Piledriver not coming until 2013 now?

[lifeblood]

I believe you will find that DX-11 games (Today and in the future) will need more GPU performance than CPU IPC

I think you picked the only 3 benchmarks were Bulldozer does well. However, your point is somewhat valid, in GPU bound games or those that benefit from lots of cores like Civ V (full render), AMD CPUs are competitive. Still, if you look at all the games, it was only in one or two where AMD beat the competition. In the rest of games it was evenly split between equaling Intel CPUs and being beat by them.

Bulldozer, as I am sure Piledriver will be, is a great server CPU. As a gaming CPU it is easily beaten by equivalently priced Intel CPUs.

 

[A5]

but what i am saying is that since 2006 cpus are stagnant...we only see some 10% to 25% increase in each generation, yet the TDP is decreasing

while gpus are stagnant since 2009, but the limit is more about the TDP and bandwidth, than nvidia\amd desire



you know, with some bios and kernel updates, and a motherboard that support IOMMU, the 7970 could actually boot up linux

If you consider a 10+% performance increase every 2 years while also reducing TDP to be "stagnant", you obviously have no clue about CPU architecture or design.

If you're going to be reductive, you could at least go all they way and say Sandy Bridge is just a "tweaked" Pentium Pro, because it is just as accurate as what you are saying right now.

 

[piesquared]

LOL I have said that for years. That WE need seperate Intel/AMD and NV/AMD CPU and GPU sectionns .

APU = AMD PROGRAMMING UNIT. Intel Isn't ever going to call their IGPU an APU not Ever. IPU yes APU not ever. I thought I was a fanboy . NO no I am not . YOU are a fanboy. I going to love what IB Igpu does to your smallish world.

lol My smallish world? ib's videolan playback unit has to compete with Trinity's discrete GPU and HSA. ib's VlanPU might be able to play a video of a game (minus a few stutters here and there), but it seems that's about it. As far as the APU moniker, I wouldn't be too sure that intel won't copy that at some point.

 

[Olikan]

If you consider a 10+% performance increase every 2 years while also reducing TDP to be "stagnant", you obviously have no clue about CPU architecture or design.

If you're going to be reductive, you could at least go all they way and say Sandy Bridge is just a "tweaked" Pentium Pro, because it is just as accurate as what you are saying right now.

sure, i am not a cpu expert, but you missed my point.

my point is...since we have a gpu fight with nvidia and amd, the gpu market does have bigger jumps in performance....compared at cpu market

you think that ivy would have TPD of 77 if bulldozer beated sandy?

 

[RavenSEAL]

I'm pretty sure my X6 can hang on decently for at least 2 more years just like my e8500 did. Not worried at all!

And IF Vishera does collapse, I should haven't any problem dishing out the money for a mid range intel build.

 

[thilanliyan]

And IF Vishera does collapse, I should haven't any problem dishing out the money for a mid range intel build.

Hehe yeah I'm giving AMD ONE more chance...if Piledriver is not much better than Bulldozer I'm gonna go back to Intel for the first time since I got my Core2 Duo E6400 in 2006.

 

[Arzachel]

To the guy who got latched on to "CMT scaling". That's absolutely not what I am talking about. I am talking about the penalty for that system, not how it scales.

We know AMD can do better as far as IPC, their last gen was better, at a similar transistor per core cost. I'm saying, when they sandwich them together, the compromises they have to make, makes for crappy single-threaded capability with their current implementation.

Do we know the transistor count for a module after they changed the total to 1.2b?

 

[exar333]

I'll be surprised to see a a Piledriver core until the earliest October of 2012. Remember the Bulldozer debacle? My suggestion to CEO Rory? Have the GPU engineers that gave us the 7970/7950 at AMD work on the Piledriver core. At least there will a legitimate contender to Intel.

And by contender you mean the PD IPC may reach C2D levels?

 

[exar333]

Hehe yeah I'm giving AMD ONE more chance...if Piledriver is not much better than Bulldozer I'm gonna go back to Intel for the first time since I got my Core2 Duo E6400 in 2006.

10% better will not cut it; it will likely be (worst case) similar to X6 and best case, decently faster. This is unless AMD really pulls a rabbit out of their hat...

 

[Rvenger]

10% better will not cut it; it will likely be (worst case) similar to X6 and best case, decently faster. This is unless AMD really pulls a rabbit out of their hat...

**Hopes for the rabbit to come out of the hat**

 

[Sweepr]

And by contender you mean the PD IPC may reach C2D levels?

Too optimistic.
3-5% better than Bulldy would still put it below Phenom II levels.

 

[blckgrffn]

10% better will not cut it; it will likely be (worst case) similar to X6 and best case, decently faster. This is unless AMD really pulls a rabbit out of their hat...

Obviously AMD has sacrificed IPC to go for parallelism and clock speed, this is a conscious decision and they made architectural trade offs that we should probably all hope pay off. To say that BD is a failure until its single threaded throughput is at Thuban (let alone SB/IVB) levels is to set an impossible goal for this architecture on this process. PD is likely to still be at least 10% lower in IPC than Thuban.

Will AMD be the best CPU for games? Probably (definitely?) not until we see another re-arch or games that really depend on many threads become the norm. For those folks whose computer does other things first and games second it is likely to be a reasonable choice for gaming still.

Will it be good for other things? Yes. If they can get power consumption under load more reasonable and increase clock speeds a bit it will be a reasonable choice at certain competitive price points and likely the go to CPU for an even smaller number of targeted applications. This is in line with what AMD has said their mission is - to win certain segments. It is sad to see BD be so narrowly focused due to its low performance/watt.

I was sad to see that there is not a 5 Module part scheduled for the server or desktop. The leaked material at CPU-World mentioned them.

They also indicated that L1 cache was back up to 64 bytes, which is a good thing, it seems like any software that has been optimized for AMD or recent Intel is going to be tuned for that cache size. I am hoping that bit stays accurate.

 

[Ferzerp]

blckgriffn, that argument would make sense if the transistor count per core were substantially lower than their previous offering, but it isn't. It is nearly exactly the same (150M transistors/"core"), but the new cores perform worse than the old.

 

[blckgrffn]

blckgriffn, that argument would make sense if the transistor count per core were substantially lower than their previous offering, but it isn't. It is nearly exactly the same (150M transistors/"core"), but the new cores perform worse than the old.

I said "hope the trade offs would pay off." I mentioned sad a lot in my post

I guess they didn't (payoff). At full tilt, BD matches Thuban in the amount of work each can do and we have yet to see really optimized software or AVX be enabled in a large number of applications, that is the only upside I can see coming for BD. In some distributed computing projects AVX could bring a BD core (thread?) inline with a Thuban core clock-for-clock. Maybe. Then you'd have a couple bonus cores in a very similar power envelope. Win?

What would BD look like on a better process? Who knows, I wish it were on a nice and mature 28 nm or similar.

 

[exar333]

Obviously AMD has sacrificed IPC to go for parallelism and clock speed, this is a conscious decision and they made architectural trade offs that we should probably all hope pay off. To say that BD is a failure until its single threaded throughput is at Thuban (let alone SB/IVB) levels is to set an impossible goal for this architecture on this process. PD is likely to still be at least 10% lower in IPC than Thuban.

Will AMD be the best CPU for games? Probably (definitely?) not until we see another re-arch or games that really depend on many threads become the norm. For those folks whose computer does other things first and games second it is likely to be a reasonable choice for gaming still.

Will it be good for other things? Yes. If they can get power consumption under load more reasonable and increase clock speeds a bit it will be a reasonable choice at certain competitive price points and likely the go to CPU for an even smaller number of targeted applications. This is in line with what AMD has said their mission is - to win certain segments. It is sad to see BD be so narrowly focused due to its low performance/watt.

I was sad to see that there is not a 5 Module part scheduled for the server or desktop. The leaked material at CPU-World mentioned them.

They also indicated that L1 cache was back up to 64 bytes, which is a good thing, it seems like any software that has been optimized for AMD or recent Intel is going to be tuned for that cache size. I am hoping that bit stays accurate.

Thats a fair assessment. If BD was miserly in regards to power, I really don't think it would have been so panned (and rightfully so). If PD is slower, but on-par efficiency-wise then it's a definite valid option for many. The fact that BD is both slow and inefficient totally kills any potential merit it has right now. The only 'plus' to BD is when it's not being used (idle power). How is that for a good chuckle?

 

[blckgrffn]

Thats a fair assessment. If BD was miserly in regards to power, I really don't think it would have been so panned (and rightfully so). If PD is slower, but on-par efficiency-wise then it's a definite valid option for many. The fact that BD is both slow and inefficient totally kills any potential merit it has right now. The only 'plus' to BD is when it's not being used (idle power). How is that for a good chuckle?

Hah. Indeed.

If you have a nice AM3+ capable setup you don't use much, that BD CPU can save you some mad power at idle over your current Phenom 2! Step right up and buy one today...

 

[Ferzerp]

Trade-off implies some benefit. They've brought a product to market that is less capable than their previous product, and somehow expect people to buy them, and they've been relying on some pretty shady marketing to try to accomplish that.

 

[dma0991]

blckgriffn, that argument would make sense if the transistor count per core were substantially lower than their previous offering, but it isn't. It is nearly exactly the same (150M transistors/"core"), but the new cores perform worse than the old.

Even if they are about the same (1.2/8 & 0.9/6), Thuban had at most 3MB of L2 and 6MB of L3, 512kB/core and 6MB shared. BD has 8MB of L2 and 8MB of L3, 2MB/core and 8MB shared. If it is true that a single BD core has about the same transistor count as a Thuban, then more of the budget is allocated to caches rather than the actual 'core' itself.

 

[blckgrffn]

Trade-off implies some benefit. They've brought a product to market that is less capable than their previous product, and somehow expect people to buy them, and they've been relying on some pretty shady marketing to try to accomplish that.

I think they were a victim of a gamble that didn't pay off as well as they thought it would (BD), sacrifices they no doubt had to make to get into their transistor budget, and a process technology that is the exact opposite of what they needed to cover their gamble.

A four module BD right now is as good (or marginally better) as a Thuban in highly threaded applications. Had the process allowed for 10% lower power consumption and 10% higher clocks it would have been fine.

We are very used to Intel consistently delivering power consumption and performance increases at each step that AMD having a misstep seems very horrible indeed.

Your statement could very easily describe Intel back in the day going from P3 @ ~1.2 ghz to P4 @ 1.4 ghz (yuck!).

 

[Vesku]

I'd expect Piledriver AM3+ and server chips late this year. What the new roadmap says to me is that OEMs want AMD to make another product cycle of drop in upgrade chips. This is my guess, AMD has to work within what their partners are willing to invest time and money-wise in every 1-2 year cycle.

 

[AtenRa]

I think you picked the only 3 benchmarks were Bulldozer does well. However, your point is somewhat valid, in GPU bound games or those that benefit from lots of cores like Civ V (full render), AMD CPUs are competitive. Still, if you look at all the games, it was only in one or two where AMD beat the competition. In the rest of games it was evenly split between equaling Intel CPUs and being beat by them.

Bulldozer, as I am sure Piledriver will be, is a great server CPU. As a gaming CPU it is easily beaten by equivalently priced Intel CPUs.

First of all, i haven't picked those benchmarks, anand did

Secondly, show me a DX-11 game (at 1080p and above) with filters enabled that is way faster with Intel Core i5/7 2500K/2600K than FX8120/1850.

You wont find any with a difference larger than 3-5 fps

I have never said that BD is the fastest or better CPU in the world, but for DX-11 Gaming (1080p and above) it is pretty even with Intel at high resolutions.

Now if you like to have triple or quad CF/SLI that is another story

 

[Sweepr]

Secondly, show me a DX-11 game (at 1080p and above) with filters enabled that is way faster with Intel Core i5/7 2500K/2600K than FX8120/1850.

I have never said that BD is the fastest or better CPU in the world, but for DX-11 Gaming (1080p and above) it is pretty even with Intel at high resolutions.

Now if you like to have triple or quad CF/SLI that is another story

So what you're saying is that you need to be GPU-bound at 1080p+ with a single graphics card on the latest DX11 games with filters enabled to make FX come close to Sandy Bridge?

 

[lifeblood]

First of all, i haven't picked those benchmarks, anand did

The three games you picked were picked by Anand precisely in order to best show the improvements to BD with the MS patches. They do not show games that are CPU bound or heavily single threaded.

Secondly, show me a DX-11 game (at 1080p and above) with filters enabled that is way faster with Intel Core i5/7 2500K/2600K than FX8120/1850.

I have never said that BD is the fastest or better CPU in the world, but for DX-11 Gaming (1080p and above) it is pretty even with Intel at high resolutions.

Unfortunately, few current games are DX11. Most are DX9 console ports. However, I think you're saying that as DX11 games become more prevalent, BD will shine more and more. When the next-gen consoles come out supporting DX11 then I absolutely agree. AMD designed BD with the future in mind and if their crystal ball is accurate, the BD architecture will redeem itself.

Having said that, I live in the here and now, and its the here and now I use a PC. And for the majority of current games, BD is uncompetitive.

 

[Rvenger]

First of all, i haven't picked those benchmarks, anand did

Secondly, show me a DX-11 game (at 1080p and above) with filters enabled that is way faster with Intel Core i5/7 2500K/2600K than FX8120/1850.

You wont find any with a difference larger than 3-5 fps

I have never said that BD is the fastest or better CPU in the world, but for DX-11 Gaming (1080p and above) it is pretty even with Intel at high resolutions.

Now if you like to have triple or quad CF/SLI that is another story

Yes high resolutions where it could rely on multithreaded situations or GPU. BTW I read your blog in your sig and I have to say you made some good points about the single threaded performance and its future. If Piledriver can manage to kick up just single threaded performance by at least 15% across the board, it won't be a bad chip after all.

Another words, if Piledriver can get close to Sandybridge in single threaded performance and still manage the same level of multithreaded performance that Bulldozer has, I will buy one for sure

 

[exar333]

Yes high resolutions where it could rely on multithreaded situations or GPU. BTW I read your blog in your sig and I have to say you made some good points about the single threaded performance and its future. If Piledriver can manage to kick up just single threaded performance by at least 15% across the board, it won't be a bad chip after all.

Another words, if Piledriver can get close to Sandybridge in single threaded performance and still manage the same level of multithreaded performance that Bulldozer has, I will buy one for sure

I can honestly say there is no way PD will gain 30-40% IPC. You can quote me on that.