User AtenRa has made some interesting multi-CPU Thief benchmarking
With DX, the 8350 is behind the i3 4330 in this poorly-threaded game. With MANTLE the 8350 catch-up the i3 on average FPS (outperforming the i3 in min. FPS) and is immediately behind the i7 3770.
Does anyone think that the GPU race is getting a bit out of hand? The hardware and software advancements are being made at completely different paces. Sure the software is getting harder to process, however, the 700 series only came out about 5-6 months ago (If I'm correct), and Nvidia is already seeking to it's next gen, even though their GTX 760 can do great at high resolutions and high qualities. Same goes for AMD with their 200 / 200x series. An R9 270x can do about the same performance as a GTX 760.
Of course each engine is different, but many of these companies have been making software that, sure is harder to process, but for each new release, the amount of new-gen GPUs that can process it well it getting greater and greater. That being said, I don't see why Nvidia and/or AMD hold the reigns to spend more money on development of a great card, rather than a set of cards that are 15-20% better than the last gen, so that way then can pump out cards (assuming the pricing isn't at Titan Z levels), that are a little cheaper but then can pump out much more performance than the previous gen.
I don't see the problem. Instead upgrading each gen, just do it each two or three gens. From Nvidia/AMD position incremental upgrades in technology are easier and cheaper than huge step upgrades. You add X, test it, developers learn to use it, then add Y, test it, developers learn to use it...
Yep. I typically build a PC to last 4 years. My setup at launch:
i7 2600k, NVIDIA 570 GTX, 8GB DDR3
Two years later?
i7 2600k, NVIDIA 770 GTX, 8GB DDR3
Two years from now? Likely will be whatever Intel releases in two gens, as I scrap the entire thing and start over. Cheaper to build around an over-speced CPU and forget about it for the next 3-4 years.
Honestly though, if my QX9650 wasn't built around such a crappy platform (nForce 790i Ultra; wanted the DDR3 and PCI-E 2.0, and got burned badly), I would have stuck with that until Haswell and skipped SB/IB entirely. Intel rigs really should need upgrades that much if you build them right.
Development side? Yes, incremental upgrades are much cheaper. Use an old tech to test out the new node, use that now developed node to develop the new tech. There's a reason why both NVIDIA and Intel use the same approach now: It greatly lowers risk of a flop (NVIDIA 5000 series anyone?)
User AtenRa has made some interesting multi-CPU Thief benchmarking
With DX, the 8350 is behind the i3 4330 in this poorly-threaded game. With MANTLE the 8350 catch-up the i3 on average FPS (outperforming the i3 in min. FPS) and is immediately behind the i7 3770.
At last a proper review and it mimics my own experience with mantle ss bench, thanks juan. Why have no big review sites done similar?
To think this is the beginning, I wonder will we see hsa apus fully utilized in mantle (cpu+igp+dgpu) and top of the charts in future games.....
Does anyone think that the GPU race is getting a bit out of hand? The hardware and software advancements are being made at completely different paces. Sure the software is getting harder to process, however, the 700 series only came out about 5-6 months ago (If I'm correct), and Nvidia is already seeking to it's next gen, even though their GTX 760 can do great at high resolutions and high qualities. Same goes for AMD with their 200 / 200x series. An R9 270x can do about the same performance as a GTX 760.
Of course each engine is different, but many of these companies have been making software that, sure is harder to process, but for each new release, the amount of new-gen GPUs that can process it well it getting greater and greater. That being said, I don't see why Nvidia and/or AMD hold the reigns to spend more money on development of a great card, rather than a set of cards that are 15-20% better than the last gen, so that way then can pump out cards (assuming the pricing isn't at Titan Z levels), that are a little cheaper but then can pump out much more performance than the previous gen.
No it always has been this way. Quad cores are finally becoming more utilized yet have been around since at least 2007. Software always lags behind hardware and always will.
Games have not fully utilized GPUs and a HD7970/GTX680 will probably last even for the next 3 years at least before we really see a need to move on.
juanrga :
User AtenRa has made some interesting multi-CPU Thief benchmarking
With DX, the 8350 is behind the i3 4330 in this poorly-threaded game. With MANTLE the 8350 catch-up the i3 on average FPS (outperforming the i3 in min. FPS) and is immediately behind the i7 3770.
I don't think it is a threading issue. If it was then the i3 wouldn't also benefit from it. Instead Mantle is trying to bypass the CPU a bit more and utilize the GPU more same as DX12 would do.
I for one am not really impressed with it because Thief already runs very well overall and as well it only benefits specific setups enough which is just meh.
It's been things like that. Just look at the last few products AMD has released and compare it to something like FX 8150 launch.
The company no longer executes the way they used to. They have changed drastically. I don't see anyone accounting for this new AMD when making conjecture about the future.
During unending discussion in this thread about why AMD was not releasing a Steamroller FX CPU on SOI, I mentioned often that AMD had changed radically and even gave a quote from Raja Kodury. I give it again:
Raja returns to a very different AMD than the one he left. I asked him what’s different and he responded by saying the AMD he left acted like a company that was 10x its size. Today, AMD is a much smaller and more agile company. Raja believes AMD is in a better position to take advantage of new opportunities vs. being in the hopeless position of never being able to catch up in mature markets.
Yes, but I think the error is that you assume that AMD not directly competing with Intel means that they are going to abandon big cores.
A large problem with APU is that GPU and CPU want different things from the process they are made on. The most logical conclusion is to make separate dies for each, and use a CPU oriented process for the CPU and a GPU oriented process for the GPU.
But the problem is that Steamroller is on 28nm Bulk, which only offers a trade off between CPU and GPU. AMD could have either ported Steamroller to SOI (maybe 32nm SOI?) or released as 28nm Bulk for an SR FX chip.
I have said this before, but even if AMD wanted to release an SR FX chip, they have no good options to do so. SR FX would have been a huge disappointment on bulk with the reduced clock speeds.
Meanwhile, AMD is currently forced to keep everything on one die because there is no good way to get CPU and GPU talking to eachother without a ton of latency over long distances (I mean long distances in SOC terms, like across an ATX motherboard).
Their hands are tied. They are in an awkward position where the things they need aren't out yet.
I do think it's entirely possible for AMD to release a new HEDT platform that is focused around high performance while not directly competing with Intel.
"Directly competing with Intel" means trying to fight them directly with x86 CPU performance. If AMD released an HEDT HSA platform, it wouldn't be directly competing with Intel because it'd be an entirely different animal than just "lets look at these bar graphs and see the scores for SuperPI!"
It'd be sort of like what we're seeing with Mantle vs DX benchmarking right now. AMD has technology that their competition doesn't have access to and it gives them a massive advantage. Except imagine that happening with CPUs instead of GPUs.
I mean, I'd even go as far as to think that AMD could be pondering just ignoring traditional x86 performance entirely and instead pushing HSA on a future platform and instead pushing HSA enabled applications.
But, it's like Mantle on Linux, it has a massive problem with chicken and egg.
Who wants to buy hardware for HSA when there is no HSA software? Who wants to make software for HSA hardware when no one has HSA hardware?
Breaking that deadlock is key for AMD and they will do it somehow. I would not be surprised if out of the blue AMD released a new HEDT platform that's HSA enabled along side a ton of HSA applications and a massive game for Mantle in Linux that offers massive advantages over what you have in Windows (be it performance, features, etc). I also wouldn't be surprised if it ended up not being much of an improvement over Intel and Steamroller in traditional workloads, but absolutely mauled Intel and old AMDs when HSA is enabled.
If this was the AMD of FX 8150 and JF-AMD, I wouldn't think that that would have a chance of happening at all. But AMD has changed and they are ambiguous in the few things they say and they don't overly hype their products or anything anymore.
HSA is AMD's final push against Intel and Nvidia, and they've learned from the past. Why make something that can benefit their competition when they can gang up with other who have common competition and create something that leaves the competition out?
AMD completely blew amd64 instruction set, dual cores, quad cores, 3dnow, etc. HSA is a completely different approach and so is Mantle.
Mantle is a very curious case because AMD knows they talk about how Mantle is open and anyone can use it and they know Nvidia never will support it because they're too proud and too fixated on proprietary solutions they control.
HSA is meant to lock Intel out because AMD sees that their GPU advantage is shrinking and they know they can't depend on it forever. WIth HSA, even if Intel totally creams AMD CPU and AMD GPU, it won't matter. If it were OpenCL AMD would be in trouble, but not with HSA.
Locking out Intel and Nvidia is key, and having a ton of other partners is massive as well. Even VIA would be a competitor to Intel with HSA...
Yep. I typically build a PC to last 4 years. My setup at launch:
i7 2600k, NVIDIA 570 GTX, 8GB DDR3
Two years later?
i7 2600k, NVIDIA 770 GTX, 8GB DDR3
Two years from now? Likely will be whatever Intel releases in two gens, as I scrap the entire thing and start over. Cheaper to build around an over-speced CPU and forget about it for the next 3-4 years.
Honestly though, if my QX9650 wasn't built around such a crappy platform (nForce 790i Ultra; wanted the DDR3 and PCI-E 2.0, and got burned badly), I would have stuck with that until Haswell and skipped SB/IB entirely. Intel rigs really should need upgrades that much if you build them right.
I don't even feel like I need to upgrade my CPU anytime soon. You could probably get away without upgrading either.
I'm still on my Phenom II X3 and I really don't see the need to upgrade...
I can still play the latest (non-fps) games at medium/1080p; for me that's Diablo 3 RoS, Civ 5, XCOM
Apps wise ArcGIS is more IO bound that anything. That and ESRI hasn't bothered to make it that multi-threaded. 3 cores and good IPC counts for more than anything else. Probably why I haven't bought an AM3+ Piledriver CPU - it would be slower - heck we don't even know if ArcMap will be GPGPU accelerated anytime soon. I would love to see any kind of acceleration here for ArcMap. I wish Intel/AMD/Nvidia would spend some time investing in there - good multi-core support and GPU support really count towards a lot in reducing hangups in ArcMap.
Photoshop and Illustrator still don't need much to work quickly. Unless you insist on crazy.
Blender could probably use a CPU upgrade until I remember that it's probably easier to upgrade my GPU - get Cycles to render on the GPU... after all my GTX 660 is faster than consumer CPUs - probably even IVB-E... *shrugs*.
I'll probably jump on the first platform with native SATA Express. That's probably going to be Skylake-E.
I know some people like to run Fritz chess or whatever... but I just want a compelling reason to upgrade...
I'm disappointed at AMD. Where's my AM3+ w/improved single threaded perf upgrade path?
i think that amd's single core performance deficit has more to do with memory subsystems e.g. cache and system memory. single core performance isn't as important as before, but strong per-core performance goes a long way. amd can't push single thread execution performance by design due their module based (not to be confused with modular design approach) cores.
system memory performance deficit may alleviate with ddr4. i say "may" because it'll depend more on how amd redesigns carrizo and carrizo's successors' imc. but the biggest gainer from a well-performing imc would be amd's small-cores, puma and later.
i read somewhere that h.m.c. and h.b.m. tech both support ddr4 specifications....? those two can add massive bandwidth boost. i saw in an h.m.c. promo slide that 128-bit hmc(it had an h and an [h]m[/i] in the acronym iirc) can approach 51GB/s - that's lower speed gddr5 territory, likely with lower latency.
amd-hopefuls should tough it out for the next few months... or a Year (that's poking fun at glofo). once ddr4 stabilises in mainstream, we might see some fast performing cpus... hopefully. it'll depend on how amd executes.
CPU performance isn't being hindered by memory bandwidth but by memory latency. Most of the pain is the long access times to L2 and L3 caches. We're also comparing AMD, which is actually a pretty good design to Intel who has mastered cache and prediction technology. I work on non-x86 architectures and AMD's per core performance is about where the rest of the world is it, Intel just has a giant lead in that field due to their superior cache and prediction. The other ISA's can do just fine because their systems offer ridiculously high system I/O performance that can't be matched in the x86 world, but AMD is forced to directly compete with chipzilla. As long as single core performance is the primary determiner for total performance we won't see Intel knocked off it's throne anytime soon.
I don't think it is a threading issue. If it was then the i3 wouldn't also benefit from it. Instead Mantle is trying to bypass the CPU a bit more and utilize the GPU more same as DX12 would do.
Yeah. The fact the i3 gains more then the Pentium basically shows the point where the CPU is simply tapped out. Your seeing the effects of the main driver/render thread, and how much it eats CPU performance. Cut the load in half, and you can suddenly feed the GPU a lot more. Hence why the i3, with very powerful cores, gains as much as it does compared to the Pentium, which is tapped out regardless. That's also why the FX-6300 and FX-8350 gain at the same exact rate, since both will have the same benefit to freeing up one core.
Thief looks very multicore insensitive in general; see the FX-6300 vs FX-8350, or i7-3770k vs i3-4330. As a result of not using more then two cores, you see large gains to a 50% workload reduction on one of them.
Just a question - When, if ever, will AMD improve single-threaded performance? It is getting ridiculous how far behind AMD is.
Stop being a dreamer
haha probably true.
Next uarch likely...after excavator.
I expect parity single-thread perf. for second-gen post Excavator.
blackkstar :
juanrga :
blackkstar :
It's been things like that. Just look at the last few products AMD has released and compare it to something like FX 8150 launch.
The company no longer executes the way they used to. They have changed drastically. I don't see anyone accounting for this new AMD when making conjecture about the future.
During unending discussion in this thread about why AMD was not releasing a Steamroller FX CPU on SOI, I mentioned often that AMD had changed radically and even gave a quote from Raja Kodury. I give it again:
Raja returns to a very different AMD than the one he left. I asked him what’s different and he responded by saying the AMD he left acted like a company that was 10x its size. Today, AMD is a much smaller and more agile company. Raja believes AMD is in a better position to take advantage of new opportunities vs. being in the hopeless position of never being able to catch up in mature markets.
Yes, but I think the error is that you assume that AMD not directly competing with Intel means that they are going to abandon big cores.
A large problem with APU is that GPU and CPU want different things from the process they are made on. The most logical conclusion is to make separate dies for each, and use a CPU oriented process for the CPU and a GPU oriented process for the GPU.
But the problem is that Steamroller is on 28nm Bulk, which only offers a trade off between CPU and GPU. AMD could have either ported Steamroller to SOI (maybe 32nm SOI?) or released as 28nm Bulk for an SR FX chip.
I have said this before, but even if AMD wanted to release an SR FX chip, they have no good options to do so. SR FX would have been a huge disappointment on bulk with the reduced clock speeds.
Meanwhile, AMD is currently forced to keep everything on one die because there is no good way to get CPU and GPU talking to eachother without a ton of latency over long distances (I mean long distances in SOC terms, like across an ATX motherboard).
Their hands are tied. They are in an awkward position where the things they need aren't out yet.
I do think it's entirely possible for AMD to release a new HEDT platform that is focused around high performance while not directly competing with Intel.
"Directly competing with Intel" means trying to fight them directly with x86 CPU performance. If AMD released an HEDT HSA platform, it wouldn't be directly competing with Intel because it'd be an entirely different animal than just "lets look at these bar graphs and see the scores for SuperPI!"
It'd be sort of like what we're seeing with Mantle vs DX benchmarking right now. AMD has technology that their competition doesn't have access to and it gives them a massive advantage. Except imagine that happening with CPUs instead of GPUs.
I mean, I'd even go as far as to think that AMD could be pondering just ignoring traditional x86 performance entirely and instead pushing HSA on a future platform and instead pushing HSA enabled applications.
But, it's like Mantle on Linux, it has a massive problem with chicken and egg.
Who wants to buy hardware for HSA when there is no HSA software? Who wants to make software for HSA hardware when no one has HSA hardware?
Breaking that deadlock is key for AMD and they will do it somehow. I would not be surprised if out of the blue AMD released a new HEDT platform that's HSA enabled along side a ton of HSA applications and a massive game for Mantle in Linux that offers massive advantages over what you have in Windows (be it performance, features, etc). I also wouldn't be surprised if it ended up not being much of an improvement over Intel and Steamroller in traditional workloads, but absolutely mauled Intel and old AMDs when HSA is enabled.
If this was the AMD of FX 8150 and JF-AMD, I wouldn't think that that would have a chance of happening at all. But AMD has changed and they are ambiguous in the few things they say and they don't overly hype their products or anything anymore.
HSA is AMD's final push against Intel and Nvidia, and they've learned from the past. Why make something that can benefit their competition when they can gang up with other who have common competition and create something that leaves the competition out?
AMD completely blew amd64 instruction set, dual cores, quad cores, 3dnow, etc. HSA is a completely different approach and so is Mantle.
Mantle is a very curious case because AMD knows they talk about how Mantle is open and anyone can use it and they know Nvidia never will support it because they're too proud and too fixated on proprietary solutions they control.
HSA is meant to lock Intel out because AMD sees that their GPU advantage is shrinking and they know they can't depend on it forever. WIth HSA, even if Intel totally creams AMD CPU and AMD GPU, it won't matter. If it were OpenCL AMD would be in trouble, but not with HSA.
Locking out Intel and Nvidia is key, and having a ton of other partners is massive as well. Even VIA would be a competitor to Intel with HSA...
I don't expect AMD to abandon big cores. Big cores are needed for HSA. I except them to abandon the Bulldozer module design. I think AMD did learn the lesson and is replacing the old brute force approach to performance (which is a disaster) by the efficient approach to raw performance. That is why are replacing jaguar by A57 on servers. The new core is both faster and efficient than the core replaced.
GPU and CPU can be made on same process on same die. The problem was AMD went with the wrong SOI route in the past (former AMD management took lots of silly decisions). They are now correcting the mistake with their move to bulk and then to FINFET. It is broadly believed that FINFETs are required for maximum performance applications. This is why 95% of foundries have chosen FINFETs including Glofo.
The reduction of clock speed of Kaveri is partially related to the reduction of TDP from 100W to 95W. A hypothetical 100W Kaveri could be clocked at about 3.9GHz.
Releasing Steamroller FX series on 32 nm SOI makes no sense, because it would be so expensive that nobody would buy them. And maintain in mind that sales for current PD FX chips are already small (less than 25% of total CPU/APU/SoC sales).
Releasing a HSA enabled HEDT platform has been AMD goal since they acquired ATI. I already gave link to AMD chief engineer mentioning why they chose a 10 TFLOP APU for their future supercomputer project. The original document details both hardware and software of the ultra-high-performance APU. I already mentioned details of the hardware: 8-cores, 256-bit wide FPU units, six channel RAM plus stacked RAM, L3 cache coherence between CPU and GPU...
That APU would be about 10x faster than any imagined CPU + discrete GPU.
The part describing software is interesting as well. It details a FSA approach, which is clearly the precursor of the current HSA.
I think Nvidia has to be very worried by HSA, basically because they have nothing to compete right now. Intel has neo-heterogeneity to compete against HSA.
Neo-heterogeneity is based in the old dream of single-ISA heterogeneous compute, which beats the current CPU+GPU model. However, the HSA approach introduces something similar: the HSAIL ISA, which has to be supported by both the HSA enabled CPU and the HSA enabled GPU.
I am still undecided on which is the best approach.
another day, another lopsided set of benchmarks pitting kabini's igpu against bonaire....
http://semiaccurate.com/2014/04/16/adding-discrete-graphics-amds-am1-platform/
poor showing from mantle. mantle should be boosting fps for unbalanced combinations like this.
another day, another lopsided set of benchmarks pitting kabini's igpu against bonaire....
http://semiaccurate.com/2014/04/16/adding-discrete-graphics-amds-am1-platform/
poor showing from mantle. mantle should be boosting fps for unbalanced combinations like this.
Here's the issue: Too weak a CPU, and the CPU benefits are lost, as you still have a CPU bottleneck. Too weak a GPU, and the same issue occurs: You give an already overburdened GPU MORE work, so no benefit. There's a point where the HW is simply too weak to perform, period, and there's no magic bullet to fix that problem in software. For a CPU heavy game like BF4, freeing up 20% or so CPU time won't benefit you if your CPU is a very weak Athlon; that 20% still leaves the CPU badly overworked.
Just like there's a point where the HW is too powerful to see a benefit to Mantle, there's also a point where the HW is too weak to see a benefit.
another day, another lopsided set of benchmarks pitting kabini's igpu against bonaire....
http://semiaccurate.com/2014/04/16/adding-discrete-graphics-amds-am1-platform/
poor showing from mantle. mantle should be boosting fps for unbalanced combinations like this.
Mantle numbers aside it's not that bad for the price.
Just a question - When, if ever, will AMD improve single-threaded performance? It is getting ridiculous how far behind AMD is.
They will, I just don't think they are trying to leapfrog Intel like they did with their K8 arch again. Just stay enough behind that they look good by price comparison.
de5_Roy :
another day, another lopsided set of benchmarks pitting kabini's igpu against bonaire....
http://semiaccurate.com/2014/04/16/adding-discrete-graphics-amds-am1-platform/
poor showing from mantle. mantle should be boosting fps for unbalanced combinations like this.
With the low gaming numbers I would still put it in a HTPC so long as it doesn't give choppy video playback. Another good option is to just throw a HD 7750 LP into the rig and have a low end gamer/Blu-ray movie player.
wiinippongamer :
If AMD released an 8-core steamroller at 4,5ghz without a ridiculous TDP I'd be all over it.
I don't think that will happen. Even if it does, I don't think they will have a CPU that all of the sudden per clock beats Intel as Steamroller is still well behind in IPC from Intel.
I don't think AMD is fighting Intel that way anymore. If they were we should have seen something about what is next for just CPUs but mostly we se just APU stuff. Hell we know more from Intels plans 3 years from now than AMDs plans next year for CPUs.
I think they are banking everything on APUs and HSA. Problem is if Intels Skylake does what they want it to do then HSA may not be enough for AMD.
I don't think AMD is fighting Intel that way anymore. If they were we should have seen something about what is next for just CPUs but mostly we se just APU stuff. Hell we know more from Intels plans 3 years from now than AMDs plans next year for CPUs.
I would expect that trend to continue as they're shifting to semi-custom design. Those designs, like PS4/XBone won't get leaked until much later in the design cycles. And depending on the market segment (gov/telecom) may not get leaked at all.
Tomorrows earnings report should be interesting although the expecting earnings of $0/share doesn't bode too well.
I think they are banking everything on APUs and HSA.
and GPUs, MANTLE, ARM, SoCs...
jimmysmitty :
Problem is if Intels Skylake does what they want it to do then HSA may not be enough for AMD.
It will be an interesting fight because both are very close technologies and because Skylake will likely compete against the new CPUs and GPUs currently designed by Raja, Keller, and company.
Real loser here is Nvidia, which has nothing to compete against them.
The funny thing about seekingalpha is for every bad report there's usually a good one too. With such bipolar results does anyone even subscribe to get the 2nd pages of those articles?
But yeah I agree. With Intel trying so hard to get into that market segment AMD is just getting trampled. Thus the recently introduced socketed Kabini platform.
Mantle, not so much. If they were banking on Mantle then they wouldn't be supporting DX12 and rather would be pushing Mantle forward. Since they are supporting DX12 it is obvious they don't expect Mantle to take over.
ARM, I wouldn't consider that when competing against Intel. I meant on a CPU level that AMD is banking on HSA to carry their competition forward instead of working on a very strong CPU. As it stands with HSA, they could have a weak CPU core but a strong iGPU will help to negate that so long as the software can use HSA.
ARM and the SoCs are more other arenas to get into and for SoCs they will have just as much trouble pushing their solutions against ARM as Intel will have.
As for GPUs, again that is not against Intel but NVidia and there they have to stay competitive since their GPU designs are the same as what they employ in the ever more profitable workstation environment GPUs. Yet NVidia still owns the GPU market in comparison. I see more workstations with Quadros than I do with FireGLs but that may be due to the fact that a lot of software in that market supports CUDA but not all support OpenCL.
juanrga :
It will be an interesting fight because both are very close technologies and because Skylake will likely compete against the new CPUs and GPUs currently designed by Raja, Keller, and company.
Real loser here is Nvidia, which has nothing to compete against them.
It will be interesting to see if Intel can finally push out a IGPU that can beat or compete at least evenly with AMD. As they have a process advantage and already have the ability to put stacked DRAM on CPUs they might come out on top if the iGPU itself is actually good.
That said, it will be interesting only in that aspect as I do not think AMD is looking to improve their CPU design as much as Intel is. For example, Intels cache speeds are greatly faster than AMD. While we wont notice them for now, in the future Intels CPUs will fare better with applications that need more bandwidth while operating than AMD will.
As for GPUs, I think NVidia is fine. For discrete GPUs they still command a larger market share than AMD and have Tegra which some people seem to like. I think overall they will survive just fine so long as Intel and AMD don't find a way to put the equivalent to a 290X/780Ti onto a CPU.
Facebook
Twitter
Instagram