The Kaveri Pre-Launch Thread (A10-7800 and A10-6800k @3,5 Ghz)

[Lepton87]

Honestly I didn't expect much from steamroller because otherwise cancelling FX products based on it wouldn't make sense, but I hoped that I was wrong. Unfortunately it's indeed a minuscule improvement in IPC that can hardly make up for the frequency deficit I won't even comment on graphics performance. They really needed a more efficient memory controller that would at least be on par with Intel's.

 

[AtenRa]

So, a 65W TDP is faster than 100W TDP in iGPU gaming. Not bad, lets see what the 95W TDP can do.

Also, lets wait for CPU + dGPU gaming performance

 

[Enigmoid]

So, a 65W TDP is faster than 100W TDP in iGPU gaming. Not bad, lets see what the 95W TDP can do.

Also, lets wait for CPU + dGPU gaming performance

Because that was so true for Richland?

http://techreport.com/review/24954/amd-a10-6800k-and-a10-6700-richland-apus-reviewed

http://www.tomshardware.com/reviews/a10-6700-a10-6800k-richland-review,3528-11.html

a10-6800k 100W vs. a10-6700. IGP benchmarks- identical, CPU benchmarks, nearly identical. Yet the a10-6700 uses less power.

Not saying that it isn't but richland had no problem matching a 100W tdp with 65W (better binning).

 

[DaZeeMan]

Saw this image in another thread. Assuming the info is correct, 3.9/3.5 Mhz at 65W for the 7800, and 4.0/3.7 MHz at 95W for the 7850. . This is one of the images in the 'early review' mentioned at the top of this thread.

 

[mikk]

DDR3-2133 confirmed now? I didn't know that the 720 Mhz GPU is a Turbo, so in some heavy CPU gaming GPU frequency could drop below 700 Mhz. No wonder Richland is so competitive with 384 VLIW4 Shaders. I assume the 65W model suffers a bit more.

 

[DaZeeMan]

A poster in another thread pointed out that 2400 MHz has no 'official' JEDEC spec yet.
I tried downloading the 'official' JEDEC document, but it requires a login... so here's what Wikipedia has to say on the matter:
http://en.wikipedia.org/wiki/DDR3_SDRAM

Scroll to about the middle, and you'll note that 2133 is the highest listed.

So the 2133 may be the maximum 'official' support they can list for Kaveri, as 2400 has no JEDEC spec yet.

 

[Homeles]

For the JEDEC site, you just need to register.

 

[ElFenix]

A poster in another thread pointed out that 2400 MHz has no 'official' JEDEC spec yet.
I tried downloading the 'official' JEDEC document, but it requires a login... so here's what Wikipedia has to say on the matter:
http://en.wikipedia.org/wiki/DDR3_SDRAM

Scroll to about the middle, and you'll note that 2133 is the highest listed.

So the 2133 may be the maximum 'official' support they can list for Kaveri, as 2400 has no JEDEC spec yet.

lack of a final jedec standard never stopped Via from officially supporting 133 mhz sdram back in the day.

 

[mikk]

Yes AMD could support higher memory speeds, they are responsible and could theoretically validate their platform/CPU combo for DDR3-2400. It is surely unusual to validate for something that JEDEC didn't specify. The chinese spec sheet listed only DDR3-2133 though.

 

[DaZeeMan]

I am pretty sure that AMD is a signatory to the JEDEC standard, which may limit what they can 'officially' support. Don't really know though. Several Richlands have been able to run DDR3 2400 memory, or even higher speed (one of the reviewers mentioned a successful 2566 or so overclock when I was googling around yesterday), depending on the brand and such.

That being said, we've seen Motherboard manufacturers support higher ram speeds in their specs. Also, with some memory brands not playing nice at 2400 MHz with Richland, AMD may just be playing it safe.

It'd be nice if JEDEC would get off their butt and certify a 2400 MHz DDR3 standard!

 

[Homeles]

I can't imagine JEDEC forseeing that DDR3 would scale to such high speeds. Their specifications have kind of been thrown out the window for several years now.

 

[SiliconWars]

DDR3-2133 confirmed now? I didn't know that the 720 Mhz GPU is a Turbo, so in some heavy CPU gaming GPU frequency could drop below 700 Mhz. No wonder Richland is so competitive with 384 VLIW4 Shaders. I assume the 65W model suffers a bit more.

512 GCN at 654 MHz should still be enough to go well past 384 VLIW4 at 844 MHz under normal conditions (should still be ~30% faster if you consider GCN is around 25% faster per shader). This is a hard limit on bandwidth I'm afraid. With that said I would expect to see some very decent gains with gaming perf/W and in shader bound games Kaveri will win some by over 20%.

 

[Arkaign]

I can't imagine JEDEC forseeing that DDR3 would scale to such high speeds. Their specifications have kind of been thrown out the window for several years now.

Very true. Official JEDEC is kind of a useless spec given that OEMs barely touch anything past 1600 anyway (price + lack of education with the average consumer = you don't see high performance memory in mass market PCs/laptops anyway).

I have 32GB 2400, and according to ARK my CPU doesn't officially support such a speed. 1333 and 1600 only, lol. That doesn't stop me from running 2400 with no problems.

For systems with discrete this mostly doesn't amount to much anyway. But for APUs, fast memory can mean a world of difference (1066 vs. 2133 on an A10 is a massive massive difference hahah). The problem with mass market APU systems is that they're often paired with terrible memory. AMD should address this somewhat I would hope.

 

[sniffin]

The problem with mass market APU systems is that they're often paired with terrible memory. AMD should address this somewhat I would hope.

For whatever reason they seem to be incapable of doing this. After watching the Trinity launch I'm convinced AMD either don't have enough influence to convince OEMs to put the chips into machines that aren't cheap garbage, or they are just happy with OEMs buying their stuff at all and aren't concerned with the lack of quality shells.

Either way I expect mobile Kaveri to suffer the same fate. You won't find A10s in anything less than 15.6" machines with terrible screens and resolutions. MSI's GX notebook line will be the sole exception.

 

[AtenRa]

Because that was so true for Richland?

http://techreport.com/review/24954/amd-a10-6800k-and-a10-6700-richland-apus-reviewed

http://www.tomshardware.com/reviews/a10-6700-a10-6800k-richland-review,3528-11.html

a10-6800k 100W vs. a10-6700. IGP benchmarks- identical, CPU benchmarks, nearly identical. Yet the a10-6700 uses less power.

Not saying that it isn't but richland had no problem matching a 100W tdp with 65W (better binning).

Techreport uses 1600MHz memory , it is only natural that when both 6800K and 6700 will be memory Bandwidth limited they will perform the same since they have exactly the same iGPU and Clocks.

Toms review uses 1866MHz for both, but they also include 2133MHz for the 6800K. With 2133MHz which is the official memory, 6800K is 15-20% faster than 6700. If you use 2133MHz with 6700 then you raise the TDP/consumption.

Kaveri is rated at 65W TDP with 2133MHz memory, not only that but it is also a little faster than 6800K 100W TDP Ritchland.

Edit: Techreport also has the 6800K with 2133MHz

 

[xSneak]

I imagine next week at ces we will get some actual benchmarks to confirm or deny these ones. The graph only says 7800 so maybe it's not the 7850 model with the 512 stream processors.

 

[AtenRa]

I imagine next week at ces we will get some actual benchmarks to confirm or deny these ones. The graph only says 7800 so maybe it's not the 7850 model with the 512 stream processors.

7800 is a different CPU than 7850K

 

[inf64]

7800 is locked, lower clocked and lower TDP part(65W). It will also command a higher price. From what we can observe in these benchmarks it seems it doesn't Turbo much (if at all) in CPU benchmarks and I would warrant a guess it does similar in GPU benchmarks. After all it is 65W part.

 

[Galatian]

7800 is a different CPU than 7850K

Yes but both use the 512 Stream processor graphic chip. The difference shouldn't be to big.

 

[AtenRa]

7800 is locked, lower clocked and lower TDP part(65W). It will also command a higher price. From what we can observe in these benchmarks it seems it doesn't Turbo much (if at all) in CPU benchmarks and I would warrant a guess it does similar in GPU benchmarks. After all it is 65W part.

Even so, it seams to me that both 65W and 95W are bandwidth limited. I dont expect a lot more iGPU Gaming performance with the 95W SKUs.
GPGPU performance will be much higher, especially vs Richland/Haswell.

 

[AtenRa]

Yes but both use the 512 Stream processor graphic chip. The difference shouldn't be to big.

In memory bandwidth limited workloads yes it will not be that much more.

 

[Fjodor2001]

Deletedhttp://tieba.baidu.com/p/2795083406

 

[NostaSeronx]

That SiSoft Financial Analysis Benchmark is HSA/GP²U+CPU enabled.

 

[inf64]

More AMD Kaveri benchmarks:

http://wccftech.com/amd-a10-7800-ka...pared-a10-6800k-tested-hd-7750-dual-graphics/

Original source in Chinese:

http://tieba.baidu.com/p/2795083406

There are no "more" Kaveri benchmarks. That is the very link from the OP that we are discussing... I hate I had to click on wcftech giving them a page visit and knowing up in front they are late and they've just ripped off some other news source who found it first. Also there is no dual graphics, those boons from wcftech cannot even read the simple charts . It's 7800+ discrete card to show how much behind the dGPU APU still is.

 

[Fjodor2001]

There are no "more" Kaveri benchmarks. That is the very link from the OP that we are discussing... I hate I had to click on wcftech giving them a page visit and knowing up in front they are late and they've just ripped off some other news source who found it first. Also there is no dual graphics, those boons from wcftech cannot even read the simple charts . It's 7800+ discrete card to show how much behind the dGPU APU still is.

Ok, sorry. I marked my post as Deleted now.