AMD CPU speculation... and expert conjecture

[de5_Roy]

http://www.fudzilla.com/home/item/33587-amd-formally-unveils-kaveri

I apologise if this has already been posted.

most of the stuff was posted but, that article has the one slide i was looking for, the one with full test pc specs.
both amd a10 7850k and 6800k were tested using 2x 8GB ddr3 2133 ram, windows 8.1, 512GB ssd.
the 2 pass x264 hd benchmark was done with version 5.0.1, both apus locked at 3.7Ghz. that's the one amd used to claim 20% ipc improvement over richland (8.53fps vs 7.09fps). definitively confirms it's a clock-for-clock improvement (multicore).

amusingly, the intel test rig had hd4000 igpu on core i5 4670k with 2x 8GB ddr3 1600 ram for the pcmark, 3dmark, basemark benchmarks... they shoulda tested against iris pro, to ensure igpu performance advantage. still, they made sure to hide the igpu model number except R7 graphics. wth. just give us the model number...

 

[tracker45]

 

[gamerk316]

^^ Frankly, not impressed. Looking at the first chart, a combined CPU/GPU upgrade is only buying a few FPS at low settings. Granted, AMD still crushes intel in the iGPU deparment.

Second chart shows Intel still winning on CPU performance head to head by a decent margin. Not shocked in the least.

Third one shows Kaveri starts to reach max potential at 1866MHz RAM. Some gains going faster, but for most of the tests, <3 FPS gains from that point onward. More likely secondary improvements at that point, rather then the GPU still being starved. Clearly no issues with Kaveri getting starved on DDR3.

I think most here are/were expecting a bigger improvement then those charts indicate.

Also FYI: Source?

 

[de5_Roy]

^^ Frankly, not impressed.

Spoiler

Looking at the first chart, a combined CPU/GPU upgrade is only buying a few FPS at low settings. Granted, AMD still crushes intel in the iGPU deparment.

Second chart shows Intel still winning on CPU performance head to head by a decent margin. Not shocked in the least.

Third one shows Kaveri starts to reach max potential at 1866MHz RAM. Some gains going faster, but for most of the tests, <3 FPS gains from that point onward. More likely secondary improvements at that point, rather then the GPU still being starved. Clearly no issues with Kaveri getting starved on DDR3.

I think most here are/were expecting a bigger improvement then those charts indicate.

Also FYI: Source?

those low-res igpu gaming benches may as well be cpu benches. it seemed more like the games were cpu-bottlenecked (or memory/imc) on the apus and igpu bottlenecked on core i5. chances are there's performance headroom left for the igpu.

as for ddr3 1866, it has long been the perf-memory-speed sweet spot for igpu-gaming since llano. then again, i'll call cpu bottleneck in those benchmarks. none of the benches seen so far properly removed platform bottleneck(s). amd's official discreet gfx gaming benchmark had discreet gfx bottleneck, for example. that's why i'll wait till toms or at benches. come tuesday.

i'm still expecting to see some viable benchmark data. none of the currently available one seem overall credible. it's not that i don't believe them, i just keep finding (intentionally put?) limiting factors. the less of those, the better.

 

[tracker45]

http://www.pugetsystems.com/labs/articles/AMD-A10-7850K-Performance-Review-529/

 

[gamerk316]

Eek, I don't like this chart at all:

I know the clock is lower, but this would indicate IPC is basically flatline for the arch, and most of the improvements are in the frontend to improve multi-core performance, as seen:

Without adjusting for clockspeed, the performance changes for Geekbench:

Single Core tests:
Integer: -6.34%
Floating Point: -3.79%

Multi Core tests:
Integer: 4.04%
Floating Point: 1.25%

Also, GPU performance looks about GT640 level, which is a lot lower then some were expecting (7750, etc). CPU may be holding the GPU back though.

In terms of CPU performance, the A10-7850K is overall an improvement over the A10-6800K. Especially in PCMark 08, the A10-7850K was about 10% faster than the A10-6800K. Our GeekBench benchmark indicated lower single-core performance, but as more and more programs use multi-core technology, that is really not even much of an issue.

However, there are a few issues we have with the A-series line in general. First, while the gaming performance is better than Intel, it is only better if you are using integrated graphics. As soon as you add a discrete video card, an equivalently priced Intel-based system is going to give you overall better performance in most games. And while the gaming performance is very good for integrated graphics, if you consider yourself even a moderate gamer you likely would want to invest in a discrete video card.

Similarly, if you are mainly concerned about performance in programs that do not use the GPU, an Intel-based system is going to give you much better performance for your dollar. Finally, in terms of future upgrade potential, the A-series APUs hit a wall much, much sooner than their Intel equivalents. The A10-7850K is the fastest A-series APU currently available, yet the CPU performance is much lower than even the Intel Core i5-4440 which is only a mid-range Intel CPU.

So I'll stick to my 15% CPU performance, even if initial testing indicates I may be optimistic. Based on ONE sample, not impressed. Lets see if this result set holds up or not.

 

[logainofhades]

So basically kaveri will finally allow an APU to reach the phenom II x4 level of performance.

noob take a hike ... there is no need to troll here.

Next time you won't get a warning ... you will be absent for a week.

I actually see nothing wrong with this statement. Bulldozer was worse than PhII. Piledriver/Vishera was much improved, but the lack of L3 cache hurts Trinity and Richland to the point that a PhII 965 is actually faster than a 750k. Clock for clock PhII is superior. No offense, but for this statement from noob at least, your warnings of a ban are unwarranted. Frankly, imo, his assumption isn't really that far fetched.

 

[etayorius]

So basically kaveri will finally allow an APU to reach the phenom II x4 level of performance.

noob take a hike ... there is no need to troll here.

Next time you won't get a warning ... you will be absent for a week.

I actually see nothing wrong with this statement. Bulldozer was worse than PhII. Piledriver/Vishera was much improved, but the lack of L3 cache hurts Trinity and Richland to the point that a PhII 965 is actually faster than a 750k. Clock for clock PhII is superior. No offense, but for this statement from noob at least, your warnings of a ban are unwarranted. Frankly, imo, his assumption isn't really that far fetched.

Phenom II is legendary, but their days are long gone... all i can see in the benchmarks you posted is that the FX4350 beats the 965BE any time, both at stock.

Both CPUs OCed, the FX is still ahead... don´t compare clock for clock, just compare the final performance.

 

[Ranth]

Phenom II is legendary, but their days are long gone... all i can see in the benchmarks you posted is that the FX4350 beats the 965BE any time, both at stock.

Both CPUs OCed, the FX is still ahead... don´t compare clock for clock, just compare the final performance.

But we are talking apu's here, l3 cacheless chips, since there are no steamroller fx cpus.

 

[etayorius]

What the Heck... double post.

 

[etayorius]

The 750k may be FM2, but it is not an APU and the 7850k will be far ahead than any PhenomII, and since i did not see Richland in those benchmarks, the closest thing is the FX4350.

7850K would be 10% average and up 20% faster than the FX4350 and about 30-35% faster than the fastest PhenomII with 4 Cores.

Some people in here claimed that Kaveri would be 10-15% (at BEST) faster than Richland, we know Kaveri architectural improvements are 20% in regard to Richland, and it would had been up to 35% faster if it had the same Speed and Max Turbo as Richland, there is no 10-15% loss in performance with Kaveri not Having the same speed as Richland, so those people were DEAD WRONG.

Some even went as far as to claim Kaveri would be slower than Richland because of the loss of Speed, i had a small debate with that person that Steamroller plans were to bring 20% IPC gains and a 10-15% more speed than previous Piledriver, and it seems AMD partially delivered... they were able to add the architectural Changes to Kaveri that improve 20% IPC but they fell short with Kaveri on the 10-15% more speed (Brought to you by GlobalScrewndies) compared to Richland, none the less the 20% Gains are there.

I can`t seem to find the link to an article about AMD improving L1 and L2 to make up to the lack of L3:

http://www.tomshardware.com/news/AMD-Steamroller-Piledriver-Kaveri-processors,17217.html


I see WIN-WIN in Kaveri, AMD did a damn good job with it, personally the only thing i don´t like is the 175 USD Price... i was expecting a 160, but come on! is still a very good price.

I have a feeling AMD in the next 8 months will release 7870k or 7890k APU that will add that extra 10-15% more speed.

 

[gamerk316]

Some people in here claimed that Kaveri would be 10-15% (at BEST) faster than Richland, we know Kaveri architectural improvements are 20% in regard to Richland, and it would had been up to 35% faster if it had the same Speed and Max Turbo as Richland, there is no 10-15% loss in performance with Kaveri not Having the same speed as Richland, so those people were DEAD WRONG.

IPC gains are not flat. Sure, one benchmark could be 40% faster, but the average performance could be far lower then that.

My take is simple: You advertise best case numbers. AMD is advertising 20%. Figure two-thirds of the "best case" is the "nominal case" result. That equates to just over 13%.

People REALLY buy into marketing way to easily. You'd think after the BD launch debacle, people would learn their lessons.

 

[Cazalan]

VLIW4 -> GCN should be giving bigger gains than that. Drivers must still need work.

Currently though, adding 1.1 B transistors for a few % isn't looking very cost effective at the moment. They almost could have fit 2x Richland with that many transistors giving an 8 core APU.

Sometimes you have to wonder why they even bother with NDAs this close to launch.

"Despite never communicating so to us, it has become clear that AMD intends for performance data to be withheld from the public until January 14th. In good faith we have proactively decided to take down this review until that time. Sorry everyone!"

Too late, it's already been mirrored by dozens of sites.

 

[gamerk316]

^^ Wow; that's saying something.

In regards to the GPU performance, its possible the CPU is holding it back. No way to prove this though...

 

[logainofhades]

Phenom II is legendary, but their days are long gone... all i can see in the benchmarks you posted is that the FX4350 beats the 965BE any time, both at stock.

Both CPUs OCed, the FX is still ahead... don´t compare clock for clock, just compare the final performance.

But we are talking apu's here, l3 cacheless chips, since there are no steamroller fx cpus.

^This. The 750k is no different than an A10 5700 other than unlocked multiplier and lacks the IGP. Richland really didn't improve on the CPU side of the APU. 750k loses to a PhII 965 at stock speeds and even when overclocked to a 300mhz higher clock than PhII 965. Kaveri will not beat FX and the statement that APU's will finally reach PhII performance is actually a pretty fair assumption to make. In all fairness, clock for clock, PhII is still pretty competetive against piledriver. A 4.0ghz PhII X4 isn't much slower than 4.2ghz FX 4350. FX's strength is more cores for 6300 and up and the fact that they can clock so high.

 

[etayorius]

[/quotemsg]
PhII is still pretty competetive against piledriver. A 4.0ghz PhII X4 isn't much slower than 4.2ghz FX 4350. [/quotemsg]

So you´re comparing a OCed Top CPU of 2010 with a Lowly 4 cores CPU of 2013 at Default speed? My take is... the FX beats it ANY TIME, and Kaveri will mop the floor with any PhenomII 4 Cores.

At the end of the day it does not matter if Clock for Clock PhenomII is faster, because the FX just pulls ahead anytime.

Oh and Gamer, how come your 15% prediction stands up? i remember quite clearly that your prediction was 5-10 AVG and up to 15% in the very best case scenario... we already got a few benchmarks and we seen a 20% average and up to 30% improvements in some areas... i would say your prediction is falling apart very quick... and i also see you fixing your prediction from 15 Max% to a 15% Avg... it´s not the same... you know? just saying.

 

[gamerk316]

See previous page. Right now, Kaveri not looking as good as predicted in actual use, even worse then I suspected.

Of course, of you look only at best case benchmarks, every product is 20% better then the one that came before.

 

[logainofhades]

Kaveri won't beat FX. FX Piledriver/Vishera will still be the enthusiast class performance chips from AMD for this year. Trinity and Richland are slower than Phenom II due to the lack of L3 cache. They are actually somewhere in between Athlon II and PhII for performance. Faildozer was worse than PhII. In all fairness, even Piledriver really doesn't improve IPC much over Ph II either. Matching PhII performance is a fair assumption at this point based on what we know.

 

[noob2222]

The 750k default speed is 3.4 ghz. 750k is a richland die. The oc for the 750k is 4.3 ghz and gets beat by PII 4.0 ghz.

I never questioned that FX chips were faster, but more questioning how far the lack of l3 cache is going to hurt overall performance. Lack of l3 cache is what lets the aging phenom II keep up with Athlons and APUs.

Think about it this.

Intel doesnt have a l3 cacheless cpu but somehow AMD is going all l3 cacheless according to any recent roadmaps. How is this competitive in todays market other than having a better IGP?

We will see in about a week from now.

@gk

Lets hope those geekbench numbers are off, I was expecting around 10% better than richland, not less than 5%.

 

[Rum]

In other news, more info on Freesync:

http://www.pcper.com/reviews/Graphics-Cards/AMD-Variable-Refresh-FreeSync-Could-Be-Alternative-NVIDIA-G-Sync

So in short:

1: Embedded Display Port 1.3, launched in 2013, has a mechanism for supporting variable refresh rates.
2: AMD is pushing for this feature in the next version of Display Port (1.3).
3: The majority of monitors, at present, do not support variable refresh rates. In theory, you could do this via firmware update, but for end users, this still almost certainly requires getting a new monitor.
4: Will only work using a native Display Port 1.3 connection.

So it comes down to the panel makers really, to see which tech gets supported first. NVIDIA has the edge, since DP 1.3 won't even be standardized until March/April, and panels based off the won't be released until the later half of the year. Meanwhile, Gsync can be integrated today. Advantage NVIDIA.

And I STILL want to know how AMD is handling setting up the Vblank intervals directly when you can't know the FPS in games ahead of time in a way that doesn't introduce lag into the system.

EDIT

NVIDIA's response:

http://techreport.com/news/25878/nvidia-responds-to-amd-free-sync-demo

*popcorn*

So what's your point? Gsync integrated monitors won't be released till at least Q2 of 2014 with a 300-400$ premium ( DP 1.3 will be standardize around the same time with companies being given access to it). The added cost alone will keep them on the shelves as a niche product and with an open standard being released in the coming months do you honestly think that companies will want to vendor lock their monitors and lose out on profit? Nvidia has the benefit of releasing Gsync first but as soon as the people realize that they can have the same thing cheaper it will fall by the wayside! Cmon DP 1.3 allows for 4k @120hz and 8K resolution same, refresh tech and cheaper who the heck would pass that up for Gsync?

Gsync is just a way for Nvidia to try and stunt the marketing/ implementation of an open standard so they can milk their customers for more money. " Hey buy our exclusive technology that will be free soon" -_- For someone who always chastises others about believing company marketing hype you sure have swallowed Nvidias hook line and sinker...

I am with you on one thing though, I am curious to see how AMD is going to implement FreeSync.

 

[gamerk316]

You think a single FPGA is going to add $400 to new monitors? Are you kidding me? $50, maybe.

And AMD's implementation isn't going to be free either. 60Hz has been ingrained for a LONG time, and adding circuitry for setting a dynamic Vblank interval, controlled by an external device [in this case, the PC], is not arbitrary. Hence why NVIDIA made their own chip to do it. And what its more then likely will be required for Freesync to be implemented by panel makers. Both are going to require a hardware mod to current screens, the only difference in NVIDIA has a two quarter head start.

And I still want to know how AMD is implemented Vsync, since I'm growing more convinced latency is going to be an issue.

 

[Rum]

You think a single FPGA is going to add $400 to new monitors? Are you kidding me? $50, maybe.

And AMD's implementation isn't going to be free either. 60Hz has been ingrained for a LONG time, and adding circuitry for setting a dynamic Vblank interval, controlled by an external device [in this case, the PC], is not arbitrary. Hence why NVIDIA made their own chip to do it. And what its more then likely will be required for Freesync to be implemented by panel makers. Both are going to require a hardware mod to current screens, the only difference in NVIDIA has a two quarter head start.

And I still want to know how AMD is implemented Vsync, since I'm growing more convinced latency is going to be an issue.

This is Nvidia we are talking about you know they are going to charge a premium for Gsync, they have a track record of over charging and to think otherwise is foolish! 300-400 was more of an inflammatory remark to towards Nvidias pricing practices and I wouldn't be surprised if it added 120.00 like Fudzilla posted back in October. Nvidia will get support for this because it is a money maker and it is to the market sooner, but as soon as an open standard is released it will crash like physx has and relegated to a small corner of the market...

As for the possible latency we will not know till we can get our hands on it. from the demos people have seen they said it looked pretty good.

 

[jdwii]

^^ Frankly, not impressed. Looking at the first chart, a combined CPU/GPU upgrade is only buying a few FPS at low settings. Granted, AMD still crushes intel in the iGPU deparment.

Second chart shows Intel still winning on CPU performance head to head by a decent margin. Not shocked in the least.

Third one shows Kaveri starts to reach max potential at 1866MHz RAM. Some gains going faster, but for most of the tests, <3 FPS gains from that point onward. More likely secondary improvements at that point, rather then the GPU still being starved. Clearly no issues with Kaveri getting starved on DDR3.

I think most here are/were expecting a bigger improvement then those charts indicate.

Also FYI: Source?

But...But 30% more IGPU power lol if anything my estimates where to much WTF Amd. 512 GCN cores really? look at the memory test as predicted the memory bandwidth is bottlenecking the GPU to much.

 

[jdwii]

Some people in here claimed that Kaveri would be 10-15% (at BEST) faster than Richland, we know Kaveri architectural improvements are 20% in regard to Richland, and it would had been up to 35% faster if it had the same Speed and Max Turbo as Richland, there is no 10-15% loss in performance with Kaveri not Having the same speed as Richland, so those people were DEAD WRONG.

IPC gains are not flat. Sure, one benchmark could be 40% faster, but the average performance could be far lower then that.

My take is simple: You advertise best case numbers. AMD is advertising 20%. Figure two-thirds of the "best case" is the "nominal case" result. That equates to just over 13%.

People REALLY buy into marketing way to easily. You'd think after the BD launch debacle, people would learn their lessons.

I stood by 15% improvements on the CPU on average not max not using upto statements or just synthetic benchmarks. I honestly thought the GPU would be 25% more powerful on average. But if some haven't noticed Bulldozer is ram hungry compared to even Phenom i noticed this with Llano to trinity. When to are fighting for the bandwidth performance will be bottle necked and yes we are seeing this.
Even though Amd fanboys who go to far will say i hate Amd or you do even though all i own is Amd and if anything want them to continue since i can build a better gaming machine per dollar with them. Unless they get rid of their 8 cores for these 4 cores.

 

[etayorius]

Phenom II is legendary, but their days are long gone... all i can see in the benchmarks you posted is that the FX4350 beats the 965BE any time, both at stock.

Both CPUs OCed, the FX is still ahead... don´t compare clock for clock, just compare the final performance.

But we are talking apu's here, l3 cacheless chips, since there are no steamroller fx cpus.

^This. The 750k is no different than an A10 5700 other than unlocked multiplier and lacks the IGP. Richland really didn't improve on the CPU side of the APU. 750k loses to a PhII 965 at stock speeds and even when overclocked to a 300mhz higher clock than PhII 965. Kaveri will not beat FX and the statement that APU's will finally reach PhII performance is actually a pretty fair assumption to make. In all fairness, clock for clock, PhII is still pretty competetive against piledriver. A 4.0ghz PhII X4 isn't much slower than 4.2ghz FX 4350. FX's strength is more cores for 6300 and up and the fact that they can clock so high.

Kaveri won't beat FX. FX Piledriver/Vishera will still be the enthusiast class performance chips from AMD for this year. Trinity and Richland are slower than Phenom II due to the lack of L3 cache. They are actually somewhere in between Athlon II and PhII for performance. Faildozer was worse than PhII. In all fairness, even Piledriver really doesn't improve IPC much over Ph II either. Matching PhII performance is a fair assumption at this point based on what we know.

Really? because CPUBOSS is reporting otherwise from your claim, it loses in Cinebench 3D Mark11 and wins in GeekBench and 3D PassMark with the same percentage.

http://cpuboss.com/cpus/AMD-Phenom-II-X4-Black-980-vs-AMD-A10-6800K


And PerformanceTest also says you`re wrong since it gives an overall CPU Mark score of 5,414 to Richland and 4,680 to the 980.

https://www.dropbox.com/s/eotom6rspg7menp/PT-6800k-1.jpg

https://www.dropbox.com/s/vidvt89tnzijyvn/PT-6800k-2.jpg


I dont know if you see that... but Richland Thrashed around the 980 in 7 scenarios and losing only in 2, so i would say Richland beats the 980 most of the time.

Here is Trinity against the 980, and they seem to be kinda on the same level, but 980 winning 3 more than Trinity... too bad they don`t have Richland available in their "benchmarks", we know Richland is up to 10% faster than Trinity.

http://www.anandtech.com/bench/product/362?vs=675


I trust PerformanceTest more than most other Benchmarking Tools, i can OC a mere 5% my PhenomII 980 and PerTest will notice that and add the boost to my score accordingly, so i can compare against other User Setups.