AMD CPU speculation... and expert conjecture

[jdwii]

^^ But remember, Intel focuses on the iGPU this go around, and got results. I'm seeing numbers that average to abut 40% performance increases generationally, which is significant to say the least. CPU side, maybe 10% strictly on IPC, or about what we've been seeing for a while now.

http://www.hardwarecanucks.com/forum/hardware-canucks-reviews/61510-amd-richland-review-a10-6800k-a10-6700-benchmarked-20.html
looks like 10-15% improvement from trinity to richland on 1600mhz ram. Haswell is still worse than llano in terms of gaming performance.

Ha ha Intel, i know its a fallacy argument but is it safe to say Amd>Intel for PURE IGPU graphics performance for ever and ever, when Amd and even Nvidia are actually invading into the graphics sector and their hard work with game developers you will never see Intel be better(overall) unless Intel makes video cards so they can get better driver support.

 

[sarinaide]

Steamroller is starting to tease me I don't like waiting.

Anyways the interesting news of the day is that a lot of distributors and resellers are pulling their HD7970's, could this mean that we can expect a new graphics line up.

HD7850 now the fastest passively cooled GPU with no auxillary power ....I think I have found heaven.

 

[de5_Roy]

Am I the only one who thinks that it would be in AMD's best interest to ship the $100 worth of faster RAM to reviewers to get better reviews? I mean, WTH is AMD thinking?

most richland reviewers seem to use 2133 ram with 6800k. some didn't. some didn't even mention ram spec. it's mixed.
i guess amd did supply 2133 ram, but s/a didn't get it....? or may be they trust their own hardware, i dunno...

Still, CPU benchmarks basically show little change; still about 30% slower then IB, and 40% slower then HW. So it will be interesting to see how much faster RAM helps the GPU side of the house.Though I do find it funny the people who were claiming 40% improvements in a single generation haven't chimed in yet. Just saying, again, never make assumptions based on "best case" benchmarking.

this review uses 2133 ram in all test pcs except with 6700:
http://www.legitreviews.com/article/2209/1/
may have finally gotten the explanation why s/a's ram wasn't oc'ed. they should've at least mentioned it in the reivew, if it really happened... l.r. also have rather amusing bits of info:

We contacted AMD about our performance numbers and they said that we did not see huge performance gains since we ran the old AMD A10-5800K APU with 2133MHz memory and not 1866MHz, which is the highest memory clock speeds officially supports on it. We ran both the 5800K and 6800K at 2133MHz to keep the test systems as close as possible and hoped to run it on the A10-6700. Hoped is the key word as we were unable to get the A10-6700 stable with the Corsair Dominator Platinum 2133MHz CL9 1.65V memory kit. AMD specifies that the A10-6800K officially supports 2133MHz memory with 1.50V, so maybe our kit requires too much voltage. Just a heads up to anyone looking to run 2133MHz memory with a Richland APU. Our experience with Richland is that you should run 2133MHz memory to get the most performance out of it, but the A10-6800K seems more receptive to it and we suggest a 1.50V kit.

you have to use slower ram with 5800k to make 6800k look much faster. some other reviewers used ddr3 1600/1866 in all test rigs but these guys did better. :lol:

 

[Cazalan]

i bet it's intel's fault again, they hijacked amd review kits midway and replaced 2133 ram with lower specced ram and edited the review guide! corporate espionage is at play. brace yourselves people!

AMD's infamous marketers that are probably interns clueless about hardware. They did fire 33% of their staff.

 

[de5_Roy]

AMD's infamous marketers that are probably interns clueless about hardware. They did fire 33% of their staff.

nah, i got my answers already. that bit was not written seriously anyway. let it rest.

 

[gamerk316]

this review uses 2133 ram in all test pcs except with 6700:
http://www.legitreviews.com/article/2209/1/

And I don't see significant performance changes that can't be explained by the 6700k being 400MHz slower then the 6800k. In short, I don't see a performance increase going from 1866 to 2133.

Its worse then that actually:

http://www.legitreviews.com/article/2209/11/

Despite a 300MHz advantage over the 5800k, the 6800k generates basically the same exact performance with 2133Mhz RAM. Really, I see ZERO performance increase generation to generation in games.

EDIT

Translation: IPC looks worse, if you get the same performance at a higher clockspeed. Either that, or there's a MAJOR system bottleneck somewhere.

 

[Cazalan]

you have to use slower ram with 5800k to make 6800k look much faster. some other reviewers used ddr3 1600/1866 in all test rigs but these guys did better. :lol:

That's quite the conundrum. Officially the A10-5800K supports 1866 so that legit reviews benchmark was essentially overclocked. What it tells budget enthusiasts is that $20 price difference between 5800/6800 is better spent on faster DRAM. Not really good for AMD.

AMD marketing has dropped the ball in signifying how important the DRAM is to getting peak APU performance. The chips are going in the lower end laptops/desktops and ODMs/OEMs are bundling the cheapest RAM they can get which makes for under performing reviews.

It also shows that Kaveri REALLY needs some eDRAM (32MB minimum) or it will still be hamstrung in memory bandwidth. The initial die shots don't appear to have any.

 

[sarinaide]

this review uses 2133 ram in all test pcs except with 6700:
http://www.legitreviews.com/article/2209/1/

And I don't see significant performance changes that can't be explained by the 6700k being 400MHz slower then the 6800k. In short, I don't see a performance increase going from 1866 to 2133.

Its worse then that actually:

http://www.legitreviews.com/article/2209/11/

Despite a 300MHz advantage over the 5800k, the 6800k generates basically the same exact performance with 2133Mhz RAM. Really, I see ZERO performance increase generation to generation in games.

EDIT

Translation: IPC looks worse, if you get the same performance at a higher clockspeed. Either that, or there's a MAJOR system bottleneck somewhere.

HardwareCanucks

S/A

XBit

All computing and systems benches improved universally, IPC improved, what the 6800K suffers from is what the 5800K suffers from in Gaming terms and that is severe bandwidth limitations. HD6670 DDR3 vs the DDR5 version is around 20% difference, basically Richland and Trinity have hit the system memory bandwidth brickwall and as they said there is nothing new or different in richland on a microarchitecture level.

 

[earl45]

this review uses 2133 ram in all test pcs except with 6700:
http://www.legitreviews.com/article/2209/1/

And I don't see significant performance changes that can't be explained by the 6700k being 400MHz slower then the 6800k. In short, I don't see a performance increase going from 1866 to 2133.

Its worse then that actually:

http://www.legitreviews.com/article/2209/11/

Despite a 300MHz advantage over the 5800k, the 6800k generates basically the same exact performance with 2133Mhz RAM. Really, I see ZERO performance increase generation to generation in games.

EDIT

Translation: IPC looks worse, if you get the same performance at a higher clockspeed. Either that, or there's a MAJOR system bottleneck somewhere.

you have to use slower ram with 5800k to make 6800k look much faster. some other reviewers used ddr3 1600/1866 in all test rigs but these guys did better. :lol:

That's quite the conundrum. Officially the A10-5800K supports 1866 so that legit reviews benchmark was essentially overclocked. What it tells budget enthusiasts is that $20 price difference between 5800/6800 is better spent on faster DRAM. Not really good for AMD.

AMD marketing has dropped the ball in signifying how important the DRAM is to getting peak APU performance. The chips are going in the lower end laptops/desktops and ODMs/OEMs are bundling the cheapest RAM they can get which makes for under performing reviews.

It also shows that Kaveri REALLY needs some eDRAM (32MB minimum) or it will still be hamstrung in memory bandwidth. The initial die shots don't appear to have any.

Can we finally start calling the new AMD chips (Ditchland)? :lol:

 

[Cataclysm_ZA]

Anyways the interesting news of the day is that a lot of distributors and resellers are pulling their HD7970's, could this mean that we can expect a new graphics line up.

You've also noticed that? I've even asked a couple of online retailers about it and they either don't know themselves, or won't tell me because they're under NDA. Distributors just flat out ignore me, so that may be a sign.

That's quite the conundrum. Officially the A10-5800K supports 1866 so that legit reviews benchmark was essentially overclocked.

PC Perspective noted that the BIOS updates for boards to support Richland also unlock DDR3-2133 support for the A10-5800K. Anand broke out DDR3-2400 kits for Intel when testing Iris and Iris Pro and the HD4600, but Trinity still maintained its performance advantage. Bit-Tech's testing showed a massive performance gulf because they included scores for cheaper i3 and i5 processors running on HD2500 and HD3000 graphics.

 

[de5_Roy]

.....
AMD marketing has dropped the ball in signifying how important the DRAM is to getting peak APU performance. The chips are going in the lower end laptops/desktops and ODMs/OEMs are bundling the cheapest RAM they can get which makes for under performing reviews.
....

i think that this corner-cutting strategy is the reason why intel cpus often outperform mobile amd apus. since the apu needs to be fed memory bw by faster ram, laptop perf takes a hit. meanwhile, intel can squeeze out more bw out of low speed ram, so they can close distance or pull ahead.

 

[sarinaide]

you have to use slower ram with 5800k to make 6800k look much faster. some other reviewers used ddr3 1600/1866 in all test rigs but these guys did better. :lol:

That's quite the conundrum. Officially the A10-5800K supports 1866 so that legit reviews benchmark was essentially overclocked. What it tells budget enthusiasts is that $20 price difference between 5800/6800 is better spent on faster DRAM. Not really good for AMD.

AMD marketing has dropped the ball in signifying how important the DRAM is to getting peak APU performance. The chips are going in the lower end laptops/desktops and ODMs/OEMs are bundling the cheapest RAM they can get which makes for under performing reviews.

It also shows that Kaveri REALLY needs some eDRAM (32MB minimum) or it will still be hamstrung in memory bandwidth. The initial die shots don't appear to have any.

1) Very simple people knew what bandwidth meant to GPU's, ie: the reason GDDR5 replaced GDDR3, a HD7970 with GDDR3 will perform around a 3rd of its current performance due to the bandwidth limitation.

2) Hence why the talk is about embedded GDDR5 or some form of L4 Cache or the likes, I still believe embedding GDDR5 on a motherboard is the cheapest and 256-512 is more than enough and you can just add extra controllers to the NB to allow it exclusive usage by the iGPU.

 

[earl45]

I would like to see the A10-5800 vs A10-6800 running at the same clock speeds
performing cpu intense tasks.
http://www.techspot.com/review/681-amd-a10-6800k-a4-4000/page9.html
in that area it don't look like such a good upgrade.

 

[griptwister]

palladin9479: NICE CATCH!

So, My question is (it might be a noob question) Will Steamroller APUs be a modification or Richland and the FX line a modification of Piledriver? Or did I just mis-understand something with the naming scheme?

Yes, Steamroller APU's will be a modification of Richland, but much more so than Richland was of Trinity and still even moreso than Trinity was of Llano. Steamroller will be the biggest change in the Module architecture in it's history. Hopefully, this will result in the performance increase that so many of us are anticipating.

My biggest hope is in the duplicating of the decode hardware in each module so that each core has it's own 4-wide instruction decoder, which will run in parallel rather than alternating cycles. This means that a single Steamroller Module (which I compare to a single Intel core) can potentially have a rate of 8 instructions/cycle as opposed to the previous 4. This should technically put Steamroller finally on par with Intel in that area, actually beyond with 8 cores vs Intel's 6, and hell, it should technically put Steamroller finally beyond Phenom II, which has outperformed anything AMD has released in that area since (at least beyond single core performance). Here, I edited a slide from one of Anand's articles, adding in a Steamroller column...

+1 to you for the elaborate yet easily understood explanation, and +1 to 8350rocks for another easily understood explanation.

 

[Cataclysm_ZA]

I would like to see the A10-5800 vs A10-6800 running at the same clock speeds
performing cpu intense tasks.

Keep your eye open for Tech Report's review. Damage is busy running the chips through their new testing suite and its going to be really, really interesting.

In the meantime, TechpowerUp managed a smashing 5.0GHz overclock with little trouble.

http://www.techpowerup.com/reviews/AMD/A10-6800K/1.html

 

[Cazalan]

i think that this corner-cutting strategy is the reason why intel cpus often outperform mobile amd apus. since the apu needs to be fed memory bw by faster ram, laptop perf takes a hit. meanwhile, intel can squeeze out more bw out of low speed ram, so they can close distance or pull ahead.

Intel APUs still have L3 as well. It helps a bit. I noticed the FX-4350 had some CPU benches that were much higher than A10-6800k, when they should be really close.

http://www.techspot.com/review/681-amd-a10-6800k-a4-4000/page4.html

 

[gamerk316]

this review uses 2133 ram in all test pcs except with 6700:
http://www.legitreviews.com/article/2209/1/

And I don't see significant performance changes that can't be explained by the 6700k being 400MHz slower then the 6800k. In short, I don't see a performance increase going from 1866 to 2133.

Its worse then that actually:

http://www.legitreviews.com/article/2209/11/

Despite a 300MHz advantage over the 5800k, the 6800k generates basically the same exact performance with 2133Mhz RAM. Really, I see ZERO performance increase generation to generation in games.

EDIT

Translation: IPC looks worse, if you get the same performance at a higher clockspeed. Either that, or there's a MAJOR system bottleneck somewhere.

HardwareCanucks

S/A

XBit

All computing and systems benches improved universally, IPC improved, what the 6800K suffers from is what the 5800K suffers from in Gaming terms and that is severe bandwidth limitations. HD6670 DDR3 vs the DDR5 version is around 20% difference, basically Richland and Trinity have hit the system memory bandwidth brickwall and as they said there is nothing new or different in richland on a microarchitecture level.

Not from what I'm seeing; The improvement is small, and could be easily attributable to the extra 400mhz clock. IPC is either unchanged, or very minor changes in either direction. I REALLY want to see a review at the same clocks now.

As for RAM, if the GPU WAS starved, then performance should jump going from 1866 to 2133. I don't see that, which indicates the bottleneck may be somewhere else.

 

[Cazalan]

2) Hence why the talk is about embedded GDDR5 or some form of L4 Cache or the likes, I still believe embedding GDDR5 on a motherboard is the cheapest and 256-512 is more than enough and you can just add extra controllers to the NB to allow it exclusive usage by the iGPU.

The pics of FM2+ are out now. The pin count is essentially the same so that rules out additional memory interfaces. eDRAM is still possible but the die shots don't look big enough to have 32MB eDRAM.

FM2+ Kaveri

FM2/FM1 Trinity/Llano

 

[gamerk316]

Not shocked; EDRAM is expensive. Probably the same reason GDDR5 has vanished from the charts.

Although I find some vindication; I predicted a few years back the memory bus was going to start being a major system bottleneck again.

 

[8350rocks]

Not shocked; EDRAM is expensive. Probably the same reason GDDR5 has vanished from the charts.

Although I find some vindication; I predicted a few years back the memory bus was going to start being a major system bottleneck again.

It always comes full circle...when something is a bottleneck, hardware comes out improving that...then something is the new bottleneck, they improve that, and then it's something else. Eventually the first thing they improved is worse than the other things that were improved after it.

 

[Cataclysm_ZA]

Squint your eyes and the pin counts look like oversized heads of the monsters in Minecraft...!

FM2+ is interesting though. That's what... 901 pins?

 

[The Q6660 Inside]

Squint your eyes and the pin counts look like oversized heads of the monsters in Minecraft...!

FM2+ is interesting though. That's what... 901 pins?

Awww, I wanted 939 Pins. Anyways, if Kaveri is oriented towards gamers (7830/8770 levels), AMD could make a killing in the $150-200 APU/integrated GFX CPU range. Richland is a decent upgrade and I can basically quote Yuka's "DAT PERFORMANCE" quote.
Update: @Palladin9479, *cough*4GB DDR3@1333 on 6800K *cough cough*

 

[Cataclysm_ZA]

http://wccftech.com/amd-demonstrates-fourth-generation-kaveri-apu-computex-2013-ruby-techdemo-showcased/

Aside from Ruby, the Kaveri chips are interesting. The chip itself is very rectangular.

Compared to Trinity's die, it's only slightly smaller, but I'm itching to see what it's made up of. Now that we have Steamroller on its way, this thread will grow even larger!

 

[Ranth]

Hopefully we won't have problems with the the thread due to the overload of pages! As it happened with the other AMD thread.

 

[-Fran-]

Man... That first pic in the link screams for a "I give you the finger" 'shop, hahaha.

Other than that, it's a very nice showing. Looks like AMD might indeed have something very interesting to show next quarter.

Cheers!