest, i can understand people who need more than 2 cores using them, but as a gamer i see little need of their piledriver CPUs. Also games are barely starting to use 4 cores, not even MANTLE can take advantage of more than 4 as we seen in benchmarks with the i7 with 8 cores, i was expecting waaay more performance from that CPU, but all the gains in FPS was due to the super fast IPC instead of the 8 cores with HT.

What is everyone smoking to claim 60% lead in IPC from intel? Seriously...are you all from Amsterdam, Colorado, or Washington?

http://www.tomshardware.com/charts/cpu-charts-2013/compare,3143.html?prod%5B6234%5D=on&prod%5B5877%5D=on&prod%5B5755%5D=on

There are a few edge cases where the 4770k is about 20-25% faster than the 8350 at stock clocks for both CPUs.

However, that is not 60%, nor is it anywhere near that.

Clock for clock is irrelevant as well...because intel puts a processor out where it is comfortable being compared, and so does AMD. If you want to talk about overclocking something, compare both overclocked, or both stock. Otherwise, you are comparing apples and oranges. As it sits...we are looking at a 25% difference between haswell and FX at best.

Now, if we assume 15% for broadwell and skylake combined...

25+15 = 40%

Zen improvements = 40%

Hmm...interesting...who said they would be close when zen launched?

Then i have no idea which benchmarks are you looking at:

http://www.anandtech.com/bench/product/1260?vs=1289

http://cpuboss.com/cpus/Intel-Core-i7-4790K-vs-AMD-FX-9590

I don`t know man, did you checked the link you posted? because your own link benchmarks don´t agree with your claims.

Cinebench 11.5 reports FX9590 has 11.5 Single Thread performance while the 4790k has 1.74 which is about 60% more IPC, pretty much in line with what CPUboss and Anandtech are reporting.

May want to check this one out too:

https://www.dropbox.com/s/py1mvuofeihf4sq/xd.jpg?dl=0


We got to be honest, AMD as of now is 60% behind in regards to IPC... it kinda sucks because Intel takes advantage of everyone by charging super premium prices, but that is just how things are now... AMD CPUs as of now gets the job done, the Intel CPUs just smokes the floor with any CPU AMD has to offer.

I do gaming, and even in gaming i seen my older brother 4670k smoke my Younger Brother FX8320 OCed to 4.4 by almost double the FPS in Fifa 15, NFS Most Wanted, Battlefield4 and basically any games, the advantage is real and it is there on Intel side.

My Older brother has a i5 4670k coupled with a GTX750 Ti while my younger brother has a FX8320 coupled with a GTX650 Ti Boost, while the 650Ti Boost is faster than the 750 i am seeing much higher Minimum FPS on my elder brother PC with the titles they both play, though they are both mostly casual PCGamers and i consider myself a hardcore PCGamer and my PC is worst than theirs... shit happens.

 

[juanrga]

Well i don't know, Haswell is already about 60% ahead of Piledriver in IPC... if we take into account 5% more from Broadwell and 5% (minimum) more from Skylake that is a good 70% ahead of Piledriver, i honestly don't think AMD can manage a 100% more in one generation, i seriously don't buy it.

40% is good and reasonable, i would definitely buy a Zen system if it were out right now considering AMD prices seem to be cheaper most of the time compared to damn Intel, sadly 2016 is waay to late (at least for me and i have been waiting since 2011).

Every little gains in IPC helps AMD, and i really hope they manage at least 40% because anything less would just be a another fail and i doubt AMD can survive another Faildozer.

I really want to see AMD succeed, i definitely do not want to see them go or transform into VIA but the state in which their current CPUs is honestly shameful to be honest, i can understand people who need more than 2 cores using them, but as a gamer i see little need of their piledriver CPUs. Also games are barely starting to use 4 cores, not even MANTLE can take advantage of more than 4 as we seen in benchmarks with the i7 with 8 cores, i was expecting waaay more performance from that CPU, but all the gains in FPS was due to the super fast IPC instead of the 8 cores with HT.

What is everyone smoking to claim 60% lead in IPC from intel? Seriously...are you all from Amsterdam, Colorado, or Washington?

http://www.tomshardware.com/charts/cpu-charts-2013/compare,3143.html?prod%5B6234%5D=on&prod%5B5877%5D=on&prod%5B5755%5D=on

There are a few edge cases where the 4770k is about 20-25% faster than the 8350 at stock clocks for both CPUs.

However, that is not 60%, nor is it anywhere near that.

Clock for clock is irrelevant as well...because intel puts a processor out where it is comfortable being compared, and so does AMD. If you want to talk about overclocking something, compare both overclocked, or both stock. Otherwise, you are comparing apples and oranges. As it sits...we are looking at a 25% difference between haswell and FX at best.

Now, if we assume 15% for broadwell and skylake combined...

25+15 = 40%

Zen improvements = 40%

Hmm...interesting...who said they would be close when zen launched?

Your numbers are plain wrong. Haswell is about ~60% ahead of Piledriver per core and another extra ~20% from the CMT module penalty. This is reason why AMD trows 8 cores @ 4GHz against Intel quad cores @ 3.5GHz...

Even if Zen adds a good 40% on top of Piledriver it will be behind Haswell.

This is all wrong. First, increasing 40% IPC over Piledriver, starting from scratch, is feasible (but hard) because the Piledriver arch is a complete disaster and because a blank piece of paper means engineers have no backward compatibility compromises. You confound that with Increasing IPC by evolving/tweaking an existent architecture as SB.

In fact whereas improving 40% over Piledriver starting from scratch is a feasible target. AMD is only improving 5--15% per gen by evolving/tweaking Piledriver. Check Steamroller and Excavator. Excavator brings a mere 5% over Steamroller.

The same comments about Intel, improving 5--10% per gen by evolving evolving/tweaking SB is just as expected, but with one remark. SB is actually one of the best archs in the world (only IBM can compete) and adding 5% on top of Haswell is much more complex and difficult than adding 5% on top of Steamroller.

100% over Piledriver is pure nonsense even starting a new design from scratch. Some day I will say the name of the engineer behind the hype. I guess you already know whom he is.

I think you need to check your math, friend. 10% increase from BD -> PD -> SR -> EX is a 33.1% increase in performance. 15% increase is a 52% IPC increase from BD to EX.

If you assume EX has 15% IPC over SR and SR has 15% over PD, then EX has 32% better IPC than PD. So really, I don't think 40% over PD with Zen is anything exciting. In fact, it wouldn't be much better than EX.

You did it again, Juan, you disproved another AMD product with your marvellous mathematical abilities. Someone give this man a Nobel for his brilliant math already!

Your post is full of mistakes.

First, I did start from Piledriver. You start from Bulldozer.

Second, you are assuming that you can add 15% to each gen. E.g. you assume Excavator is 15% faster than Steamroller, but your numbers are divorced from reality, because Excavator is only 5% better than Steamroller.

Third, the mentioned 40% over piledriver for Zen is per core. Steamroller 15% IPC gain was only for multitreaded workloads thanks to doubling the decoder and eliminating most of the module penalty in Piledriver, You are mixing multitread numbers for modules with single-thread numbers per core. Another mistake.

Steamroller increase was only 0--5% per core. Excavator brings another 5% IPC gain. Thus Excavator is about 5--10% faster than Piledriver per core.

That 5--10% per core is very far from the improvement that AMD needs to be competitive, that is why Keller is designing Zen from scratch. Bulldozer family ends with Excavator.

 

[juanrga]

Well i don't know, Haswell is already about 60% ahead of Piledriver in IPC... if we take into account 5% more from Broadwell and 5% (minimum) more from Skylake that is a good 70% ahead of Piledriver, i honestly don't think AMD can manage a 100% more in one generation, i seriously don't buy it.

40% is good and reasonable, i would definitely buy a Zen system if it were out right now considering AMD prices seem to be cheaper most of the time compared to damn Intel, sadly 2016 is waay to late (at least for me and i have been waiting since 2011).

Every little gains in IPC helps AMD, and i really hope they manage at least 40% because anything less would just be a another fail and i doubt AMD can survive another Faildozer.

I really want to see AMD succeed, i definitely do not want to see them go or transform into VIA but the state in which their current CPUs is honestly shameful to be honest, i can understand people who need more than 2 cores using them, but as a gamer i see little need of their piledriver CPUs. Also games are barely starting to use 4 cores, not even MANTLE can take advantage of more than 4 as we seen in benchmarks with the i7 with 8 cores, i was expecting waaay more performance from that CPU, but all the gains in FPS was due to the super fast IPC instead of the 8 cores with HT.

Yes, Haswell is about 60% faster than Piledriver clock for clock. But this 60% is per core and only for integer workloads.

Since Zen is being designed for servers and will first appear on servers, it is good to compare server workloads between Intel and AMD

http://spec.org/cpu2006/results/res2014q3/cpu2006-20140715-30431.html
http://spec.org/cpu2006/results/res2014q3/cpu2006-20140728-30675.html

Haswell quad-core @3.4GHz scores 59.6
Steamroller quad-core @3.7GHz scores 30.1

Haswell was ~2x faster than Steamroller in this server benchmark.

About Skylake, first leaked benchmarks suggest that Skylake is 20% faster than Haswell on integer.