AMD chief say BD will offer only 35% not 50% more performance than previous gen

[yottabit]

Basically, Core 2 45nm is 11% faster than K10.5, Nehalem is 13% faster than Core 2, and Sandy Bridge is 11% faster than Nehalem. My math was a little off, so sorry about that. In the end, Intel has a 35% advantage in IPC, so I wasn't off in the overall numbers.

JFWIW, you know you can't just add percentages like that right?

 

[Lonbjerg]

JFWIW, you know you can't just add percentages like that right?

Yeah, doing it that way will get a smaller number than the actual (~39,2&#37 number of increased IPC performance.

 

[Riek]

JFWIW, you know you can't just add percentages like that right?

You also can't propagate ipc in general since it is dependant on the clockspeed also. So the difference in ipc between two cpu's clocked at 2GHz can already be different then comparing the ipc between two architectures running at 3GHz.

But with any discussion about such a variable you take approximations.

 

[grkM3]

Everything dose not look well for BD so far.You have to wonder if they had a killer they would leak out the numbers to slow down people going intel or they have a monster but AMD marketing is dumb.

I remember when intels sandys bridge chips reviews went up before launch and in that same 2 weeks the ceo of AMD stepped down.

It all dosnt look well for AMD with intel about to go 22nm,it sucks because now intel will price high like always and will delay ivy to make as much with sandy.

 

[Riek]

Everything dose not look well for BD so far.You have to wonder if they had a killer they would leak out the numbers to slow down people going intel or they have a monster but AMD marketing is dumb.

AMd has a better track record when they shut up about their design and performance than when they talk big. (this is true for their CPU as well as their GPU devision..)

Whe they talk big 2900XTX and phenom..
When they don't -> K7, K8, 4xxx, 5xxx, 6xxx

 

[grkM3]

AMd has a better track record when they shut up about their design and performance than when they talk big. (this is true for their CPU as well as their GPU devision..)

Whe they talk big 2900XTX and phenom..
When they don't -> K7, K8, 4xxx, 5xxx, 6xxx

yeah but I clearly remember athlon 64 was leaked way before launch and I remember how it picked up crazzy gains in 3d01,it was that chips that made a lot of us break 20k on air cooling.

they had a monster and leaked the numbers back then,they also were charging 900 for the top end FX until intel came out swinging.

Its sad I have my first A64 sitting right next to me in my desk and I can remember how excited I was reading up on the reviews about it.

I couldnt wait to buy one,now I dont get that same buzz with BD and it dosnt look very well.

 

[Riek]

yeah but I clearly remember athlon 64 was leaked way before launch and I remember how it picked up crazzy gains in 3d01,it was that chips that made a lot of us break 20k on air cooling.

they had a monster and leaked the numbers back then,they also were charging 900 for the top end FX until intel came out swinging.

Its sad I have my first A64 sitting right next to me in my desk and I can remember how excited I was reading up on the reviews about it.

I couldnt wait to buy one,now I dont get that same buzz with BD and it dosnt look very well.

The result were from opteron back then. Only leaks pre release leaks were from a 800MHz part.

AMD also delayed A64 due to process problems (SOI), they went to IBM for this. Sounds familiar?

Also AMd64 launched at a higher price you say..

For those of us with more pedestrian spending limits, the Athlon 64 3200+ looks like a great value. Yes, it costs over 400 bucks, but the stock Pentium 4 3.2GHz is selling for more than $600 right now.

The p4 which was competive costed 50% more. Now we are complaining that BD will suck because the leaked prcies are only 7$ above intel fastest mainstream cpu.. (2600)

 

[grkM3]

There were many leaks from the A64 that showed amazing gaming performance,there was even one guy that had a 3200 that made a video that showed him run 3d01 and broke 20k with ease while everyone were saying the leaks are fake.

AMD at that time was trying to build up against intel and in the beggining they were cheaper but once they took off and everyone wanted an A64 the prices went way up.I clrealy remember an FX chip on newegg for 1100 bucks since at the time it dominated anything from intel.

if BD had 40% gains over sandy you cpuld bet your life they would charge a premium for there chips

 

[Nemesis 1]

Wlee15 is correct . Although it has been said countless of times on this forum.. i'm still suprised to see the same people make the same mistake over and over again.

Ya Riek is correct and Its exactly what I said but in the manner I choose to say it. Clearly I stated .That I also did the other none HPC % . Yes people do keep making the same mistake and it is surprising . 35% gain in HPC on unknown HPC benchmark. That says it all . Than we have some dude going getting benchmarks of intels Hpc improvements in known benchmarks and were suppose to connect the dots . I think we all know AT has been handling AMD review previews with kid gloves . Which is fine. Nothing has changed other than the SB release in jan. clock for clock BD is 10% slower than nehalem as I told you in Dec.

 

[Riek]

Ya Riek is correct and Its exactly what I said but in the manner I choose to say it. Clearly I stated .That I also did the other none HPC % . Yes people do keep making the same mistake and its is surprising . 35% gain in HPC on unknown HPC benchmark. That says it all . Than we have some dude going getting benchmarks of intels Hpc improvements in known benchmarks and were suppose to connect the dots . I think we all know AT has been handling AMD review previews with kid gloves . Which is fine. Nothing has changed other than the SB release in jan. clock for clock BD is 10% slower than nehalem as I told you in Dec.

Well using non applications for comparing a possible gain in HPC is not usefull. I hope you agree there. So him saying the non HPC application speed increase with cores is irrelevant since we only have the statement about HPC.

The dot he wants to connect is: look at typical HPC applications, they do not scale lineairly with more cores. If a typical HPC application would scale great it would show in multiple HPC applications. Thus you cannot do 35% - 33% and say that there is no performance increase per core. So it points out that the statement isn't an indication of bad performance. (which ofcourse does not rule it out either).

Yeah, doing it that way will get a smaller number than the actual (~39,2%) number of increased IPC performance.

Yeah with his numbers. But Core2 isn't 11% faster than thuban cores so that levels things a bit.

 

[Black96ws6]

According to PC Mag, they won't be ready until Sep 26:
http://www.pcmag.com/article2/0,2817,2390410,00.asp

Advanced Micro Devices will release the first Opteron server processors to feature its new Bulldozer cores on Sept. 26, PCMag has learned. AMD's FX-series Bulldozer chips for high-end desktops are also set for release in either September or October, according to a report from DigiTimes.

It does say the desktop versions will be ready in September or October. The "or October" part is scary. Another month slip potentially?

 

[bryanW1995]

Sounds like Phenom I all over again.

At least there's no tlb error yet...knocks on wood

Yea, I bet they are locking out the overclocking options in Ivybridge as we speak. This sucks

Nah, they'll have it. But it will be a lot more than $20 extra for a K edition. Probably more like double the price...

Basically, Core 2 45nm is 11% faster than K10.5, Nehalem is 13% faster than Core 2, and Sandy Bridge is 11% faster than Nehalem. My math was a little off, so sorry about that. In the end, Intel has a 35% advantage in IPC, so I wasn't off in the overall numbers.

The most telling thing about these numbers is how much every new CPU architecture is an evolution, rather than a revolution, compared to the one it succeeds.

Also, Bulldozer doesn't need to have the same IPC as Sandy Bridge to succeed. If it has Nehalem IPC it'll be a lot faster in multi-threaded apps and somewhat slower in single-threaded, like the 990X.

You must be American like me. 100*1.11*1.13*1.11 = 139.2%, so a 39.2% increase.

 

[Vesku]

90% of the IPC of Nehalem would be fine with me. 8 of those cores running at about 4GHz, I don't think I'd have to juggle compute heavy tasks anymore just run them all at once.

 

[bryanW1995]

Everything dose not look well for BD so far.You have to wonder if they had a killer they would leak out the numbers to slow down people going intel or they have a monster but AMD marketing is dumb.

I remember when intels sandys bridge chips reviews went up before launch and in that same 2 weeks the ceo of AMD stepped down.

It all dosnt look well for AMD with intel about to go 22nm,it sucks because now intel will price high like always and will delay ivy to make as much with sandy.

No, the real problem is that even if BD is awesome, if they leak out the info too soon then intel will just come out with a 2700/2800/2900k lineup that is still faster. Even assuming BD has a 12% IPC improvement for single-threaded apps, the odds that it will clock up to 5.5 ghz + are lower than JHH throwing a birthday party with bunnies and rainbows/etc for his assistant during secretaries week.

According to PC Mag, they won't be ready until Sep 26:
http://www.pcmag.com/article2/0,2817,2390410,00.asp

It does say the desktop versions will be ready in September or October. The "or October" part is scary. Another month slip potentially?

At this point I'll just be happy if BD comes out before SB-E.

 

[Idontcare]

JFWIW, you know you can't just add percentages like that right?

Sure you can, you just have to make sure you clarify in fine print somewhere that the instruction set you used to accomplish said percentage propagation is not, itself, necessarily IEEE-754 compliant...instead we just take a page out of the american capitalism playbook and we call it GAAP compliant

Or even more abusive of the reader, we can call it non-GAAP compliant and then we can do whatever the hell kind of math pleases us in best serving our intentions!

 

[grkM3]

looks like intel is making sandy E right now...this was taken from someone that works at intel over at extremsystems.

 

[Nemesis 1]

Well using non applications for comparing a possible gain in HPC is not usefull. I hope you agree there. So him saying the non HPC application speed increase with cores is irrelevant since we only have the statement about HPC.

The dot he wants to connect is: look at typical HPC applications, they do not scale lineairly with more cores. If a typical HPC application would scale great it would show in multiple HPC applications. Thus you cannot do 35% - 33% and say that there is no performance increase per core. So it points out that the statement isn't an indication of bad performance. (which ofcourse does not rule it out either).



Yeah with his numbers. But Core2 isn't 11% faster than thuban cores so that levels things a bit.

Yes I know . Thats why I did the sissoft % . Intel showed a 48% with 50%core increase.
Amd increased core count by 33% . . Thats why I used the 25% increase in cores . As already stated 2 BD cores do not = 2 cores in scaling . So the very best your going to see from BD is 25% increase . and thats a huge maybe. Bob says 1 BD core is not = to one PHII core its less according to him . I personnelly have not seen a BD core in action but he has.

 

[classy]

In stock form if the top model processor is 35% faster than the 1100T across the board or in the areas where the Thuban is weak, it will be among the fastest desktop processors on the market. That is nothing to sneeze about. Now the product is late, but the architecture looks like it may have a lot of life and plenty of room for improvement. It has been a long time coming, its not going to be the King of the castle, but for foundation, it appears to be well suited.

 

[LOL_Wut_Axel]

What about Starcraft II benchmark? Or these benches?

2500k 3.3ghz is 40-50% faster than X1100T (6 core!) 3.3ghz.

The problem is that the 1100T is creating a GPU bottleneck, hence making the numbers look higher than normal in games. In normal CPU benchmarks Intel has a 35% advantage in IPC, as you can see. Games aren't really effective at comparing architectures, to be honest. It's best to stick to CPU only benchmarks to remove the GPU equation. The Phenom II X6 is faster than the Core i5 and i7 in DiRT 3. Does that mean it's a better CPU for gaming? No.

Also, remember that the Phenom II X6 was never really meant for gaming. It's more of a CPU for encoding, 3D rendering and content creation. In those situations, at the same clocks as the 2500 and 2400, it's the same speed. That's why I think it's a good, cost effective solution if you get something like an X6 1055T and do a mild OC.

 

[LOL_Wut_Axel]

At least there's no tlb error yet...knocks on wood



Nah, they'll have it. But it will be a lot more than $20 extra for a K edition. Probably more like double the price...



You must be American like me. 100*1.11*1.13*1.11 = 139.2%, so a 39.2% increase.

:hmm:

Aren't you supposed to add the numbers? 11+13+11=35...

 

[Nemesis 1]

Nah, they'll have it. But it will be a lot more than $20 extra for a K edition. Probably more like double the price...

Why would you say this ? Same was said about SB befor pricing was announced. Ever since penryn Intel has been pretty consistant on pricing . 2x more cost is just SCI-fi.

AMDs 8 core no matter what the performance assures us that the IB k will be priced the same . Which I have no problem calling a bit of a rip off considering its a die shrink that gives intel more cores per wafer. So it should be lower priced which likely will not happen .

 

[Nemesis 1]

The problem is that the 1100T is creating a GPU bottleneck, hence making the numbers look higher than normal in games. In normal CPU benchmarks Intel has a 35% advantage in IPC, as you can see. Games aren't really effective at comparing architectures, to be honest. It's best to stick to CPU only benchmarks to remove the GPU equation. The Phenom II X6 is faster than the Core i5 and i7 in DiRT 3. Does that mean it's a better CPU for gaming? No.

Also, remember that the Phenom II X6 was never really meant for gaming. It's more of a CPU for encoding, 3D rendering and content creation. In those situations, at the same clocks as the 2500 and 2400, it's the same speed. That's why I think it's a good, cost effective solution if you get something like an X6 1055T and do a mild OC.

WHAT? I got to go back and reread your llano post. What the hell. Be consistant. Your all over the place.

 

[Nemesis 1]

The problem is that the 1100T is creating a GPU bottleneck, hence making the numbers look higher than normal in games. In normal CPU benchmarks Intel has a 35% advantage in IPC, as you can see. Games aren't really effective at comparing architectures, to be honest. It's best to stick to CPU only benchmarks to remove the GPU equation. The Phenom II X6 is faster than the Core i5 and i7 in DiRT 3. Does that mean it's a better CPU for gaming? No.

Also, remember that the Phenom II X6 was never really meant for gaming. It's more of a CPU for encoding, 3D rendering and content creation. In those situations, at the same clocks as the 2500 and 2400, it's the same speed. That's why I think it's a good, cost effective solution if you get something like an X6 1055T and do a mild OC.

Its not cost effective at all . It lacks performance. There for in order to sell them amd has to cut pricies to the bone. Its not like AMD is pricing their cpus because they love the consumer . It because thats all there worth and I still find them overpriced when ya factor in cost of ownership for a 5 year period. One other thing when you guys figure % of improvement over BD your throwing out the high and keeping the low . You can't do that . If you throw out the high you have to throw out the low. I can make % based on the Benchies I choose. Thats not fair either.

% Thats the problem with hardware reviews. Were at the mercy of the reviewers. Good example is AMD prerelease benchies and videos of zacata /llano

 

[ed29a]

AMDs 8 core no matter what the performance assures us that the IB k will be priced the same . Which I have no problem calling a bit of a rip off considering its a die shrink that gives intel more cores per wafer. So it should be lower priced which likely will not happen .

You can confirm that because you magically travel to the future and see the prices of the 8 core FX and IB processors? Interesting.

 

[Phynaz]

:hmm:

Aren't you supposed to add the numbers? 11+13+11=35...

Nope.

If the first increase is 11%, then you have to add 13% onto the result of the 11% increase, so it's actually larger. Then you do the multiplication and then add the 11%, multiply, add the 15%.