-
We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.
PileDriver Performance only 10% -15% (maybe)
Page 4 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
frostedflakes
Diamond Member
They did die shrink Thuban, it's called Stars and is what they used in Llano.
Clearly they thought Piledriver offered better perf/watt or whatever they want from a CPU architecture, otherwise they would have just stuck with Stars.
The idea behind Bulldozer makes sense if you consider the direction AMD is heading with GPU compute. The thinking is that all AMD APUs have a very powerful floating point processor built in (the GPU), so a beefy FPU on the CPU is a waste of die space. That's why Bulldozer shares FP resources but each core has dedicated integer resources, in the future AMD sees the processor being used less for floating point calculations and the GPU being used more. If the GPU compute thing catches on it makes sense for the future. Unfortunately it makes little sense currently since the vast majority of floating point calculations are still done by the CPU. Between the greater focus on integer performance over floating point, the architecture being designed to scale well with additional cores, and the architecture being designed to scale well with additional clock speed as process technology allows, AMD is well positioned for the future with Bulldozer (and its derivatives) IMO. But that also means it's a kind of "meh" architecture currently, it's a bit too forward thinking in some ways and hinges on some ifs, such as GPU compute panning out and becoming very popular for workloads that benefit from it.
Clearly they thought Piledriver offered better perf/watt or whatever they want from a CPU architecture, otherwise they would have just stuck with Stars.
The idea behind Bulldozer makes sense if you consider the direction AMD is heading with GPU compute. The thinking is that all AMD APUs have a very powerful floating point processor built in (the GPU), so a beefy FPU on the CPU is a waste of die space. That's why Bulldozer shares FP resources but each core has dedicated integer resources, in the future AMD sees the processor being used less for floating point calculations and the GPU being used more. If the GPU compute thing catches on it makes sense for the future. Unfortunately it makes little sense currently since the vast majority of floating point calculations are still done by the CPU. Between the greater focus on integer performance over floating point, the architecture being designed to scale well with additional cores, and the architecture being designed to scale well with additional clock speed as process technology allows, AMD is well positioned for the future with Bulldozer (and its derivatives) IMO. But that also means it's a kind of "meh" architecture currently, it's a bit too forward thinking in some ways and hinges on some ifs, such as GPU compute panning out and becoming very popular for workloads that benefit from it.
frostedflakes
Diamond Member
Better performance per watt than Bulldozer at least, seems like AMD has managed to improve perf/watt quite a bit with Piledriver. I was kind of skeptical they could improve it 10-15% in one year as they had planned in their roadmap, but between the resonant clock mesh and other improvements looks like they've not only met but have maybe even exceeded this goal, suspect it might end up being 20-25% better than Bulldozer.
Still won't be able to come close to Intel in perf/watt, especially with Ivy Bridge on the way, but at least Piledriver's perf/watt should hopefully be a lot less embarrassing than Bulldozer's was.
Still won't be able to come close to Intel in perf/watt, especially with Ivy Bridge on the way, but at least Piledriver's perf/watt should hopefully be a lot less embarrassing than Bulldozer's was.
IntelUser2000
Elite Member
Sandy Bridge's GPU takes up 38mm2 in 32nm process. Ivy Bridge is similar but at 22nm. The 20EU version of Haswell is at ~60mm2, and the 40EU version looks to be around 100mm2.
Also, don't count out Ivy Bridge yet. There may yet be a positive side for Ivy Bridge coming close to May rather than January. Same time last year the new driver for Sandy Bridge that brought nice gains were released: http://www.intel.com/performance/desktop/2ndgencore/hdgraphics/index.htm
At least on the 3770K benchmark, its faster than Llano: http://www.anandtech.com/show/5626/ivy-bridge-preview-core-i7-3770k/16
Last edited:
I'd suggest you read that article properly before you bring it up to prove a point.
Seems to me the difference in cpu performance is about the same as the difference in igpu performance, i.e. they are trailing eachother by a generation or so.
A very good situation for consumers in that area. Although bad for me waiting for desktop Piledriver, since AMD has to spend so many resources maintaining it's igpu lead which can't be spent awesomizing the cpu.
Edit: lol you were only talking about GPGPU, I'll take my own advice & read properly in the future 😉
Last edited:
Sadly by that time Trinity should start to trickle out to retailers. What's worse, I hadn't read the preview on Anandtech before and was kinda considering the iGPU on Ivy to be within 5% of Llano. Seeing that the gap was actually 20-50% was a bit of a shock to say the least. 10% is a reasonable performance delta over a generation due to better drivers, but even that wouldn't bring parity. Also, it confirms my fear that Ivy Bridge can't keep up in image quality. First time in my life I feel disappointed in a hardware manufacturer, because I hoped for a nice competition when I'm shopping for a laptop this summer.
Actually, that's the only GPGPU benchmark that I could find of Ivy Bridge's iGPU. Considering that OpenCL support on the GPU was one of the touted features, I would hope to see more benches when the parts launch.
For me that's a horrid situation, because I've never needed more CPU power than a decent Core 2 Duo, the GPU is always the bottleneck on laptops. I don't care that archives extract 20 seconds longer, when the iGPU can't push 30 fps on anything above medium.
Last edited:
IntelUser2000
Elite Member
Richaron: Nah, I think you should read my reply properly instead. My response was to Arzachel's comment that GPGPU is far behind with Ivy Bridge, when it doesn't show that in Anand's bench.
Last edited:
Editing as you posted, my bad. MOVE ALONG KIDS NOTHING TO SEE HERE
http://www.xbitlabs.com/news/cpu/di...nity_Fusion_APUs_for_Notebooks_on_May_15.html
If it's really 25% faster/better than Bulldozer it might be ok.
If it's really 25% faster/better than Bulldozer it might be ok.
gmaster456
Golden Member
Oh noezzzzzzzzz AMD sukz sooo much lolzzzzzzzzz.
They state that the 35W Trinity is up to 25% better in CPU performance than the 35W Llano - it is not about Piledriver vs Bulldozer.
Dresdenboy
Golden Member
Thanks ;-) So you've seen secret bench results? ^^
formulav8
Diamond Member
Well Trinity mobile is rumored to be May 18th or something like that. FX will be quite a bit later than that unfortunately.
Plus, they could have easily been working on PD for quite awhile now. They've had working BD samples for along time before they were released. AMD knew from the start that there were tweaks that need to be done. So PD coming this year isn't to soon or anything like that.
Plus, they could have easily been working on PD for quite awhile now. They've had working BD samples for along time before they were released. AMD knew from the start that there were tweaks that need to be done. So PD coming this year isn't to soon or anything like that.
This point gets overlooked all too often - is it any surprise that Llano performs so much better when its iGPU uses over twice the die size of SNB? What is surprising is that SNB performance is pretty close to Llano in terms of area efficiency given how it compares to the A4 line.
While it's certainly not proof of any pending performance gains, it's already known that there's a driver update coming either with the 7 series chipset launch or IVB - http://downloadcenter.intel.com/Det...vers+.&OSFullname=Windows+7,+64-bit*&lang=eng
Despite the improvement in Ivy's GPU I don't think anyone expected it to be a desktop video card replacement. Where Ivy will shine is in the notebook area. Remember the HD 4000 GPU in the desktop Ivy will be virtually identical to the HD 4000 GPU on the mobile version. That means the GPU in a 17W Ivy should perform the same as the 3770K Ivy that Anand tested (assuming its not CPU limited).
While Llano desktop maybe faster then Ivy that may not transfer over to the notebook segment. The A8-3870K desktop Llano that Anand used to compare to Ivy has 400 Shaders and 600 MHz speed while the top of the line A8 mobile Llano only has 444 MHz speed. If you factor that in the mobile Ivy should perform about the same the mobile A8 Llano and beat the mobile A6 and A4 Llanos that have 320 and 240 shaders.
How exactly is it surprising? SB is missing a ton of features and is made on a mature 32nm process compared to Llano, which is made on a new process node for both AMD and GF that was plagued with a ton of issues.
Only if you expect the HD4000 to perform as well at ~500mhz as it does at ~1000mhz. Conveniently enough, a 320 shader part at 444mhz has around half the performance of a 400 shader part at 600mhz.
Unless I'm mistaken Trinity will be out by the end of Q2, it's already in production, and new FX chips with the Piledriver core are scheduled for Q3.
Yes, SB is lacking a few features in terms of what's actually enabled/fully implemented, but how much does that actually impact die size?
Because Intel's 17W parts have that much lower of graphics turbo frequency? Yeah, I guess they do - 1.2GHz instead of 1.35GHz. Yes, they also have some 17W skus that have lesser graphics turbo frequencies, with the lowest I see being 800MHz.
Saying SB is lacking a few features might be a slight exaggeration. UVD and a tessellator already take up non-zero die space, and SB doesn't support neither OpenCL nor Direct Compute using the iGPU. ~3 times smaller for 40% of the performance with worse IQ and less features doesn't seem that stellar to me, diminishing returns considered.
I'll be blunt, gpu turbo frequencies are totally meaningless for a 17w sku, even if you enjoy gaming in a freezer.
I'll give you tessellation seeing as how it's not implemented at all in SNB, but neither OpenCL nor 'UVD' can be counted against SNB die space. OpenCL is more a matter of a few bug fixes than actual die size, and last I checked the SNB decode was on par with both AMD and NVIDIA offerings. As for IQ, no question that it's behind there due to its lack of proper anisotropic filtering, but that's effectively fixed with IVB and again didn't affect area much as all the logic was present already, just not working properly.
It is? That must by why according to notebookcheck.net the 2557m in a macbook air gets 1360 on 3dmark vantage P GPU while the 2620m in a macbook pro gets 1477... Oh wait, that's pretty much exactly the difference between their 1.2GHz vs 1.3GHz turbo speeds.
cytg111
Lifer
- Maybe the "rumored" ps4 deal will be a step in that direction, who knows.
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes + WCL Discussion Threads
- Started by Tigerick
- Replies: 25K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 24K
-
-
