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

[Galatian]

if dual graphics this time finally works, it could be interesting with another 7750, but then again, how much less expensive (once you buy nicer memory and etc not needed without IGP usage) it is compared to a 7790/260x or something?

from what I can see at the moment, A10 kaveri is only attractive if you need the IGP (for gaming and/or OpenCL), for basic IGP usage or discrete graphics the i3 4130 is simply a better solution, and you would need lower priced parts, like the "athlon X4" line again,

Now this is an interesting point. AMD and a lot of users here claim that OpenCL is such a game changer and that AMD is taking the right route with combining weaker cores with a faster GPU. Now Tomshardware did a nice OpenCL test with the HD4600. Unfortunately they don't use a newer Richland APU for comparison, but the HD4600 wins by a huge margin actually. It's even more evident when they use both the CPU and OpenCL.

Here comes the big caveat: I have yet to see a truly fair comparison between a new A10-6xxx APU and a comparable Core i3-4xxx. They only one I saw was at X-bit Labs but they don't really test OpenCL and they still use Sysmark, which - arguably - is not a perfectly fair benchmark tool.

To summarize: One thing is for sure. For most people the Core i3-4xxx is just a much better CPU. It get's everyday work done faster, uses a lot less energy and - at least here in Germany - is cheaper then the top of the line A10 APUs (and that includes total system cost). The one and only usage scenario for an APU I can come up with is gaming on a budget. But again you usually are better off with just getting a cheaper Athlon or Pentium and pair it with a cheaper discrete GPU.

Things might change with Kaveri. We'll have to wait at least until tomorrow to see some more thorough benchmarks (and I hope Anandtech or the like will finally make a fair comparison between an A10-7xxx APU and a Core i3-4xxx, including OpenCL). The increase in single threaded performance is nothing to sneeze at, but I highly doubt it will get anywhere near the Haswell Performance. HSA sounds nice on paper, but we will need software support for this to really take off and here I also doubt many software vendors will jump on the ship, given the market relevance of Kaveri APUs at least in the foreseeable future.

Don't get me wrong, I'd love to support AMD as an underdog, but I really have to be realistic. I'm currently building an miniITX Office PC for my BF and I tried to gather as much information as possible on both the i3-4130 and the A8-6700. Again not only is the i3-4130 vastly superior, it also comes at a cheaper price for the entire system and uses a lot less energy. For me the entire Bulldozer Architecture is just a big fail. I don't care if it's because they are stuck with GloFo or if the architecture itself is to blame (and according to AMDs slides themselves they know where the problem is, but seems to take one step each year for each revision, instead of just doing it all at once and releasing a Bulldozer revision as it should have been), what matters is that I can't recommend the chip at all, when compared to a similar Intel chip.

 

[itsmydamnation]

the big change here for openCL is vliw vs GCN

https://compubench.com/compare.jsp?config_0=11993914&config_1=11906533

1536 vs 1792 shaders yet avg of about 3x the opencl perf.

 

[KompuKare]

Seems to perform better than the earlier leaks suggested. I really think AMD should try to push hybrid Crossfire and actually market specific cards for this (R9 255 for the A10-7850K, R7 250 for the A8-7700K etc) but if there is any way that they could get asymmetric shaders to work well it could help them to sell some more graphic cards.

Of course, the driver team do have their hands full ATM so they have to prioritise stuff:
-Mantle
-Enduro
-Frame pacing CF drivers for DX9 & Eyefinity
-Hybrid Crossfire
From a pure revenue point of view, Enduro should probably be their #1 but that's been obvious for a while now and it doesn't look it has been. Their lack of progress with Enduro has cost them a huge marketshare in mobiles. Yes, there are some differences with Kepler vs GCN but as W1zzard's perf/watt charts over at TPU show Cape Verde, Pitcairn score very close to their Kepler equivalents (Tahiti XT GHz edition and 90°C running Hawaii are mostly an exception but enthusiasts on forums do tend to focus on the very high-end) so I imagine Enduro vs Optimus is the main factor in their marketshare loss.

 

[Abwx]

Open CL perf on the slide bottom :

 

[USER8000]

Now this is an interesting point. AMD and a lot of users here claim that OpenCL is such a game changer and that AMD is taking the right route with combining weaker cores with a faster GPU. Now Tomshardware did a nice OpenCL test with the HD4600. Unfortunately they don't use a newer Richland APU for comparison, but the HD4600 wins by a huge margin actually. It's even more evident when they use both the CPU and OpenCL.

Here comes the big caveat: I have yet to see a truly fair comparison between a new A10-6xxx APU and a comparable Core i3-4xxx. They only one I saw was at X-bit Labs but they don't really test OpenCL and they still use Sysmark, which - arguably - is not a perfectly fair benchmark tool.

To summarize: One thing is for sure. For most people the Core i3-4xxx is just a much better CPU. It get's everyday work done faster, uses a lot less energy and - at least here in Germany - is cheaper then the top of the line A10 APUs (and that includes total system cost). The one and only usage scenario for an APU I can come up with is gaming on a budget. But again you usually are better off with just getting a cheaper Athlon or Pentium and pair it with a cheaper discrete GPU.

Things might change with Kaveri. We'll have to wait at least until tomorrow to see some more thorough benchmarks (and I hope Anandtech or the like will finally make a fair comparison between an A10-7xxx APU and a Core i3-4xxx, including OpenCL). The increase in single threaded performance is nothing to sneeze at, but I highly doubt it will get anywhere near the Haswell Performance. HSA sounds nice on paper, but we will need software support for this to really take off and here I also doubt many software vendors will jump on the ship, given the market relevance of Kaveri APUs at least in the foreseeable future.

Don't get me wrong, I'd love to support AMD as an underdog, but I really have to be realistic. I'm currently building an miniITX Office PC for my BF and I tried to gather as much information as possible on both the i3-4130 and the A8-6700. Again not only is the i3-4130 vastly superior, it also comes at a cheaper price for the entire system and uses a lot less energy. For me the entire Bulldozer Architecture is just a big fail. I don't care if it's because they are stuck with GloFo or if the architecture itself is to blame (and according to AMDs slides themselves they know where the problem is, but seems to take one step each year for each revision, instead of just doing it all at once and releasing a Bulldozer revision as it should have been), what matters is that I can't recommend the chip at all, when compared to a similar Intel chip.

At idle and low load scenarios you probably won't see much difference in power consumption at the wall with a full system. I remember,an ancient Llano A8 based system I built on a tight budget for someone with one of the 400W FSP OEM PSU,and it consumed between 34W to 47W at the wall during web browsing and watching videos,and it was using an mATX motherboard and that PSU was not the most efficient at low loads. I have built 20+ mini-ITX and Shuttle SFF PC builds in the last 7 to 8 years,and people always seem to look at artificial scenarios for power consumption. I have done both Intel and AMD based ones.

Even things like video encoding are peak operations and CPUs like the Core i3 and A10 are not really the target anyway for people doing much of it anyway,ie,you would use a Core i5 and the like anyway. When it comes IGP performance the AMD CPUs are faster in games even if more power is consumed.

The thing is though for web browsing tasks anything over 2GHZ Core2 level performance is more than enough. A Q6600 is more than enough for most web browsing and office purposes nowadays,as long as the IGP is capable of video decode and flash acceleration and this is why you are seeing more and more laptop sales being lost to lower performance tablets. People are sticking with their laptops longer,and now you are seeing tablet sales figures being added to laptop ones IIRC,as the lines get blurred.

I could argue that even a cheap Celeron dual core(or AMD equivalent) is probably enough for most basic purposes anyway.

Moreover,usage habit are the most important factor regarding power consumption. If you are the sort who just leaves their rig on 24/7 without doing anything,that should be your concern not a few watts here and there.

Nothing can beat a tablet for power consumption.

You want to save energy. Don't use a desktop.

Your monitor alone consumes 25+ watts. That will double your idle power draw of the entire desktop system.

That is more than netbooks and tablets under full load if you include the monitor.

PS:I live in Western Europe too and electricity is not cheap as the US.

 

[schmuckley]

Slide graphs are NOT screenshots. :whiste:
Just sayin'..
Screenshot with CPU-Z in it..That's what I'd call reliable.

 

[Abwx]

Slide graphs are NOT screenshots. :whiste:
Just sayin'..
Screenshot with CPU-Z in it..That's what I'd call reliable.

http://foro.noticias3d.com/vbulletin/showthread.php?t=421821

 

[USER8000]

It will be interesting to see how the single module Kaveri CPUs perform. The Richland and Trinity ones were meh,since once the second thread was used,the performance dropped due to the sharing penalty. I still expect the Celeron and Pentium dual cores to be faster,but OTH,the AMD IGPs should be better for basic compute purposes. It would be interesting to see how it pans out.

Sadly,hardly any review websites bother to compare the low end chips anymore.

 

[Slomo4shO]

http://wccftech.com/amd-kaveri-apu-...erformance-unveiled-mantle-45-faster-directx/

 

[ShintaiDK]

It will be interesting to see how the single module Kaveri CPUs perform. The Richland and Trinity ones were meh,since once the second thread was used,the performance dropped due to the sharing penalty. I still expect the Celeron and Pentium dual cores to be faster,but OTH,the AMD IGPs should be better for basic compute purposes. It would be interesting to see how it pans out.

Sadly,hardly any review websites bother to compare the low end chips anymore.

2M/4T always competed against dualcore Intels. 1M/2T is born hopeless. 1 FPU etc.

 

[sniffin]

From a pure revenue point of view, Enduro should probably be their #1 but that's been obvious for a while now and it doesn't look it has been. Their lack of progress with Enduro has cost them a huge marketshare in mobiles. Yes, there are some differences with Kepler vs GCN but as W1zzard's perf/watt charts over at TPU show Cape Verde, Pitcairn score very close to their Kepler equivalents (Tahiti XT GHz edition and 90°C running Hawaii are mostly an exception but enthusiasts on forums do tend to focus on the very high-end) so I imagine Enduro vs Optimus is the main factor in their marketshare loss.

Enduro is mostly fixed now, it just took them a stupidly long time to do it. They released the 7970M, the perfect mobile gpu (small, fast and extremely efficient) but literally sat there doing nothing for 6 months while the software side completely ruined the product.

They fixed it mostly with 13.1, but the damage has been done. The Enduro name has a stigma attached to it, the same kind of stigma thats stayed with AMD drivers in general, even if they have been quite good for some time now (desktop side).

If I was Rory Read I'd have scrapped and rebuilt the driver team. They are about as incompetent as you can get and have cost AMD almost the entire notebook market, and now Nvidia has essentially free reign.

 

[inf64]

I'm really curious how async. crossfire will work in practice. They have built a special block for CF (CF XDMA) so in theory Kaveri should work better when coupled with some similar GCN dGPU products.

 

[PPB]

I'm really curious how async. crossfire will work in practice. They have built a special block for CF (CF XDMA) so in theory Kaveri should work better when coupled with some similar GCN dGPU products.

It would be a interesting development if only 1 part of the equation needed the XDMA block to make bridgeless, correctly frame-paced crossfire. Because only 260x and 290/x will have that XDMA block to begin with.

 

[AtenRa]

Those numbers for the 7750 look awfully low for GDDR5. In any case, kaveri even with fast ram and highly overclocked is not adequate at 1080p.

With the settings the reviewer is using even the HD7750 is not adequate for 1080p(less than 30fps in most of the above benchmarks).

 

[JDG1980]

Those numbers for the 7750 look awfully low for GDDR5. In any case, kaveri even with fast ram and highly overclocked is not adequate at 1080p.

A discrete 7750 is not adequate at 1080p in the vast majority of modern "AAA" titles. You need a 7850 or better for that.

 

[Gikaseixas]

With the settings the reviewer is using even the HD7750 is not adequate for 1080p(less than 30fps in most of the above benchmarks).

I noticed that when the 7750 is above 30fps so is the APU in those games. I would say this is quite a nice jump on GPU performance.

 

[SlowSpyder]

It is probably worth noting that DDR3 2400 costs more or less the same price as DDR3 1333/1600. I'm looking forward to one of these APU's connected to my living room TV for some light gaming/media server responsibilities.

 

[Slomo4shO]

It is probably worth noting that DDR3 2400 costs more or less the same price as DDR3 1333/1600.

Yes, 2400 seems to be a ~5-15% premium over 1600, definitely worth the extra costs.

 

[inf64]

It is probably worth noting that DDR3 2400 costs more or less the same price as DDR3 1333/1600. I'm looking forward to one of these APU's connected to my living room TV for some light gaming/media server responsibilities.

Yup, the price of 2400Mhz memory has dropped a lot and now it doesn't carry such a premium over 1600Mhz memory modules. If there is a thing Kaveri needs it's moar memory BW.

 

[cbn]

It is probably worth noting that DDR3 2400 costs more or less the same price as DDR3 1333/1600. I'm looking forward to one of these APU's connected to my living room TV for some light gaming/media server responsibilities.

That is true for the 2 x 4 GB kits, but on the 2 x 2 GB kits 2400 is not available and the 1866 carries a significant price premium in most cases.

 

[AtenRa]

Prints here out off Germany. I assume AMD will start selling the A10-7000 series at the same price or a slight premium:

A10-6800K: 125€ + 100€ for 8GB DDR3-2666 + 90 € FM2+ mITX Mainboard
A10-6700K: 107€ + 60€ for 8GB DDR3-1600 + 90€ for same Mainboard + 105€ for R7 260X
i3-4130: 100€ + 60€ for 8 GB DDR3-1600 + 80€ for a mITX Mainboard + 105€ for R7 260X

What actually shocked me is the fact that I only have 4 mITX mainboards on FM2+ to choose from on one of Germany's biggest and cheapest reseller. On socket 1550 I can go down as low as 60€, but I wanted the features to be at least a little bit more comparable. The Intel solution comes out cheaper, uses less power, while paired with a discrete chip is even faster on GPU workloads and the CPU is almost always faster in CPU bound tasks anyway.

Since we are going to use a dGPU and you only used the lowest Core i3 Haswell, why did you chose the 6800K and not the A8-6600K that is also a unlocked Quad core at a price of only 86,64 euros at Hardware Versand ???

Lets see the prices. (only available models)

A8-6600K = 86,64 Euros
MSI FM2-A75IA-E53 = 72,94 Euros

Total = 159,58 Euros

FM2+ X88 with tones of features and Kaveri upgradability.
ASRock FM2A88X-ITX = 90,47 Euros
A8-6600K = 86,64 Euros

Total = 177, 11 Euros

Cheapest Core i3 Haswell.
Core i3 4310 = 101,73 Euros
MSI H81I = 65,57

Total = 167,3 Euros

So, the AMD APU is cheapest than the cheaper core i3 + H81. The APU can be OCed, it has more features and faster iGPU if you will use it.
For 10 Euros more over the cheapest Core i3 4310, you get the top of the line mini-iTX X88 motherboard that its also upgradable to Kaveri and it has tones more features than the H81.

But lets wait and see the Kaveri reviews and how much it will gain from CF with R7 250/HD7750.

 

[KompuKare]

It would be a interesting development if only 1 part of the equation needed the XDMA block to make bridgeless, correctly frame-paced crossfire. Because only 260x and 290/x will have that XDMA block to begin with.

The XDMA block is confirmed in Bonaire? What about Oland? Either way, that leaves A10-7850K without a matching part since Oland only goes up to 384 shaders and Bonaire only starts at 768. Or they have to come up with genius moment to make an asymmetric actually perform consistently well.

Of course, if PCs were a fixed target and AMD had the money to write their own engine they could design something which could offload some work to the onboard and a dGPU for the main graphics. But that applies to Intel's iGPU as well (I for one am tired of all those posts from people who have bought an Intel i5/i7 and are constantly complaining about the die space 'wasted' on HD3000/4000/4600 etc.). AMD are the only ones who have a presence in both CPUs and GPUs but they cannot afford to try something so risky while Intel have (near) infinite funds but reason to help owners of GPUs (even PCI-E is something they might get rid of on consumer hardware as soon as they can).

If I was Rory Read I'd have scrapped and rebuilt the driver team. They are about as incompetent as you can get and have cost AMD almost the entire notebook market, and now Nvidia has essentially free reign.

Well, I would imagine that it has a lot do with talking stuff so far in advance and then not dedicating enough resources to get that stuff delivered. So probably either some manager for hardware has unrealistic expectation about how long these things take or a manager for drivers likes to promise stuff with unrealistic deadlines. Could be either or even both.

 

[SlowSpyder]

That is true for the 2 x 4 GB kits, but on the 2 x 2 GB kits 2400 is not available and the 1866 carries a significant price premium in most cases.

Yea, but I doubt too many people are buying these systems with 4GB of ram. Remember, you have to share your video and system memory needs from that 4GB pool. I suppose there could be some people that is an issue for, but I doubt it is a huge percentage of people who are considering these parts for a build/new PC (at least not the desktop parts).

 

[Galatian]

Sorry you guys keep posting and posting the same AMD marketing slides over and over again without actually taking the time to take a closer look.

Open CL perf on the slide bottom :

Although it's hard to tell it looks like they are using the Accelerated Home Benchmark on PCMark 8. I'm not sure what they did, but the Core i5-4560K performs much worse in their test, then actual real world examples:

http://foro.noticias3d.com/vbulletin/showthread.php?t=421821

Ok your slides show a nice improvement over Richland. Compare to results from Intel...

This is from my Core i5-3570K with HD4000:

It's not even the newer HD4600 or above. I'd say they'll probably be on par with the A10-7xxx

At idle and low load scenarios you probably won't see much difference in power consumption at the wall with a full system. I remember,an ancient Llano A8 based system I built on a tight budget for someone with one of the 400W FSP OEM PSU,and it consumed between 34W to 47W at the wall during web browsing and watching videos,and it was using an mATX motherboard and that PSU was not the most efficient at low loads. I have built 20+ mini-ITX and Shuttle SFF PC builds in the last 7 to 8 years,and people always seem to look at artificial scenarios for power consumption. I have done both Intel and AMD based ones.

I suggest you take a look at the following graphs:

I guess you are completely right about the idle power usage. But you could further decrease this by using T models. Now the Core i3-4130T (35W TDP) only runs 5€ more here in Germany (105€). And still the difference is fairly big in my eyes. Also note that the Core i3 - in most cases - idles much faster whenever you do something, since it's simply faster:

Even things like video encoding are peak operations and CPUs like the Core i3 and A10 are not really the target anyway for people doing much of it anyway,ie,you would use a Core i5 and the like anyway. When it comes IGP performance the AMD CPUs are faster in games even if more power is consumed.

One word: Quick Sync. Seriously try it out. It's fast and doesn't use a lot of power. It's one big edge Intel CPUs have, even for your everyday Office Built.

Also - and I keep asking that question until I get a satisfying answer - what exactly is the point of a faster IGP on a desktop, when it's not really faster then a cheaper CPU + discrete chip? You can always get the same OpenCL performance enhancements from using a discrete chip. In fact as I have shown the HD4600 and HD4000 OpenCL performance is quite good, while actually consuming a lot less power. Where powerful IGP shine is mobile. And this is where it get's really blurry really fast. I can't find many benchmarks of the A10-5757M with Radeon HD-8650G and a comparable Core i3-4158U with Iris 5100. Notebookcheck for example show the Iris 5100 being only 2% slower then the top model from AMD, while actually having a lower TDP. In my eyes Intels way with the HD line of IGPs is to be as efficient as possible and no brute force the performance crown. It works out for them. Again this is exactly my point: I think AMD is simply missing the point with their APUs. It seems like they acquired ATI and just had to figure out how to use their IP. Fusion is nice if it is efficient. But their CPU designs simply sucks and a good IGP is only really relevant on mobile, where Intel is actually in my book leading when you take efficiency into account.

Since we are going to use a dGPU and you only used the lowest Core i3 Haswell, why did you chose the 6800K and not the A8-6600K that is also a unlocked Quad core at a price of only 86,64 euros at Hardware Versand ???

...snip...

Are you kidding me? You asked:

Also,

How much it cost to get the Core i3 + HD7750 ???
How much it cost to get the A10-7850K + DDR3-2666MHz ???
How much it cost to get the A10-7700K + HD7750 ???

Don't ask rhetorical questions when you're not sure about the outcome :whiste:

And look at it the other way: taking all the Benchmarks into account not even the most expensive AMD APU can beat the cheapest Intel Core i3...

 

[DaZeeMan]

My man, you are grasping at straws here. iGPUs are already choked for bandwidth and you want it to use DDR3-1600? What's wrong with rebates? Finally A10-6800K isn't priced at $173... it's a place holder because the site doesn't have unreleased APUs.

My point in bringing this up is that if the i3 system and the Kaveri system are similarly priced, without a iGPU, well then if you can add the dGPU to one you'll be adding it to the other as well, at the same price point.

Hence why my A10-7850K versus i3 comparison becomes relevant. As I noted above, the Cyberpower PC i3-4340 system comes in at $531, with only the integrated GPU, versus the A10-7850K system, at $528. At that point, it comes down to how well the HD4600 integrated graphics fare against the R7 integrated graphics of the A10-7850.

The i3 does seem to do better with a dGPU than the A10-6800K, according to some benches I found, but as noted in the benches above, the A10-7850K has increased performance over the A10-6800K, which reduce the i3-4340's lead, and in some cases it should achieve parity or exceed the i3, depending on the bench.

However, a lot of entry level consumers at this price point may not need/be willing to pay for a dGPU, as the Kaveri or i3 already have the integrated GPU which is sufficient for their needs.

At that point, it comes down to what other features the Kaveri brings to the table vs. the i3. If asymmetric crossfire for Kaveri is implemented properly (which is a Mantle feature apparently, according to one of the AMD slides), that could also change things... of course that's an unknown for now though.

So for enthusiasts, well y'all are going to to the dGPU route anyways, so the i3 may look better. But for entry level consumers, the Kaveri hits a sweet spot that the i3 doesn't quite fill, due to the superior integrated graphics of the Kaveri vs the i3.

Serious AMD enthusiasts will still probably go the FX route in any case... at least for desktops. And you Intel guys will continue to do your thing with your pricier i5's/i7's. Unless Mantle ends up being a game changer for Kaveri of course (we shall see)...