There is no chance for AMD to do that much in a single generation, i expect around 40% over Piledriver and close to Haswell.
Well, it's created from scratch. And 8350 even didn't define what is "within 5% of Skylake". IPC? Performance of ST? Overall performance? Performance/watt?
etayorius :
What would they do with all the Piledriver CPUs they still got if they present something 60% faster?
That is what really bothers me about it. They should have clarified early to avoid the mess entirely, but that is not the nvidia way. They just hoped that it would be a while before the 4 GB frame buffer flaw was actually shown for what it was.
You don't have any proof of this, true?
8350rocks :
The other thing I dislike about the solution is the design requires the GPU to be unable to read or write from both sections of memory simultaneously. Sure, it would have been more complex to implement a second memory controller for the second chunk of memory, but now you end up with a gpu that thinks it has 4 GB, but really only has fast access to 3.5 GB
It resembles certain CPUs that think that have 8 independent cores, but really only have fast access to half the cores whereas access to the other half of the cores has a 20% performance penalty. The reason was that implementing a second decoder and a smarter fetch for those cores was "more complex".
Moreover, the engineers that build that CPU didn't claim that have 8 cores, but they used the term "8 clusters". However, the folks of marketing department 'simplified' the specs and said everyone that the CPU has 8 cores. And this was the starting point of unending discussions on forums. You can see still find people in forums discussing if those CPUs have 8 real cores or 8 fake cores or what.
The analogy with the Nvidia case is amazing, isn't?
I cannot say that there is any proof...
As for your analogy...it is a poor analogy. You can still access the second core of a module on the BD/PD/SR line of CPUs, on the NVidia GPU, you literally cannot access the 512 MB portion if you are accessing the 3.5 GB side. It is an either or, not both with one at a penalty.
Hence, your ignorance proves my point. Even if you could do both at a penalty to one side, it would still be better than having it as a binary type of scenario.
yes you can, otherwise there wouldn't be benchmarks showing the card using more than 3.5 gb of ram, but in many benchmarks it does use the whole 4gb,besides it's already explained in much detail how the memory allocation works hence his analogy stands
Besides have you forgotten how amd marketed their bulldozer as the i7 killer, that never happened, now your sources are saying it will go head to head with skylake, which i imagine come from amd, perhaps on igpu it will and some specific workload, but they have over hyped their parts far to much in the past for me to trust anything that comes from amd be it engineers or their pr team
I do hope that zen will at least match sandy performance in terms of ipc and clocks good i will be going back to amd again, but there comes a time when one has to realize that dreaming alone won't help and it doesn't help
as it stands now amd hadn't had a HEDT part since phenom II x6
If this was possible they would have done this instead of Crapdozer.
Did you read story of Bulldozer?
http://semiaccurate.com/2011/10/17/bulldozer-doesnt-have-just-a-single-problem/
Core 2 Duo was released in 2006, K10 wasn't even designed to compete with that. Then they failed with BD and stuck with it for 5 years. To be honest they tried to defeat Intel only two times: with Bulldozer and K7. 50% succes rate.
That is what really bothers me about it. They should have clarified early to avoid the mess entirely, but that is not the nvidia way. They just hoped that it would be a while before the 4 GB frame buffer flaw was actually shown for what it was.
You don't have any proof of this, true?
8350rocks :
The other thing I dislike about the solution is the design requires the GPU to be unable to read or write from both sections of memory simultaneously. Sure, it would have been more complex to implement a second memory controller for the second chunk of memory, but now you end up with a gpu that thinks it has 4 GB, but really only has fast access to 3.5 GB
It resembles certain CPUs that think that have 8 independent cores, but really only have fast access to half the cores whereas access to the other half of the cores has a 20% performance penalty. The reason was that implementing a second decoder and a smarter fetch for those cores was "more complex".
Moreover, the engineers that build that CPU didn't claim that have 8 cores, but they used the term "8 clusters". However, the folks of marketing department 'simplified' the specs and said everyone that the CPU has 8 cores. And this was the starting point of unending discussions on forums. You can see still find people in forums discussing if those CPUs have 8 real cores or 8 fake cores or what.
The analogy with the Nvidia case is amazing, isn't?
I cannot say that there is any proof...
As for your analogy...it is a poor analogy. You can still access the second core of a module on the BD/PD/SR line of CPUs, on the NVidia GPU, you literally cannot access the 512 MB portion if you are accessing the 3.5 GB side. It is an either or, not both with one at a penalty.
Hence, your ignorance proves my point. Even if you could do both at a penalty to one side, it would still be better than having it as a binary type of scenario.
Analogies are incomplete, otherwise wouldn't be analogies but equivalences. I tried to put AMD in the best possible shape by considering a case with penalty, but we can consider a poor scenario: what you call "binary type".
On fused mode the FPU on BD/PD/SR modules is only accessible to one of the cores in the module at a given clock beat. Either one core or the other can access the fused FPU but not both at same time.
Once said that, I will add that your evaluation of Nvidia problems is wrong. It is false that the card cannot access both polls of memory at the same time:
I wanted to clarify a point on the GTX 970's ability to access both the 3.5GB and 0.5GB pools of data at the same. Despite some other outlets reporting that the GPU cannot do that, Alben confirmed to me that because the L2 has multiple request busses, the 7th L2 can indeed access both memories that are attached to it at the same time.
As the author correctly notes, Nvidia did a simple mistake with the specs but the card perform exactly as expected:
But from a purely practical standpoint, this doesn’t really change anything for the end user. The GeForce GTX 970 remains one of the best graphics card buys on the market. It performs the same way it did at launch — which is really good. As such, we will continue to recommend it until there is a better-performing option for the price.
We can empathize with buyers who feel betrayed, though. Nvidia definitely has some mind-share to earn back. But to us the price/performance ratio trumps everything else, and that is no different today than it has been since the GeForce GTX 970 was released.
Juan went from an AMD advocate to disillusioned and can't seem to get past the anger phase.
etayorius :
8350rocks :
szatkus :
8350rocks :
My source said that Zen would be..."within 5% of skylake across the board"
Do they know real performance of Skylake or is it just a prediction?
I would imagine it would be a very educated guess...
There is no chance for AMD to do that much in a single generation, i expect around 40% over Piledriver and maybe close to Haswell in the best scenario.
What would they do with all the Piledriver CPUs they still got if they present something 60% faster?
If this was possible they would have done this instead of Crapdozer.
Also, how can they know the performance of Skylake? there is no chance for AMD even matching Haswell, and if they do i would probably buy one but i can`t wait till 2016 to release another miracle... ill just grab the i5 4690k or wait a little more till Skylake arrives and grab the cheapest i5.
Piledriver is 3 generations old by the time Zen comes. The are effectively bargain bin parts even today. If you think of a 20% increase per gen, 1.20^3 = 1.73, the chips would be at least 70% better. Of course a lot of the gains will be on the energy efficiency side but the performance should still be much better than piledriver which is ancient even now.
I also know that they will not apply the same standard to their own predictions. For instance the people from the second camp that said that "Zen will be faster than Skylake" will try to negate he said that. This kind of stuff happened before.
That is not what was said...let us be fair now Juan.
My source said that Zen would be..."within 5% of skylake across the board"
I didn't mention any name. But you identified yourself because you said that the goal was "better than skylake" and latter added that even if they fall short of predictions they will be "on par". I repeat some few replies you received then:
@8350rocks
You most understand why people aren't jumping right onboard on the "AMDs next x86 arch will be better than skylake". We have little to no information on the architecture itself.
I have stated, that I doubt AMD's next x86 arch will reach skylakes performance (first, we don't even know skylakes performance yet, we are only estimating based on their earlier architectures, and some of the things it features (which we know for certain, like AVX512)).
Don't hype the hypetrain to much without anything backing it up.
I'm not saying I'm not excited about their upcoming x86 arch, I'm just waiting for some more information, to properly place and estimation on it's performance.
In short, this idea of AMD's 2016 x86 core outperforming Skylake on performance doesn't make sense: neither technologically, nor economically, nor strategically.
I agree, there is no chance AMD cant even match Skylake, even less outperform it with their next Arch... BUT, AMD does not have to beat Intel in the IPC game, it only needs to be competitive and offer better prices and AMD would be back at the game.
Skylake will probably be a good 15-25% faster than what ever AMD Throws with Keller, but hey! 25% behind is much better than being 60% behind, just as they are now.
The Bulldozer is sh1t line was born because Bulldozer was sh1t. I can't wait for AMD to ditch it and focus all their resources on post BD lines.
To expect them to reach Intel current gen levels is wishful thinking. For them to reach parity with Skylake (simultaneously) from their current position would require a miracle that would leave Jesus scratching his head, wondering how they'd managed it.
Juan went from an AMD advocate to disillusioned and can't seem to get past the anger phase.
etayorius :
8350rocks :
szatkus :
8350rocks :
My source said that Zen would be..."within 5% of skylake across the board"
Do they know real performance of Skylake or is it just a prediction?
I would imagine it would be a very educated guess...
There is no chance for AMD to do that much in a single generation, i expect around 40% over Piledriver and maybe close to Haswell in the best scenario.
What would they do with all the Piledriver CPUs they still got if they present something 60% faster?
If this was possible they would have done this instead of Crapdozer.
Also, how can they know the performance of Skylake? there is no chance for AMD even matching Haswell, and if they do i would probably buy one but i can`t wait till 2016 to release another miracle... ill just grab the i5 4690k or wait a little more till Skylake arrives and grab the cheapest i5.
Piledriver is 3 generations old by the time Zen comes. The are effectively bargain bin parts even today. If you think of a 20% increase per gen, 1.20^3 = 1.73, the chips would be at least 70% better. Of course a lot of the gains will be on the energy efficiency side but the performance should still be much better than piledriver which is ancient even now.
The 60% that he mentions is for ST IPC. Steamroller increased IPC by about 20% on MT by eliminating partially the module penalty, but ST performance did not change significantly. Excavator will bring 10--15% IPC gains.
This means that Excavator will be only ~15% faster than Piledriver on ST clock for clock. Zen would increase IPC by 45--50% over Excavator only to match Haswell IPC on certain workloads clock for clock.
My source said that Zen would be..."within 5% of skylake across the board"
Do they know real performance of Skylake or is it just a prediction?
Very simply, extract ~8-10% generation over generation performance increases, and you can figure out roughly where it will fall.
Hence, the very educated guesses.
As for why...it is a brand new uarch, with cues taken from the wealth of research AMD has done. I honestly hope the branch prediction logic and MC get some serious attention. That would certainly shore up some massive gains with solid revisions.
I would imagine it would be a very educated guess...
There is no chance for AMD to do that much in a single generation, i expect around 40% over Piledriver and maybe close to Haswell in the best scenario.
What would they do with all the Piledriver CPUs they still got if they present something 60% faster?
If this was possible they would have done this instead of Crapdozer.
Also, how can they know the performance of Skylake? there is no chance for AMD even matching Haswell, and if they do i would probably buy one but i can`t wait till 2016 to release another miracle... ill just grab the i5 4690k or wait a little more till Skylake arrives and grab the cheapest i5.
My source said that Zen would be..."within 5% of skylake across the board"
Do they know real performance of Skylake or is it just a prediction?
Very simply, extract ~8-10% generation over generation performance increases, and you can figure out roughly where it will fall.
Hence, the very educated guesses.
As for why...it is a brand new uarch, with cues taken from the wealth of research AMD has done. I honestly hope the branch prediction logic and MC get some serious attention. That would certainly shore up some massive gains with solid revisions.
I would imagine it would be a very educated guess...
So in fact they are targeting at 5-15% above Broadwell? Amusing
My source said that Zen would be..."within 5% of skylake across the board"
Do they know real performance of Skylake or is it just a prediction?
I would imagine it would be a very educated guess...
We are all hoping for that including Amd engineers but that would have took a lot of resources to do. Its possible granted. But we are talking about a 50% boost in IPC if so over piledriver-Zen.
If i had more money i'd probably bet that they won't get anymore then 20-30% boost on average in IPC from piledriver to zen. Some rumors are claiming its just going to be a 8 core which is actually fantastic news if so that means they might try for more single core performance. Like others are saying i doubt having a 256bit FPU per core will really hurt them much since the market for anything else is pretty small.
Zen will beat skylake in some things and loses in others. I doubt skylake will be 8 cores in mainstream so Zen will win multithreading pretty hands down.
So^ we are now claiming Zen will offer similar integer performance in single core output using SSE? Since it takes twice the cores now from piledriver to even come close to a 4 core Intel CPU guessing the new claim is twice the performance per clock in integer base workloads compared to piledriver?
Like i stated i expect a 20-30% boost in IPC overall and lower clock speeds to. Scaling per core will probably be similar to phenom which was great. With skylake being 50% better in IPC compared to piledriver(rough estimate probably a bit more ahead). I honestly don't see anything winning from a 8 core Zen CPU except when all 6-8 cores are used heavily, which is so rare its not even funny. When 6-8 cores are being used its usually still being bottle necked by 1 core.
However i could be wrong i just don't think gains of 50% are possible from one generation with very limited money to spend on a complicated architecture. Even the rumors don't hold to support this with it being rumored as being a small core.
I do however expect it to be way more efficient in performance per watt maybe even 50% better at performance per watt then piledriver.
With skylake being 50% better in IPC compared to piledriver(rough estimate probably a bit more ahead)
I think Haswell is already a good 60% ahead of Piledriver, i expect Skylake to be at least 10% faster than Haswell, no one knows for sure how much performance in regards to IPC compared to Haswell Skylake will have, so AMD saying they expect to be within 5% of Haswell seems extremely optimist... No, really.
But i been saying this for a while, AMD does not need to match Skylake or even Haswell, just add at least a 30% IPC as final performance over Piledriver and cram as many Cores as they can and sell it cheaper than Intel, that would be awesome.. sadly i cannot wait for AMD miracles so my next Build will be an i5.
With skylake being 50% better in IPC compared to piledriver(rough estimate probably a bit more ahead)
I think Haswell is already a good 60% ahead of Piledriver, i expect Skylake to be at least 10% faster than Haswell, no one knows for sure how much performance in regards to IPC compared to Haswell Skylake will have, so AMD saying they expect to be within 5% of Haswell seems extremely optimist... No, really.
But i been saying this for a while, AMD does not need to match Skylake or even Haswell, just add at least a 30% IPC as final performance over Piledriver and cram as many Cores as they can and sell it cheaper than Intel, that would be awesome.. sadly i cannot wait for AMD miracles so my next Build will be an i5.
Sounds like me
Anyways yeah i agree it can be upto 60% stronger but i used quite a lot of benchmarks since i have a lot of free time and its generally averaged out to about 45%. Amd will probably be 30% weaker in IPC when zen comes around with a 3.5ghz clock speed. Intel isn't known to make parts that are clocked higher then 3.5Ghz often anyways.
Why i think they will see a major 20-30% boost is over Jim keller improving cache speeds which is one area piledriver looks embarrassing in. palladin9479 gave some good reasons why Amd is behind and it really comes down to their branch predictor something that is hard to fix and something i bet hasn't been improved a lot. But cache improvements and other small fixes should increase IPC by 20-30% on average.
Then we should be done with this CMT nonsense. I still think SMT isn't happening since it doesn't make since on a small core. Its all repeated nonsense and we will hear more of it the days to come.
Then we might even get trolls after the product comes out and they will cherry pick results to confirm their bias, or worse they will call the sites hosting the benchmarks liers. I used to do this myself until i started benchmarking the products myself over me having the resources to do so.
That is what really bothers me about it. They should have clarified early to avoid the mess entirely, but that is not the nvidia way. They just hoped that it would be a while before the 4 GB frame buffer flaw was actually shown for what it was.
You don't have any proof of this, true?
8350rocks :
The other thing I dislike about the solution is the design requires the GPU to be unable to read or write from both sections of memory simultaneously. Sure, it would have been more complex to implement a second memory controller for the second chunk of memory, but now you end up with a gpu that thinks it has 4 GB, but really only has fast access to 3.5 GB
It resembles certain CPUs that think that have 8 independent cores, but really only have fast access to half the cores whereas access to the other half of the cores has a 20% performance penalty. The reason was that implementing a second decoder and a smarter fetch for those cores was "more complex".
Moreover, the engineers that build that CPU didn't claim that have 8 cores, but they used the term "8 clusters". However, the folks of marketing department 'simplified' the specs and said everyone that the CPU has 8 cores. And this was the starting point of unending discussions on forums. You can see still find people in forums discussing if those CPUs have 8 real cores or 8 fake cores or what.
The analogy with the Nvidia case is amazing, isn't?
I cannot say that there is any proof...
As for your analogy...it is a poor analogy. You can still access the second core of a module on the BD/PD/SR line of CPUs, on the NVidia GPU, you literally cannot access the 512 MB portion if you are accessing the 3.5 GB side. It is an either or, not both with one at a penalty.
Hence, your ignorance proves my point. Even if you could do both at a penalty to one side, it would still be better than having it as a binary type of scenario.
yes you can, otherwise there wouldn't be benchmarks showing the card using more than 3.5 gb of ram, but in many benchmarks it does use the whole 4gb,besides it's already explained in much detail how the memory allocation works hence his analogy stands
Besides have you forgotten how amd marketed their bulldozer as the i7 killer, that never happened, now your sources are saying it will go head to head with skylake, which i imagine come from amd, perhaps on igpu it will and some specific workload, but they have over hyped their parts far to much in the past for me to trust anything that comes from amd be it engineers or their pr team
I do hope that zen will at least match sandy performance in terms of ipc and clocks good i will be going back to amd again, but there comes a time when one has to realize that dreaming alone won't help and it doesn't help
as it stands now amd hadn't had a HEDT part since phenom II x6
I can confirm that "the faster than skylake" hype/nonsense is not coming from the PR folks but from one CPU engineer of the K12/Zen project.
etayorius :
jdwii :
With skylake being 50% better in IPC compared to piledriver(rough estimate probably a bit more ahead)
I think Haswell is already a good 60% ahead of Piledriver, i expect Skylake to be at least 10% faster than Haswell, no one knows for sure how much performance in regards to IPC compared to Haswell Skylake will have, so AMD saying they expect to be within 5% of Haswell seems extremely optimist... No, really.
But i been saying this for a while, AMD does not need to match Skylake or even Haswell, just add at least a 30% IPC as final performance over Piledriver and cram as many Cores as they can and sell it cheaper than Intel, that would be awesome.. sadly i cannot wait for AMD miracles so my next Build will be an i5.
Yes, Haswell is about 60% ahead of Piledriver on integer single-thread. Leaked benchmarks suggest Skylake will be 10--15% ahead of Haswell on integer single thread. Also Skylake is coming this year (Q3) whereas Zen is late 2016 early 2017.
Wondering what the Carrizo TDPs will be. You reckon we'll see 17-19W and 35W, or will there be a wider range / greater variety?
Also, does the actual power usage ever rise above the TDP when using "turbo"? If so, approximately how much could it (a guesstimate)?
Personally I'm hyped about the idea of getting a nice 1080p freesync screen with a flagship Carrizo APU in a laptop. Your estimates for gaming performance increase over fx-7600p?
I go back to the days of Athlon XPs
I remember OCing the heck out of Bartons and Thunderbirds
I also do try to have an Intel rig also
Last year I had a 1150 socket Xeon
but I made a choice to sell that one and keep my PhII x 6 1100T BE rig
however as of now the way things are if I decided I needed a platform change I would go 1150 or 2011
Pretty sad
I do not even own an AMD GPU anymore
I still have a loyalty to AMD but my money goes to the best performance for the dollar
I go back to the days of Athlon XPs
I remember OCing the heck out of Bartons and Thunderbirds
I also do try to have an Intel rig also
Last year I had a 1150 socket Xeon
but I made a choice to sell that one and keep my PhII x 6 1100T BE rig
however as of now the way things are if I decided I needed a platform change I would go 1150 or 2011
Pretty sad
I do not even own an AMD GPU anymore
I still have a loyalty to AMD but my money goes to the best performance for the dollar
Somewhere out there is a developer considering utilizing HSA features for their software... maybe even a game... who knows, maybe 3 years from now there will be a game that runs better on an A10 (or future-APU equivalent) than an i7.
I can confirm that "the faster than skylake" hype/nonsense is not coming from the PR folks but from one CPU engineer of the K12/Zen project.
Then why comment on it at all really? It's common practice to ignore the extraneous high and low water marks.
Because one person said here the "faster than skylake". Several people reacted to that, and one (the one whom I replied) asked if the hype was coming from the PR dept. or from the engineers.
Reepca :
Wondering what the Carrizo TDPs will be. You reckon we'll see 17-19W and 35W, or will there be a wider range / greater variety?
Also, does the actual power usage ever rise above the TDP when using "turbo"? If so, approximately how much could it (a guesstimate)?
Personally I'm hyped about the idea of getting a nice 1080p freesync screen with a flagship Carrizo APU in a laptop. Your estimates for gaming performance increase over fx-7600p?
Carrizo TDP has a span of 15--35W.
The CPU will be about the same than FX-7600P even could be slower in some benchmarks. The GPU is expected to bring a good ~30% boost. Gaming will depend of if the game is CPU bound, GPU bound...
I go back to the days of Athlon XPs
I remember OCing the heck out of Bartons and Thunderbirds
I also do try to have an Intel rig also
Last year I had a 1150 socket Xeon
but I made a choice to sell that one and keep my PhII x 6 1100T BE rig
however as of now the way things are if I decided I needed a platform change I would go 1150 or 2011
Pretty sad
I do not even own an AMD GPU anymore
I still have a loyalty to AMD but my money goes to the best performance for the dollar
Best perf/$ and you do not own an AMD GPU???
Color me confused, AMD was only the best bang for your buck in every price range for the last 6 months, and only recently was unseated, and even then, that was only if you were spending more than $400-500 on a GPU.