AMD CPU speculation... and expert conjecture

[de5_Roy]

Spoiler

Well, they did their homework with PD and increased FP throughout put by 25%. If they increase around stuf f(cache latency, clock, etc) the overall package will be good IMO. IB was in striking range from AMD's stong point in PD (integer perf), so they'll be focusing on Integer now. Seems logical for the server fight.

Also, not news, but a good piece of history:
http://arstechnica.com/business/2013/04/the-rise-and-fall-of-amd-how-an-underdog-stuck-it-to-intel/

Cheers!

thanks for the link.
interesting read, to say the least. confirmed quite a few of my suspicions. amd's over-producing of chips, [strike]loser[/strike] 'second best' mentality, chronic poor execution, excessive 'multi-tasking' etc. although, their 'now, intel runs circles around amd cpus' bit in the intro was a little harsh....
edit: i hope amd fanboys don't denounce ars as well.

 

[8350rocks]

http://www.theregister.co.uk/2012/08/29/amd_papermaster_steamroller_seamicro_opteron/

Add all the changes up on the integer side of the core module and there will be a 30 per cent increase in operations per clock cycle for Steamroller over Piledriver. That's a process node or so right there without actually having to go through it. There's hope for AMD yet. It's a pity that these Steamroller changes can't be combined with a jump from 28 nanometer wafer baking processes down to 22 nanometer or 20 nanometer processes at the same time.

ah, that one. it still doesn't say overall 30% speedup from piledriver, unless i've misunderstood what '30% max improvement on integer 'side'' means...unless the cpu has no other 'sides' or 'cache' and the performance trickles down (or bursts out, if it's that big) all the way to end user without any hindrance. looks a lot like a single area (of the cpu) performance improvement before other factors like cache, component choice comes into play - just like other individual, double-digit performance improvements were being discussed. i'll even say that the source that this article misinterprets (and you're seemingly trying to misrepresent ) doesn't even mention piledriver in that context, it just says '30% improvement'.
as for processes, i don't know much about them but i keep reading that there isn't 'much difference' between 28nm bulk and 32nm soi processes, so it could end up being another '"look we're 28nm but not telling which kind!! ". i am just happy that i probably won't be confused and disappointed if the cpu behaves like a 32nm one. i'll still be ignorant about what those differences mean though. :lol:

btw, can anyone help me understand what the heck does integer side even mean? seems like a carefully chosen phrase to appease the audience or to confuse 'less knowing' journalists (looking atcha wccfghijklmnocredtech). does integer side mean the integer clusters? integer calculations? "bulldozer: the electric boogaloo"?

anyway, my point was to not question what amd can or cannot achieve, just the spreading of what seems like misinformation.

A 30% improvement in OPC is not a theoretical increase, it's an increase in actual performance...the theoretical gains were actually a bit higher than than that 37.5% to be exact...so the increase in throughput will scale closely with the chip's theoretical performance increase. They were talking about the advantages to the increase in integer calculation areas of the architecture will increase OPC because it will be operating more efficiently...The only other thing to discuss besides integer was FPU and they've addressed that even more aggressively than the integer side of the CPU, as I discuss below.

@gamerk316: To answer the question about FPUs, in Steamroller, the FPUs will be beefed up as well...each separate FPU will be able to do 1 additional DP MAD FP calculation...which means effectively a 50% increase in FP capability...(going from 2 DP MAD FP calculations per unit to 3 per unit is a substantial % increase over PD). So, steamroller is a large increase across the board...hence the 30% figure.

 

[de5_Roy]

A 30% improvement in OPC is not a theoretical increase, it's an increase in actual performance...the theoretical gains were actually a bit higher than than that 37.5% to be exact...so the increase in throughput will scale closely with the chip's theoretical performance increase. They were talking about the advantages to the increase in integer calculation areas of the architecture will increase OPC because it will be operating more efficiently...The only other thing to discuss besides integer was FPU and they've addressed that even more aggressively than the integer side of the CPU, as I discuss below.

@gamerk316: To answer the question about FPUs, in Steamroller, the FPUs will be beefed up as well...each separate FPU will be able to do 1 additional DP MAD FP calculation...which means effectively a 50% increase in FP capability...(going from 2 DP MAD FP calculations per unit to 3 per unit is a substantial % increase over PD). So, steamroller is a large increase across the board...hence the 30% figure.

i can't find anywhere saying performance improvement over piledriver would be as linear as you make it sound like, let alone 30%. instead of an official or credible confirmation, it looks like conjecture. also, most of these just went from 'oh i know about these too.' to 'i don't know most of these because i never paid attention to them.". the best thing to do right now is to wait for steamroller (not kaveri) to come out and verify these claims with benchmarks. ^_^

 

[8350rocks]

A 30% improvement in OPC is not a theoretical increase, it's an increase in actual performance...the theoretical gains were actually a bit higher than than that 37.5% to be exact...so the increase in throughput will scale closely with the chip's theoretical performance increase. They were talking about the advantages to the increase in integer calculation areas of the architecture will increase OPC because it will be operating more efficiently...The only other thing to discuss besides integer was FPU and they've addressed that even more aggressively than the integer side of the CPU, as I discuss below.

@gamerk316: To answer the question about FPUs, in Steamroller, the FPUs will be beefed up as well...each separate FPU will be able to do 1 additional DP MAD FP calculation...which means effectively a 50% increase in FP capability...(going from 2 DP MAD FP calculations per unit to 3 per unit is a substantial % increase over PD). So, steamroller is a large increase across the board...hence the 30% figure.

i can't find anywhere saying performance improvement over piledriver would be as linear as you make it sound like, let alone 30%. instead of an official or credible confirmation, it looks like conjecture. also, most of these just went from 'oh i know about these too.' to 'i don't know most of these because i never paid attention to them.". the best thing to do right now is to wait for steamroller (not kaveri) to come out and verify these claims with benchmarks. ^_^

They quoted Mark Papermaster, AMD's CTO stating 30%...what more evidence do you need?

 

[JAYDEEJOHN]

Problem is, the good thing is also the bad thing here.
If theoretical peak was good, the real increases wouldnt be that great, good of course, but not great, BUT being the peak isnt good to begin with, it will be great, but from a much lower starting point.
If it scales well with these improvements, it will be a very nice improvement, and we dont know exactly how tied in everything is to complete great scaling, , so if the scaling is still not great, then great improvement wont be seen, good yes, but as far as great goes, expect the worse, hope for the best

 

[de5_Roy]

They quoted Mark Papermaster, AMD's CTO stating 30%...what more evidence do you need?

that's exactly the problem: i can't find anywhere(except the register article) papermaster(or amd) actually saying that 30% improvement is over piledriver. everyone else says just 30% and many hint at bd. worse, i can't find a transcript of the keynote... just to ease my conflicted mind.... the closest thing at hand (related to that keynote) says this at the end:

The information presented in this document is for informational purposes only and may contain technical inaccuracies, omissions and
typographical errors.

i am being this skeptical because in the past, i bought fully into bulldozer previews and 'official numbers'. it looked like amd had a done deal on hand - architecture that competes well aggressively and takes a more elegant approach to it. reality was different.

 

[8350rocks]

I can understand skepticism because of BD, but Jim Keller, one of the fathers of the K6/K7/K8 architecture, is overseeing this one. I think we'll be good.

 

[mayankleoboy1]

I can understand skepticism because of BD, but Jim Keller, one of the fathers of the K6/K7/K8 architecture, is overseeing this one. I think we'll be good.

We all hope it will be good, and perform 30% better than PD.
Realistically, 15% is a better estimate.

 

[gamerk316]

We all hope it will be good, and perform 30% better than PD.
Realistically, 15% is a better estimate.

Based on what I've seen, past trends, and the like:

Single core loaded: ~7.5% increase in performance at the same clockspeed
All cores loaded: ~30% increase in performance at the same clockspeed
4 cores loaded (50% load per core on average): ~17.5% increase in performance at the same clockspeed

I think you will have the same dynamic we did before: Intel wins up to the point where you have four highly loaded cores AND a clockspeed advantage of at least 400MHz. At that point, SR pulls ahead.

 

[juanrga]

I could see 45% in certain applications. But actual core performance, close to 35%. Another thing AMD didn't mention is what speed these CPUs will run at. For all we know they could be clock for clock on par with Intel. Once Kaveri comes around, we'll know what SR is capable of. I predict AMD Kaveri APUs will surpass the i3s in CPU performance and maybe be bihind the i5

If you use biased benchmarks such as Sysmark then Trinity APUs are behind the i3, but if you use fair benchmarks then trinity is already behind the i5.

 

[Cazalan]

They quoted Mark Papermaster, AMD's CTO stating 30%...what more evidence do you need?

Read the fine print.

[2] Based on AMD's internal SIMULATION results of average workloads...

These are software simulations only. Not real world hardware.

 

[lilcinw]

Here is a link to the Hot Chips presentation where Mark is discussing Steamroller. He never explicitly defines the comparison terms he only references the 'Bulldozer architecture' and the 'third generation Steamroller core'.

Personally, since the key note was made prior to Piledriver (which he never mentions) being released I think it is 30% over Bulldozer.

http://www.youtube.com/watch?feature=player_detailpage&v=oi4eq5arzp8#t=1978s

 

[jdwii]

Here is a link to the Hot Chips presentation where Mark is discussing Steamroller. He never explicitly defines the comparison terms he only references the 'Bulldozer architecture' and the 'third generation Steamroller core'.

Personally, since the key note was made prior to Piledriver (which he never mentions) being released I think it is 30% over Bulldozer.

http://www.youtube.com/watch?feature=player_detailpage&v=oi4eq5arzp8#t=1978s

Based on that information it would seem he meant 30% boost over bulldozer which would be 15%-20% boost over Piledriver.

 

[Cazalan]

Kabini for Embedded sighting. Has pricing info too.

http://www.congatec.com/products/com-express-type6/dView/conga-tcg.html

 

[8350rocks]

The Keynote speech was delivered about 30 days prior to the FX piledriver release, but after trinity launch, and after the paper launch of the FX series...I would be fairly certain he was talking about PD, since that technology was already out or arriving. For example, the Richland paper launch has already occurred, and we still do not see them released yet.

 

[mayankleoboy1]

Kabini is for tablets ?

 

[8350rocks]

Kabini is for tablets ?

Kabini is a mobile solution for tablets, etc.

 

[jdwii]

Off topic but i have to say 690>7990 based from power consumption vs performance as well as throwing in noise. Again prerelease drivers but with how old the 7970 is and the 7000 series is Amd is kinda disappointing in the high-end GPU market. Also i'm seeing Nvidia still beating Amd in latency. The 1 plus side is the free games which is kinda amazing. For that fact alone i'd say its worth it.

Read 3 reviews tomshardware techpowerup and anandtech i'd give the card itself a 7.5/10

Pro.
Price/Performance in line with the 690
free games
Con.
Temps, Noise and more importantly power draw
Took forever for the product to come out

 

[Cazalan]

Kabini is for tablets ?

Kabini is 9-25W so not low enough for tablets. It's to replace the 40nm Bobcat systems in low cost laptops/AOI/slimtops. E-350, E2-1800, etc.

That product brief shows Kabini supports ECC so it can be used in server roles.

Temash will be for tablets.

 

[Ranth]

Off topic but i have to say 690>7990 based from power consumption vs performance as well as throwing in noise. Again prerelease drivers but with how old the 7970 is and the 7000 series is Amd is kinda disappointing in the high-end GPU market. Also i'm seeing Nvidia still beating Amd in latency. The 1 plus side is the free games which is kinda amazing. For that fact alone i'd say its worth it.

Read 3 reviews tomshardware techpowerup and anandtech i'd give the card itself a 7.5/10

Pro.
Price/Performance in line with the 690
free games
Con.
Temps, Noise and more importantly power draw
Took forever for the product to come out

Some of the cons I find a bit wrong or minor..
Temps? A bit high on idle, but load? about the same as the 690.. Noise? About the same as 690.. Can't see why temps and noise is a problem, compared to 690. Powerdraw I think is a minor point.... it's a 1000$ card how much do one really care about powerconsumption when paying 1000$?

Otherwise I agree, painfully late...

 

[sarinaide]

OFF TOPIC:

The HD7990 represents enthusiast level, and trust me when I say an enthusiast building a 7990 based system is not going to give a continental about power and noise it uses and there are reasons for that;

1) Surveys suggest that most owners of these cards will use some form of water cooling which has a much higher accoustic level than the card, to compound this they are likely to be watercooling the GPU's so sound and the extra power is hardly a feasible criteria.

2) AMD got very serious with its beta drivers, considering it's beta and more improvements are to come, Crossfire is no longer a legitimate arguement.

3) The extra power corrolates into 90FPS in BF3 at enthusiast level resolutions, this mark was sizeably above the GTX690 and made the Titan look like childs play, so for that added power corrolated into substantial performance and now AMD all but has a card to control the $1000 mark by a comfortable margin.

Nvidia won the last generation, AMD has won this won. To make the nut even sweeter is the fact that AMD's never settle bundle with the HD7990 includes 7 top end titles in the cost so even at $1000 factor in 7 games with it which easily comes to $50-60 per game is a very good bargain imho.

 

[sarinaide]

OFF THE SR TOPIC:

Toms has been doing a lot of strange reviewing of late, they first reviewed budget cards namely the 7790 and 650ti Boost, The Titan and now the 7990, the criterion for assesment on overall opinion has been contradicting to the point where the line of reasoning from a objective reader with no interest or knowledge in components suggested to me the articles are being formulated on the opinion that brand x must at all cost be the favored position.

I first go back to the 7790 then the 650ti Boost(hereinafter referred to as 650), the 650 used more power than the 7850 which these cards are guaged against to replace, generated more heat to deliver almost the same performance, contrasting the 7790 used the same power as the 7770, producing within 10% of a 7850/650 performance and producing 40% less heat. For budget builders the criterion has always been most performance at the lowest possible watt, the point where they intersect determines the value. The other factor for budget builders is the existing hardware ie: Powersupplies, the 650ti needs a lot more power than the 7790 so getting along on a SFX350w may just be touch and go on a 650, possibly hitting 85% up capacity usage under load along with the voltage ripples you will be hard pressed on that capacity, while a 7790 will do it easily.

Toms skewered a result based on performance only at the budget spectrum.

The article on the 7990 based the premise of a decision on Power and Accoustics, foregoing performance, now this card whatever way we slice it is only for enthusiast builiders looking for $2000-5000+ builds, likely have custom cooling or 900D's with sound dampning, probably don't worry as the card will be watercooled. Simply put no mainstream builder in the 500-1500 bracket will ever use this card so its obselete to ever consider it. The enthusiast cares only for performance at all costs, accoustics and power is irrelevant as most are rocking 1.5kw supplies and are nowhere near capacity. On performance the 7990 backs its specs with its metrics which are industry leading and for that is the best dual graphics card and best graphics card period. They then came up with this other standard they termed "Feeling" their feelings and some randoms "feelings or perceptions". When you start having feelings and perceptions as grounds of justification you often opening upto subjectiveness and that is never good in any review. Chris and Don or Igor whomever it was are know Intel/Nvidia users so the sentiment lends to brand bias. The review is thus not based on science or metrics its formed on subjectivity and its just not good enough.

 

[gamerk316]

Off topic but i have to say 690>7990 based from power consumption vs performance as well as throwing in noise. Again prerelease drivers but with how old the 7970 is and the 7000 series is Amd is kinda disappointing in the high-end GPU market. Also i'm seeing Nvidia still beating Amd in latency. The 1 plus side is the free games which is kinda amazing. For that fact alone i'd say its worth it.

Read 3 reviews tomshardware techpowerup and anandtech i'd give the card itself a 7.5/10

Pro.
Price/Performance in line with the 690
free games
Con.
Temps, Noise and more importantly power draw
Took forever for the product to come out

On the latency problem:

http://www.pcper.com/reviews/Graphics-Cards/Frame-Rating-AMD-Improves-CrossFire-Prototype-Driver

Looks like AMD fixed CF for single monitor solutions at least, so its a start.

 

[de5_Roy]

Spoiler

OFF THE SR TOPIC:

Toms has been doing a lot of strange reviewing of late, they first reviewed budget cards namely the 7790 and 650ti Boost, The Titan and now the 7990, the criterion for assesment on overall opinion has been contradicting to the point where the line of reasoning from a objective reader with no interest or knowledge in components suggested to me the articles are being formulated on the opinion that brand x must at all cost be the favored position.

I first go back to the 7790 then the 650ti Boost(hereinafter referred to as 650), the 650 used more power than the 7850 which these cards are guaged against to replace, generated more heat to deliver almost the same performance, contrasting the 7790 used the same power as the 7770, producing within 10% of a 7850/650 performance and producing 40% less heat.

For budget builders the criterion has always been most performance at the lowest possible watt, the point where they intersect determines the value.

Spoiler

The other factor for budget builders is the existing hardware ie: Powersupplies, the 650ti needs a lot more power than the 7790 so getting along on a SFX350w may just be touch and go on a 650, possibly hitting 85% up capacity usage under load along with the voltage ripples you will be hard pressed on that capacity, while a 7790 will do it easily.

Toms skewered a result based on performance only at the budget spectrum.

The article on the 7990 based the premise of a decision on Power and Accoustics, foregoing performance, now this card whatever way we slice it is only for enthusiast builiders looking for $2000-5000+ builds, likely have custom cooling or 900D's with sound dampning, probably don't worry as the card will be watercooled. Simply put no mainstream builder in the 500-1500 bracket will ever use this card so its obselete to ever consider it. The enthusiast cares only for performance at all costs, accoustics and power is irrelevant as most are rocking 1.5kw supplies and are nowhere near capacity. On performance the 7990 backs its specs with its metrics which are industry leading and for that is the best dual graphics card and best graphics card period. They then came up with this other standard they termed "Feeling" their feelings and some randoms "feelings or perceptions". When you start having feelings and perceptions as grounds of justification you often opening upto subjectiveness and that is never good in any review. Chris and Don or Igor whomever it was are know Intel/Nvidia users so the sentiment lends to brand bias. The review is thus not based on science or metrics its formed on subjectivity and its just not good enough.

if i hadn't known about you being biased towards amd, i woulda suspected you of championing intel's desktop core i3, i5 cpus against amd's desktop fx cpus and llano and trinity apus. ;-D

 

[sarinaide]

Spoiler

OFF THE SR TOPIC:

Toms has been doing a lot of strange reviewing of late, they first reviewed budget cards namely the 7790 and 650ti Boost, The Titan and now the 7990, the criterion for assesment on overall opinion has been contradicting to the point where the line of reasoning from a objective reader with no interest or knowledge in components suggested to me the articles are being formulated on the opinion that brand x must at all cost be the favored position.

I first go back to the 7790 then the 650ti Boost(hereinafter referred to as 650), the 650 used more power than the 7850 which these cards are guaged against to replace, generated more heat to deliver almost the same performance, contrasting the 7790 used the same power as the 7770, producing within 10% of a 7850/650 performance and producing 40% less heat.

For budget builders the criterion has always been most performance at the lowest possible watt, the point where they intersect determines the value.

Spoiler

The other factor for budget builders is the existing hardware ie: Powersupplies, the 650ti needs a lot more power than the 7790 so getting along on a SFX350w may just be touch and go on a 650, possibly hitting 85% up capacity usage under load along with the voltage ripples you will be hard pressed on that capacity, while a 7790 will do it easily.

Toms skewered a result based on performance only at the budget spectrum.

The article on the 7990 based the premise of a decision on Power and Accoustics, foregoing performance, now this card whatever way we slice it is only for enthusiast builiders looking for $2000-5000+ builds, likely have custom cooling or 900D's with sound dampning, probably don't worry as the card will be watercooled. Simply put no mainstream builder in the 500-1500 bracket will ever use this card so its obselete to ever consider it. The enthusiast cares only for performance at all costs, accoustics and power is irrelevant as most are rocking 1.5kw supplies and are nowhere near capacity. On performance the 7990 backs its specs with its metrics which are industry leading and for that is the best dual graphics card and best graphics card period. They then came up with this other standard they termed "Feeling" their feelings and some randoms "feelings or perceptions". When you start having feelings and perceptions as grounds of justification you often opening upto subjectiveness and that is never good in any review. Chris and Don or Igor whomever it was are know Intel/Nvidia users so the sentiment lends to brand bias. The review is thus not based on science or metrics its formed on subjectivity and its just not good enough.

if i hadn't known about you being biased towards amd, i woulda suspected you of championing intel's desktop core i3, i5 cpus against amd's desktop fx cpus and llano and trinity apus. ;-D

Not a i3 and not integrated graphics

I don't use FX, haven't in a while and Trinity is the first time I used a APU and the results are good enough to build a ITX system which doubles as a HTPC or Media center and gamer all kept with my Home theatre setup. Since I don't need a GPU I offset power and can run a 300W PSU with ease, Since I heard SeaSonic will have a 200 and 250w Platinum Fanless coming soon and Richland out in June, I am happier than a pig in mud.

Well I do have a 32 Thread dual Xeon cruncher which has 580's and a 7970 for the best of both worlds. I am not sure about the others, I have said the i3 is only tangible when you factor in a discrete graphics solution but i3 vs APU in HTPC/Gaming lite trim the APU consumes more power but outputs far more at a lower cost. I factored this in the local costings and Intel ITX boards are significantly higher than the A85 ITX from Asrock. I also have a NAS running on these boards due to the 7 SATA ports which is a very neat offering. Don't shoot the messenger because I have found that a "lowly" APU is capable of delivering my needs and lets me pwn noobs in BF3 at the same time

PS: I tweaked the CFG file and dropped settings to complete lowest possible, at 800x600 BF3 is on steriods, get accused of aimbotting and hacking but its just a case of serious FPS and low latency, it may look bad but at the end of the day winning and killing plenty is the object of the game