kaveri bf4 good performance?

[NTMBK]

Just one question guys: It's clear that Kaverid didn't reach the intended clocks, but it seems that the GPU bore the brunt of the cuts, 10% of the CPU against 20% of the GPU. What are your guesses for that?

Not convinced that we've seen the full story on Kaveri clocks. Remember that Hawaii has similar base clocks, but boosts up to 1GHz:

In essence, it appears that AMD has a base clock rate around 727 MHz with its R9 290X, though the Hawaii GPU wants to run as close to 1000 MHz as possible

http://www.tomshardware.com/reviews/radeon-r9-290-review-benchmark,3659.html

Kaveri is based on the same GPU cores as Hawaii, and if anything boost should be even more important because APUs need to dynamically balance TDP usage between CPU and GPU portions. So I'm waiting for the official announcements and full clarification of clock speeds before I tear down my Jim Keller poster.

 

[ViRGE]

Thread cleansed and morons infracted. Let's keep it on topic.
-ViRGE

 

[NTMBK]

Wait, I thought this got moved to GPUs?

After initially moving it over there, at 3 pages in I decided it would be too disruptive at that point, and moved it back
-ViRGE

 

[Blitzvogel]

Obviously AMD is hoping HSA and hUMA bares some of the brunt of increasing bandwidth demands. It's silly of me, but I'm still hoping to see a 768+ SP/quad channel variant of Kaveri for desktop and server soon. Assuming thermals were not too high it would make sense as a Steambox APU of which there would be no competing Intel product by a long shot if it was $200, with close to 4.0 GHz on the main cores and 1000 MHz on the SIMDs. It would replace my current setup no problem when it comes to modern DX11 titles.

I think AMD has lost sight of how valuable a market the desktop really is. Builders, both OEMs and people, I think will jump onto the Steambox bandwagon and there is of course the HTPC market that always values playback performance and accuracy assuming the cost is low and thermal envelopes are decent. Kaveri as we know it better be able to push 1 GHz on it's graphics in desktop versions.

Along with Mantle, we must wait to see how HSA and it's associated software takes off and I think game developers are too much an important part of this with the PS4 and Xbox One being semi-HSA, and the trickle down and up effects of multiplatform development. Thankfully AMD is extremely involved with so many game and software developers.

While I deem it unlikely, it's certainly possible that an AMD APU + dedicated GPU could be both the cost & performance optimized set up for future PC games versus an expensive brute force 8 core i7 machine. But Intel is a monster, and their high end is beneficial to any user regardless of what they use it for and that is available right now.

 

[Xpage]

unless Kaveri has massive internal bandwidth optimizations,no way, they really need DDR4. AMD should have scooped up Sandforce for their realtime data compression schemes and use that somehow for GPU frame pushing

 

[ViRGE]

unless Kaveri has massive internal bandwidth optimizations,no way, they really need DDR4. AMD should have scooped up Sandforce for their realtime data compression schemes and use that somehow for GPU frame pushing

They also have compression in the GPU. Besides the fixed ratio texture compression, compression is available to the ROPs for both Z/Stencil ops and color ops.

 

[NTMBK]

unless Kaveri has massive internal bandwidth optimizations,no way, they really need DDR4. AMD should have scooped up Sandforce for their realtime data compression schemes and use that somehow for GPU frame pushing

Have a read of the Eurogamer interview with the XBox One architects, there's lots of interesting stuff about how they've added compressed texture support and DSPs for compressing/decompressing, to help with their low bandwidth (and fit things into eSRAM):

Some obvious things we had to deal with - a new configuration of memory, we couldn't really pass pointers from CPU to GPU so we really wanted to address that, heading towards GPGPU, compute shaders. Compression, we invested a lot in that so hence some of the Move Engines, which deal with a lot of the compression there... A lot of focus on GPU capabilities in terms of how that worked. And then really how do you allow the system services to grow over time without impacting title compatibility. The first title of the generation - how do you ensure that that works on the last console ever built while we value-enhance the system-side capabilities.

Digital Foundry: Obviously though, you are limited to just 32MB of ESRAM. Potentially you could be looking at say, four 1080p render targets, 32 bits per pixel, 32 bits of depth - that's 48MB straight away. So are you saying that you can effectively separate render targets so that some live in DDR3 and the crucial high-bandwidth ones reside in ESRAM?

Andrew Goossen: Oh, absolutely. And you can even make it so that portions of your render target that have very little overdraw... For example, if you're doing a racing game and your sky has very little overdraw, you could stick those subsets of your resources into DDR to improve ESRAM utilisation. On the GPU we added some compressed render target formats like our 6e4 [six bit mantissa and four bits exponent per component] and 7e3 HDR float formats [where the 6e4 formats] that were very, very popular on Xbox 360, which instead of doing a 16-bit float per component 64pp render target, you can do the equivalent with us using 32 bits - so we did a lot of focus on really maximizing efficiency and utilisation of that ESRAM.

Nick Baker: From a power/efficiency standpoint as well, fixed functions are more power-friendly on fixed function units. We put data compression on there as well, so we have LZ compression/decompression and also motion JPEG decode which helps with Kinect. So there's a lot more than to the Data Move Engines than moving from one block of memory to another.

http://www.eurogamer.net/articles/digitalfoundry-the-complete-xbox-one-interview

It would be nice to see some of that tech show up in Kaveri, but I doubt it would see any use outside of Mantle enabled engines.

 

[Arkadrel]

It's silly of me, but I'm still hoping to see a 768+ SP/quad channel variant of Kaveri for desktop and server soon.

You are not alone.
AMD should have had quad channel APUs along time ago, esp with kavari when their makeing the igpu portions this powerfull.

I can understand the "mobile" parts dont work well with quad channel, but screw that, what about the desktop market?

Dont AMD want to have a APU that can finally wow someone? I believe a Kavari with quad channel memeory would be such.

I think AMD has lost sight of how valuable a market the desktop really is.

Again I agree. APUs will matter in the desktop markets if their made properly, which means not being bandwidth starved.


If only AMD had kavari with Quad channel DDR3-2133 support, so we could see some ~68.3 GB/s memory bandwidth for these APUs.

 

[R0H1T]

You are not alone.
AMD should have had quad channel APUs along time ago, esp with kavari when their makeing the igpu portions this powerfull.

I can understand the "mobile" parts dont work well with quad channel, but screw that, what about the desktop market?

Dont AMD want to have a APU that can finally wow someone? I believe a Kavari with quad channel memeory would be such.



Again I agree. APUs will matter in the desktop markets if their made properly, which means not being bandwidth starved.


If only AMD had kavari with Quad channel DDR3-2133 support, so we could see some ~68.3 GB/s memory bandwidth for these APUs.

The thing with DDR4 probably makes it untenable & undesirable for Kaveri to have quad channel DDR3 memory especially since DDR4 is being commercially released this year & sometime next year we'll see desktops being laced with'em, assuming Haswell E is released on schedule.

 

[ShintaiDK]

The thing with DDR4 probably makes it untenable & undesirable for Kaveri to have quad channel DDR3 memory especially since DDR4 is being commercially released this year & sometime next year we'll see desktops being laced with'em, assuming Haswell E is released on schedule.

The only desktop with DDR4 you see in 2014 is Haswell-E. And Kaveri is not DDR4 compatible. Mainstream DDR4 usage will first start in the end of 2015.

 

[R0H1T]

The only desktop with DDR4 you see in 2014 is Haswell-E. And Kaveri is not DDR4 compatible. Mainstream DDR4 usage will first start in the end of 2015.

Not yet but who's to say that Kaveri version 2.0 won't have DDR4 support & historically AMD have done this sort of a thing in the past as well with the Phenoms being prime example.

 

[ShintaiDK]

Not yet but who's to say that Kaveri version 2.0 won't have DDR4 support & historically AMD have done this sort of a thing in the past as well with the Phenoms being prime example.

Kaveri 2.0 on what? FM3? In what year? 2015? Seems this Kaveri 2.0 is an imaginary product.

 

[CHADBOGA]

Sorry if I have missed it, but in percentage terms, what kind of increase will we see Kaveri bring over Richland for:

GPU - More or less than 30%

CPU - More or less than 15%

 

[R0H1T]

Kaveri 2.0 on what? FM3? In what year? 2015? Seems this Kaveri 2.0 is an imaginary product.

Its just speculation on my part but I do expect AMD to add DDR4 support for their APU's as soon as its economically viable for them btw do you want progress or not from AMD cause it seems like you want them to be relegated to the second tier perpetually!

 

[ShintaiDK]

Its just speculation on my part but I do expect AMD to add DDR4 support for their APU's as soon as its economically viable for them btw do you want progress or not from AMD cause it seems like you want them to be relegated to the second tier perpetually!

The reason why Haswell-E/EP/EX will be the only DDR4 parts in 2014 is due to price and volume. We are deep into 2015 before you can add it to the mainstream without having to pay 25$ per GB or something.

 

[R0H1T]

Sorry if I have missed it, but in percentage terms, what kind of increase will we see Kaveri bring over Richland for:

GPU - More or less than 30%

CPU - More or less than 15%

Its actually more in the range of 40~50% with HSA, HUMA & not to forget Mantle :sneaky:

/jk

 

[R0H1T]

The reason why Haswell-E/EP/EX will be the only DDR4 parts in 2014 is due to price and volume. We are deep into 2015 before you can add it to the mainstream without having to pay 25$ per GB or something.

I doubt it'll be that high, especially since mass production starts next month, but it won't stop DRAM makers from ripping customers off with their early adopter premium BS & what not

 

[ShintaiDK]

I doubt it'll be that high, especially since mass production starts next month, but it won't stop DRAM makers from ripping customers off with their early adopter premium BS & what not

Volume to meet demand? Or just volume? If you only make 10mio and demand is 20mio. Price will skyrocket.

 

[R0H1T]

Volume to meet demand? Or just volume? If you only make 10mio and demand is 20mio. Price will skyrocket.

That's anyone's guess because DRAM makers in the past have colluded to control production & raise prices however it isn't necessarily an indication of what's to happen in the future with DDR4.

 

[Arkadrel]

Kaveri 2.0 on what? FM3? In what year? 2015? Seems this Kaveri 2.0 is an imaginary product.

^ shintai has a point.

Which is why its beyound me, that Kavari doesnt come in its current version with quad memory channel support.

 

[R0H1T]

^ shintai has a point.

Which is why its beyound me, that Kavari doesnt come in its current version with quad memory channel support.

Probably because AMD didn't feel the need to improve their memory controller & were hoping HSA, HUMA & eventually DDR4 would bail them out on this front, basically taking a shortcut & saving money on R&D I guess :\

 

[ShintaiDK]

Which is why its beyound me, that Kavari doesnt come in its current version with quad memory channel support.

Cost on the desktop. Cost and size on the mobile.

 

[Erenhardt]

Who in their right mind would buy APU for budged gaming and then shell for quad memory kit? That way you are paying more for apu system than you would pay for cpu+gpu.
AMD is going right way. They don't follow usual PC path: Throw more hardware at the problem! They are trying to solve the problem at its roots. hUMA can save some memory bandwidth and mantle can optimize memory usage by queuing and spreading memory operations in time.
What I would like to see is Kaveri+mantle+7750 in BF4
APU is budged. When DDR4 become budged option, everyone will talk about DDR5

 

[R0H1T]

Who in their right mind would buy APU for budged gaming and then shell for quad memory kit? That way you are paying more for apu system than you would pay for cpu+gpu.
AMD is going right way. They don't follow usual PC path: Throw more hardware at the problem! They are trying to solve the problem at its roots. hUMA can save some memory bandwidth and mantle can optimize memory usage by queuing and spreading memory operations in time.
What I would like to see is Kaveri+mantle+7750 in BF4
APU is budged. When DDR4 become budged option, everyone will talk about DDR5

Now about the quad channel memory, I've seen reviews that show quad channel doesn't help mitigate the problem of memory bandwidth that much over dual channel or triple channel systems ~

The gains are in the region of 60~70% theoretically but the cost of quad channel memory kits & an improved MC would make it economically unviable for low/mid range systems based on AMD's current APU lineup !

 

[ShintaiDK]

Also remember more pins, traces etc for quadchannel.