AMD CPU speculation... and expert conjecture

[juanrga]

"HSA" won't change anything unless a program is designed to use it. I can't compile 7zip for SPARCv9 and expect it to run on an Haswell/Piledriver CPU. A program compiled for x86 without special "HSA" related code will not utilize any additional processor units associated with that code.

Of course, that is why we are mentioning "HSA enabled software" when mentioning the 500% increase in performance.

 

[8350rocks]

The 9970 tdp is suppose to be 250w same as a gtx780 or titan. So if the 9970 doesn't beat the titan then the 9970 most likely will have worse performance per watt then kepler.

TDP = Thermal Design Power

This does not equate equally to power consumption...

In some cases, devices with a given TDP (for example A10-6800k @ 100W TDP) may typically consume substantially less power than their Thermal Design Power.

In other cases, the consumption may be very close to the TDP (HD 7970 GHz/GTX 680)...

In some cases, the power consumption is actually quite a bit higher than the listed TDP (i5-3570k for example).

TDP is talking about the heat generation properties of the component...not actual power consumption.

 

[8350rocks]

The 9970 tdp is suppose to be 250w same as a gtx780 or titan. So if the 9970 doesn't beat the titan then the 9970 most likely will have worse performance per watt then kepler.

TDP = Thermal Design Power

This does not equate equally to power consumption...

In some cases, devices with a given TDP (for example A10-6800k @ 100W TDP) may typically consume substantially less power than their Thermal Design Power.

In other cases, the consumption may be very close to the TDP (HD 7970 GHz/GTX 680)...

In some cases, the power consumption is actually quite a bit higher than the listed TDP (i5-3570k for example).

TDP is talking about the heat generation properties of the component...not actual power consumption.

lol the 3570k cpu takes like 35w and the igpu makes up like 10w. The rest is the idle power consumption. Fx8350 maybe 125w tdp but takes like 150w, the 4770k tdp is 84w but takes like 50w for the cpu cores. Nvidia tdp is generally how much there power consumption consumes while amd tdp are normally lower then the power consumption.
This proves the a10-6800k cores on cou id hw monitor takes like 92w.
http://www.guru3d.com/articles_pages/amd_a10_6800k_review_apu,8.html

3570k ia cores that is cppu cores take 42.8w.
http://www.guru3d.com/articles_pages/core_i5_3570k_review,10.html

That's why the 3570k draws 161W at load right? The 3770k draws 166W too because Intel is way under their stated TDP of 77W there too right?

What about clocked at 5.0 GHz like the monster 9590 AMD? Oh it draws 220W!!!! How terrible!?

Well, the 3570k @ 5.0 GHz draws 267W @ load clocked at 5 GHz and the 3770k draws 244 @ 4.8 GHz.

Hmm...think your perf/watt went out the window there man...

http://www.bit-tech.net/hardware/2012/04/23/intel-core-i7-3770k-review/8

EDIT: FYI, that test the Intel's are using a Micro ATX board as well, if it was a normal ATX board the consumption would be about 20W higher as well:

Note: the AMD chips were tested in an ATX motherboard, while the Intel LGA1155 chips were tested in a micro-ATX board. This difference can account for up to 20W, as we found in our Energy Efficient Hardware feature.

 

[juanrga]

No one who knows anything about modern computers and software uses Superpi as a benchmark for anything. Correction, no one except people who don't know any better, people who haven't kept up with the technology since the late 1980s, and Intel fanboys use Superpi. It is a total worthless benchmark, but one that Intel fanboys love to use because Intel processors score high with it. In actual real raw performance we have seen MODERN benchmarks that put the FX-8350 above i5 and in some cases above i7 4770K. In those modern benchmarks the only processors that beat the FX-8350 are 6 core Intel that cost five times as much.

Completely agree. Only to add that current very low scores of AMD on SuperPi are due to a deficiency on BIOS. When the deficiency is eliminated with a patch, AMD processors receive a huge boost in SuperPi scores

http://www.techpowerup.com/186056/amd-super-pi-history-to-be-rewritten-courtesy-of-the-stilt.html

 

[8350rocks]

http://www.hardwarecanucks.com/forum/hardware-canucks-reviews/62166-amd-fx-9590-review-piledriver-5ghz-17.html

Look here these i7 i5 and i3 have there igpus and cpus stressed to the max in these power consumption. The 4770k with igpu stressed still takes 147w less then an fx9590.

The benchmark I posted above is done CPU only running prime95...weird how the results are so dramatically different huh?

Isn't it odd how you always find benchmarks that are the statistical outlier and contradict the majority of the others...? I wonder why that could be...?

 

[The Q6660 Inside]

8350rocks nearly every website has similar readings.

I am getting an i5 3210m laptop and this should consume at 2.9ghz around 30w full load and the cpu cores will consume about 16w. If you double the cores and threads at 2.9ghz it should need more power so lets say 3770k takes around 45w is very realistic and the i5 3570k taking around 40w is very realiistic to as these are desktop parts which are normally running at slightly high voltage.

Look, a QX9350 has a 55W TDP, the equivalent Q9450 has a 95W TDP. Laptop chips are designed around a smaller TDP. A T9900 has a TDP of 35W and a E8400 has a 65W TDP. That being said, as we scale down the process, the TDP for each core becomes lower, ala i3-3220. A realistic power consumption for the "CPU" part of the 3770K is around 80-90W.

 

[juanrga]

Our dream of a Kaveri APU powered with ultrafast GDDR5 memory in DIMMs is coming

http://www.hardware.fr/news/13287/gddr5m-4-gbps-sodimm-kaveri.html

 

[8350rocks]

http://translate.googleusercontent.com/translate_c?act=url&depth=1&hl=en&ie=UTF8&prev=_t&rurl=translate.google.com&sl=auto&tl=en&u=http://www.hardware.fr/news/13287/gddr5m-4-gbps-sodimm-kaveri.html&usg=ALkJrhjf0Z9UCQOPZkUB2eAUD7J6xH5i8w

interesting read...

 

[gamerk316]

"Guesstimating" is about as close as anyone can ever get, even then, you have statistical outliers that skew results because one architecture has it as a strength and another does not.

For example, if you compare games like Crysis 3 and Battlefield 3 (multiplayer especially)...you find that AMD and Intel are very close in performance. In other games, they are virtually the same. Then you have poorly optimized games like Planetside 2 or Skyrim that completely skew the results because something in the way they were coded is very well received by one architecture, but not the other even though AMD supports more instruction sets than Intel on their current generation CPUs.

Hence needing a large enough sample.

 

[blackkstar]

The 9970 tdp is suppose to be 250w same as a gtx780 or titan. So if the 9970 doesn't beat the titan then the 9970 most likely will have worse performance per watt then kepler.

Hi forum slider, you're really good at your job.

However I don't think the users here are dumb enough to fall for "performance per watt matters in a high end GPU." I'm not sure if you're ramping up the trolling and anger by trying to act so inept that you suddenly forgot how to use the quote feature or if you're just mad, but regardless it's fun to see you either get desperate or fall off your handle. You should probably leave this thread, you're so far below me in mental capacity that it's not even funny and all your pathetic attempts at rusing this forum and sliding it are going to die.

If you folks are going to address hajidurp, I suggest you make posts consisting of half bashing him and half talking about Steamroller, HSA, or AMD products. His job is to derail the thread by making us respond to him and ignoring AMD, but we can do both in the same post :lol:

But those GDDR5 modules look great. If the speeds can ramp up AMD APUs might be the absolute best choice for bandwidth intensive applications. It could be superior to quad channel DDR setups while using significantly less die space and costing significantly less.

I am not sure if AMD would want to drop 8 cores for 4 cores + HSA. It's going to be a tough sell to get people to take a performance hit in general workloads for big gains in HSA. AMD already has a big enough problem with getting people to go AMD FX for multi-thread because it's far more cost effective for those types of workloads compared to a similarly priced Intel. Intel's competition at that price point is i5 4570, not the K version you can overclock and it lacks HT. So it's going to lose in multi pretty much every time.

I wouldn't be surprised if we saw some sort of high end APU in the $300 to $500 segment. I think FX 9000 series was "conditioning" of the consumer to accept higher prices for AMD CPUs. It is similar to what Nvidia did with Titan and GTX 780. It was a means to make $650 for a GTX 780 seem like a good value (it's a gimped Titan and Titan costs $1000, so it's a great deal!) when GTX 680 cost $550 and GTX 580 cost $499 at launch. Notice the price creep Nvidia has been doing, and notice how no one is really upset about it? That's how you introduce more expensive to produce products to market while increasing profit margins without pissing off your customers and fans.

APUs dropped the HT link, which is needed for multi-chip-module (ergo putting two 4 core AMD APUs on a single package for a single socket). Perhaps they could add it to Steamroller APUs and then sell MCM APUs? An 8 core SR with 16CUs would be an absolute monster of a chip.

 

[griptwister]

So, I'm confused. I thought the HD 9970 was a HD 8970. Is the HD 9970 going to be a re brand, or is it going to be totally new? I saw some one say it was 20nm, which I personally think is a lie. Anyone got any info?

I'm hoping this smaller process makes for cooler GPU temps. I don't like Overclocking GPUs because of how friggin hot it gets. I also don't have money for liquid coolers lol.

The whole Nvidia Marketing scheme was in-genius. That's also what Intel does with their hexicore i7s. They sell one for $1000 because it has a higher clock.

 

[juanrga]

But those GDDR5 modules look great. If the speeds can ramp up AMD APUs might be the absolute best choice for bandwidth intensive applications. It could be superior to quad channel DDR setups while using significantly less die space and costing significantly less.

The modules are guaranteed to provide much more bandwidth than future DDR4. I have always believed that AMD would ignore DDR4 for now and work hard on developing GDDR5 memory subsystems for CPUs. They have already the experience with PS4 now.

I am not sure if AMD would want to drop 8 cores for 4 cores + HSA.

AMD is doing exactly that in servers with the new Berlin APU:

8 PD cores --> 4 SR cores + 8 GCN cores

I wouldn't be surprised if we saw some sort of high end APU in the $300 to $500 segment. I think FX 9000 series was "conditioning" of the consumer to accept higher prices for AMD CPUs. It is similar to what Nvidia did with Titan and GTX 780. It was a means to make $650 for a GTX 780 seem like a good value (it's a gimped Titan and Titan costs $1000, so it's a great deal!) when GTX 680 cost $550 and GTX 580 cost $499 at launch. Notice the price creep Nvidia has been doing, and notice how no one is really upset about it? That's how you introduce more expensive to produce products to market while increasing profit margins without pissing off your customers and fans.

This is an interesting idea. It also matches with the idea that AMD released 220W chips for facilitating the development of motherboards that could accept future high-end performance APUs with huge TDPs. Currently AMD is restricted to ~100-130W mobos.

Kaveri 4 cores matches an i5 in CPU performance. AMD would release an alternative to i7s. The same about servers, The Opterons with 4/8 Piledriver cores are replaced by Berlin 4 cores. AMD would release a replacement for the Warsaw with 12/16 Piledriver cores.

Either AMD will release the rumoured 6 core version of Kaveri or directly a 4 core Carrizo (featuring the new Excavator cores). Similar thinking about servers: either AMD will release a 6 core version of Berlin or directly something new with 4 excavator cores.

APUs dropped the HT link, which is needed for multi-chip-module (ergo putting two 4 core AMD APUs on a single package for a single socket). Perhaps they could add it to Steamroller APUs and then sell MCM APUs? An 8 core SR with 16CUs would be an absolute monster of a chip.

AMD has done something similar with the jaguar-based APU in the PS4.

 

[BeastLeeX]

So, I'm confused. I thought the HD 9970 was a HD 8970. Is the HD 9970 going to be a re brand, or is it going to be totally new? I saw some one say it was 20nm, which I personally think is a lie. Anyone got any info?

I'm hoping this smaller process makes for cooler GPU temps. I don't like Overclocking GPUs because of how friggin hot it gets. I also don't have money for liquid coolers lol.

The whole Nvidia Marketing scheme was in-genius. That's also what Intel does with their hexicore i7s. They sell one for $1000 because it has a higher clock.

28nm, GCN 1.1 (just a refresh, prolly better power consumption, new memory chips). The 28nm is happening, the GCN 1.1 refresh is expected

Source: http://wccftech.com/amd-volcanic-islands-family-hawaii-xt-hawaii-pro-possibly-based-current-gcn-architectures/

 

[hcl123]

2.1x is 110% better ipc. Superpi shows this. Thats why an i5 4670k with 4 cores beats the amd 8 core fx8320.

Perfect example haswell vs piledriver with same amount of cores:
http://www.anandtech.com/bench/product/700?vs=837

On extremely multithreaded tasks the 4670k is around 2x faster with lower clock speed. Reason is for every 2 cores of piledriver at same clock speed as 1 core of haswell, haswell cpus are just over 2x better ipc. Thats why steamroller has to improve 100% ipc for what people on these forums say a 4 core SR to be intel haswell level. or at least 80% better to be on sandy bridge level with same amount of cores.

If superpi was so bad then why does it roughly show why intel can beat amd with half the amount of cores as its accurate. I think if anyone has both an amd piledriver system and an intel haswell system, run fotosizer same files same settings and see if the intel system is 2x faster. It should be.

Good show LOL

there isn't no Superpi test in the link you point (oops wrong propaganda ! lol).

But even discounting the "biased" AT, meaning doubt if they had done a superpi test they would include this http://www.xtremesystems.org/forums/showthread.php?286448-The-Book-of-Bulldozer-Revelations-Episode-2-%28SuperPI-x87%29 , and even discounting that Windows benchmarks are already more "optimized" for intel, which makes them not suitable for a real impartial comparative, when concerning REAL HIGH PERFORMANCE, is to suspect by the less knowned tests like "Par2 multithreading" (usually not in time to be optimized in the right way LOL) that the even the FX4300 wins for the 4670K lol

UPDATE

Oops my mistake, is "time" not a "logarithmic positioning" or "GB/s"... hell will freeze over before Anantech shows an "impartial" test showing AMD winning anything ( that is what made me "look better again" ). Anyway by the price the real contender for the 4670k is not the F4300, not even the FX8320 but the FX8350 (which is now below the $200).

Yes it loses in power consumption, equivalent to have one more 40W lamp lit( at most perhaps more $5 at the end of the month, depending on INTENSIVE usage), but it wins in clock hands down ( you can't have both, low power and high clock) and it wins at multithreading with double the REAL core number(more prepared for the future)... and is cheaper...

Why would intel be better ? ( if the 4670K would be $100 it would be recomendable)

 

[The Q6660 Inside]

So, I'm confused. I thought the HD 9970 was a HD 8970. Is the HD 9970 going to be a re brand, or is it going to be totally new? I saw some one say it was 20nm, which I personally think is a lie. Anyone got any info?

I'm hoping this smaller process makes for cooler GPU temps. I don't like Overclocking GPUs because of how friggin hot it gets. I also don't have money for liquid coolers lol.

The whole Nvidia Marketing scheme was in-genius. That's also what Intel does with their hexicore i7s. They sell one for $1000 because it has a higher clock.

And add 3MB of cache... ala TITAN. The 9970 is 100% Brand New, not a desktop 8970M.

 

[hcl123]

Our dream of a Kaveri APU powered with ultrafast GDDR5 memory in DIMMs is coming

http://www.hardware.fr/news/13287/gddr5m-4-gbps-sodimm-kaveri.html

What about FM2+ ? it for sure doesn't have provisioning for GDDR5M DIMMS or SO-DIMMS... and there are boards already out there.

Doubt GDDR5M will go for the mobile SKUs, it would be much cheaper for IHV only DDR3 SO-DIMMS for those laptops, and a L3 block of some kind to compensate some lack of memory bandwidth.

For the "desktop" GDDR5M(edt) could be a solution, but not DIMMS (price of a decent GDDR5 DIMM could be the same of a APU), but more like this http://www.techpowerup.com/img/11-05-03/17a.jpg sheds a lot of intermediary price "share" compared with a DIMM.

OTOH i think ESRAM could be even cheaper... instead of 3 modules(Windows PC world doesn't take real advantage of 6 cores anyway), the same space of 1 module(edt) will be a ESRAM block... and is to suspect it could be quite cheaper than GDDR5, even the tiny cheap Wuii chip has it.

Intel eDRAM has a L4 is a mistake, it makes the chip too large and too expensive, only the L4 Flag File CAM structures are larger than any ESRAM so far, and this in spite the 22nm size.

 

[cowboy44mag]

So, I'm confused. I thought the HD 9970 was a HD 8970. Is the HD 9970 going to be a re brand, or is it going to be totally new? I saw some one say it was 20nm, which I personally think is a lie. Anyone got any info?

I'm hoping this smaller process makes for cooler GPU temps. I don't like Overclocking GPUs because of how friggin hot it gets. I also don't have money for liquid coolers lol.

The whole Nvidia Marketing scheme was in-genius. That's also what Intel does with their hexicore i7s. They sell one for $1000 because it has a higher clock.

I generally don't suggest overclocking the GPU unless you have the necessary hardware to do it. Water coolers work well, but are expensive and you need a case with enough room to install them. A much better, and cheaper solution:

http://www.newegg.com/Product/Product.aspx?Item=N82E16835213001

I have a few of these fans in my case, one right over the GPU, big overclocks need adequate cooling. You can overclock the GPU all day long and keep temperatures totally in check, without the need of water cooling

 

[hcl123]

So, I'm confused. I thought the HD 9970 was a HD 8970. Is the HD 9970 going to be a re brand, or is it going to be totally new? I saw some one say it was 20nm, which I personally think is a lie. Anyone got any info?

I'm hoping this smaller process makes for cooler GPU temps. I don't like Overclocking GPUs because of how friggin hot it gets. I also don't have money for liquid coolers lol.

The whole Nvidia Marketing scheme was in-genius. That's also what Intel does with their hexicore i7s. They sell one for $1000 because it has a higher clock.

uumm... no!... the Hawaii magic will be at the CUs, just to remember Fermi had 512sp(double clock 1400-1600Mhz) and could compete with the 1536sp of AMD, Kepler was a mistake, a step backwards in order NOT to make the 1.5B intel gave them, irrelevant for those games at the CPU side. A Fermi evolution could had dispensed the CPU almost entirely for the scheduling and a lot of geometry processing... then any i5 would perform at the same level of any i7... (yes i5=i7, but it has the potential to perform much better by bypassing the CPU, a lot of memory copies can be avoided, only it would not be good for intel business... the users F them, they can swallow any propaganda... [why kepler was a mistake] ) not good for business LOL ... and the whole Nvidia marketing "genius" is mostly Intel propaganda by inference (since the beginning the "orders" were any test is with a nvidia, no matter if the Tahiti was better than the 680) lol...

Money can sing very loud, pity this is a good illustration of stagnation forces in action... win some lose other... i think nvidia is not surprised the console guys showed the finger to all this "propaganda campaign" and discarded them... and the irony is that Intel is positioning fast "for a kill", the HSW shows the smell of dead to the nvidia cash caw, now its not only the APUs , and Broadwell is mostly BGA ( what a friends lol), and AMD platforms will be HSA (which nvidia is not), so where is nvidia going to put their boards ? LOL ... nvidia is clearly between a rock and a hard place, or Tegra and ARM64 spill in force soon to all client market sectors (i hope, good for users), or they are up for real hard times with nothing to compete in mass markets (so much for the "genius" lol)...

GCN uarch already has for each CU (64sp) 64KB x4 for the SIMDs + 8KB for the Scaler Core + 64KB of Scratchpad register extension(local share), makes 328KB of CAM oriented memory for each CU, and this without counting any L1 or L2.

This is already quite above what Fermi had, and i believe is sub-optimized. Hawaii could have 36 CUs(36x64=2304 sp) compared with the 32 of Tahiti, it will be 36x328 = 11.808KB of CU memory (almost 12MB) in CAM structures (quite above any CPU).. if those CUs perform(at 1200Mhz) at the level of a Fermi it would be at least 4x has much, so the potential is to be at least 4x Cayman, quite above Titan. 2 MB of cache could be enough, same scheme of today, only more CUs means more L1.

All quite possibly yet below the 450mm², so an optimized 28nm process is reasonable ( no need for real monster chips).

 

[griptwister]

So, I'm confused. I thought the HD 9970 was a HD 8970. Is the HD 9970 going to be a re brand, or is it going to be totally new? I saw some one say it was 20nm, which I personally think is a lie. Anyone got any info?

I'm hoping this smaller process makes for cooler GPU temps. I don't like Overclocking GPUs because of how friggin hot it gets. I also don't have money for liquid coolers lol.

The whole Nvidia Marketing scheme was in-genius. That's also what Intel does with their hexicore i7s. They sell one for $1000 because it has a higher clock.

I generally don't suggest overclocking the GPU unless you have the necessary hardware to do it. Water coolers work well, but are expensive and you need a case with enough room to install them. A much better, and cheaper solution:

http://www.newegg.com/Product/Product.aspx?Item=N82E16835213001

I have a few of these fans in my case, one right over the GPU, big overclocks need adequate cooling. You can overclock the GPU all day long and keep temperatures totally in check, without the need of water cooling

Perfect! I've been looking for something to fit the bottom of my new case. I'll probably upgrade a modular PSU when I upgrade my CPU any how. I'm looking to eventually pop in 5 of these...

http://www.newegg.com/Product/Product.aspx?Item=N82E16835553010

I'm so excited to see Kaveri. I'm looking for a good GPU and CPU for my friend who is saving up to build his PC. I told him about Kaveri and He's hoping it drops early too. he originally wanted an i5 4670k, but him and I both know the intel HD graphics suck. Lol.

@hcl123: NAIL'D IT! So much information. Haha!

 

[The Q6660 Inside]

So, I'm confused. I thought the HD 9970 was a HD 8970. Is the HD 9970 going to be a re brand, or is it going to be totally new? I saw some one say it was 20nm, which I personally think is a lie. Anyone got any info?

I'm hoping this smaller process makes for cooler GPU temps. I don't like Overclocking GPUs because of how friggin hot it gets. I also don't have money for liquid coolers lol.

The whole Nvidia Marketing scheme was in-genius. That's also what Intel does with their hexicore i7s. They sell one for $1000 because it has a higher clock.

I generally don't suggest overclocking the GPU unless you have the necessary hardware to do it. Water coolers work well, but are expensive and you need a case with enough room to install them. A much better, and cheaper solution:

http://www.newegg.com/Product/Product.aspx?Item=N82E16835213001

I have a few of these fans in my case, one right over the GPU, big overclocks need adequate cooling. You can overclock the GPU all day long and keep temperatures totally in check, without the need of water cooling

Perfect! I've been looking for something to fit the bottom of my new case. I'll probably upgrade a modular PSU when I upgrade my CPU any how. I'm looking to eventually pop in 5 of these...

http://www.newegg.com/Product/Product.aspx?Item=N82E16835553010

I'm so excited to see Kaveri. I'm looking for a good GPU and CPU for my friend who is saving up to build his PC. I told him about Kaveri and He's hoping it drops early too. he originally wanted an i5 4670k, but him and I both know the intel HD graphics suck. Lol.

@hcl123: NAIL'D IT! So much information. Haha!

Get dem 8350 and Sabertooth 990FX R2.0, it wants to hug you, oh and get a new case :3

 

[hcl123]

I'm so excited to see Kaveri. I'm looking for a good GPU and CPU for my friend who is saving up to build his PC. I told him about Kaveri and He's hoping it drops early too. he originally wanted an i5 4670k, but him and I both know the intel HD graphics suck. Lol.

@hcl123: NAIL'D IT! So much information. Haha!

I'm more curious to see what secrets the Hawaii chips holds, for when the X2 or the RD1090 chipsets shows up.

I think is point concluded that the "sideport" will disappear somehow, i wonder if they replaced it with Hypertransport HTX compatibility (perhaps capable of up to 4GHz , half of PCIe v3, but capable of the same 8 GT/s... and much better latency ).

For one point then the X2 versions could be much cheaper because they will happen "inside" the same socket/package, like the server G34, and there will be no excuse for reviewers, the "single chip" (1 socket 1 package) crown will be at hand for a looong time.

Besides i've shown already here a clear indicated AMD patent, were they present a "combo slot" HTX + PCIe, meaning those slots could function equal for a PCIe board or a HTX board, meaning if a GPGPU chip of AMD has identical interface, then a board with it could be put on an Intel mobo as if nothing is different, on a PCIe slot... but the same board could be put on an AMD mobo ( RD1090FX ?), but as a HTX slot of that combo possible config... just guess where it will be better, just because the interface is much better, not because it artificially cripples any solution ??

And HTX spells HSA on steroids, the GPU cache could be coherent with the CPU cache (which is *the* variant that the HSA hUMA official presentation had different from earlier disclosures where they talked only of system DRAM and system VM, by this it could be not only for APUs but for CPU +GPU card also)(edt), no matter if in different boards, this is Fermi approach on super-steroids lol... but if not on a HSA platform (which will be the majority of mobos out there for a while), no fun spoiled if there is a PCIe slot, that same HSA GPU board could be used ( the same with a HTX HSA ready mobo with vulgus PCIe cards) (edt), which is very good even for AMD.

This will be far more "revolutionary" than any current chip evolution... sorry to rain some on your parade lol...

 

[griptwister]

@hcl123: It took me 3 reads to really get a grasp of your paragraphs. Lol, that would truly be incredible.

Well, I was thinking about all the news and rumors. And I thought of this...

If AMD does move the "Enthusiast Processors" to the FM2+ socket, would there be a possibility that we would see SteamRoller Phenom III x6 and x8 Processors for that line up? Look at what they did with Richland Althlons. I mean it seems feasible.

@The Q6660 Inside: I desperately want to upgrade. But, as I said earlier, I'm upgrading my monitor to a 144Hz. I'm dropping 3 bills on a monitor, I'm going to be buying a decent CPU. (which is why I'm waiting for steamroller before I make a purchase) Also, I upgraded to a InWin GT1 (Fantastic case for the price, CPU runs 7C cooler and GPU runs 10C cooler under load over previous case, The dust filters are a HUGE plus)

 

[hcl123]

@hcl123: It took me 3 reads to really get a grasp of your paragraphs. Lol, that would truly be incredible.

Well, I was thinking about all the news and rumors. And I thought of this...

If AMD does move the "Enthusiast Processors" to the FM2+ socket, would there be a possibility that we would see SteamRoller Phenom III x6 and x8 Processors for that line up? Look at what they did with Richland Althlons. I mean it seems feasible.

@The Q6660 Inside: I desperately want to upgrade. But, as I said earlier, I'm upgrading my monitor to a 144Hz. I'm dropping 3 bills on a monitor, I'm going to be buying a decent CPU. (which is why I'm waiting for steamroller before I make a purchase) Also, I upgraded to a InWin GT1 (Fantastic case for the price, CPU runs 7C cooler and GPU runs 10C cooler under load over previous case, The dust filters are a HUGE plus)

I think Kaveri on FM2+ will be hUMA only on the APU side, put there a discrete card and it will not have that discrete GPU cache, coherent with the CPU cache. There are IOMMU tricks to make the GPU card see the same system virtual memory (VM) , will be good for virtualization solutions (all APUs have an IOMMU, which makes all of them better VMware players as example, than any intel chip... in this respect for the FX, only the top 990FX chipset have an IOMMU but old version by now), but will not improve performance not nearly as much as GPU card cache coherent with CPU cache, if this solution is to account for.

Don't take me wrong, kaveri seems a very good solution, already better than what is similar out there, better than Intel for sure for graphics and compute... a lot depends on how the drivers deal with the "crossfiring", if they address the discrete adapter only for graphics and video (which intel will not dispense for real gaming), then only the APU GPU will do compute, compute jobs where the lack of cache coherence is more penalizing... which is a very good interim solution... and those platforms are not for the top of "enthusiast gamers", but they can manage very well.

Kaveri will remain 4 cores top i suspect (good bet since Windows doesn't really take advantage of more now), but this time will NOT lose anything of the already ridiculous small difference for intel core to core... quite possible gain to hasfail core to core lol (of course bentmarks are already being "optimized" to show me wrong in next editions, is one of the easiest bets ever lol)

The best of the APUs is they are the best performance for the buck ever, top of those chips are around $100, is a sweat deal.

In the future is quite possible the APU platforms having HTX+PCIe combo slots ( like this: __________ x16 main slot + _____ smaller sideband on the same row, only HTX uses the sideband ) and so the APUs having also proper interfaces "on die" ... and 4 cores/threads per module (8 total)... and nearly has good for top "enthusiast gamers", but is not the case for kaveri for FM2+.

Yes i suspect AMD will keep some kind of FX platform for top enthusiasts, and that AMD will not abandon the FX brand and chips anytime soon (forever will not stand, and perhaps every chip will be an APU sooner than i anticipate some years ahead)... the "Centurium" (the 5Ghz FX9590) seems to reinforce this view i think ... the additions to those FX chips could be HTX+PCIe "on die" capabilities and likely one more DDR channel ( v3 and future v4).... i think Xfiring with double "single package X2" cards will be much more common on HSA HTX platforms, as today xfiring with 2 cards... and think those "possible" changes are quite good causes for delays, the rest is propaganda and fanboism...

[ oh ! well.. any doubt, it went WIPO already
http://www.freepatentsonline.com/WO2012138550A1.html

The drawings easily seen at faqs are self explanatory
SLOT DESIGN FOR FLEXIBLE AND EXPANDABLE SYSTEM ARCHITECTURE
http://www.faqs.org/patents/app/20120258611

images can tell a thousand words

* http://www.faqs.org/patents/imgfull/20120258611_01
http://www.faqs.org/patents/imgfull/20120258611_02
]

 

[juanrga]

So, I'm confused. I thought the HD 9970 was a HD 8970. Is the HD 9970 going to be a re brand, or is it going to be totally new? I saw some one say it was 20nm, which I personally think is a lie. Anyone got any info?

I'm hoping this smaller process makes for cooler GPU temps. I don't like Overclocking GPUs because of how friggin hot it gets. I also don't have money for liquid coolers lol.

The whole Nvidia Marketing scheme was in-genius. That's also what Intel does with their hexicore i7s. They sell one for $1000 because it has a higher clock.

28nm, GCN 1.1 (just a refresh, prolly better power consumption, new memory chips). The 28nm is happening, the GCN 1.1 refresh is expected

Source: http://wccftech.com/amd-volcanic-islands-family-hawaii-xt-hawaii-pro-possibly-based-current-gcn-architectures/

Hawai is not a refresh or update. The same link mentions that Hawai is an entire "new architecture".

Our dream of a Kaveri APU powered with ultrafast GDDR5 memory in DIMMs is coming

http://www.hardware.fr/news/13287/gddr5m-4-gbps-sodimm-kaveri.html

What about FM2+ ? it for sure doesn't have provisioning for GDDR5M DIMMS or SO-DIMMS... and there are boards already out there.

Doubt GDDR5M will go for the mobile SKUs, it would be much cheaper for IHV only DDR3 SO-DIMMS for those laptops, and a L3 block of some kind to compensate some lack of memory bandwidth.

For the "desktop" GDDR5M(edt) could be a solution, but not DIMMS (price of a decent GDDR5 DIMM could be the same of a APU), but more like this http://www.techpowerup.com/img/11-05-03/17a.jpg sheds a lot of intermediary price "share" compared with a DIMM.

OTOH i think ESRAM could be even cheaper... instead of 3 modules(Windows PC world doesn't take real advantage of 6 cores anyway), the same space of 1 module(edt) will be a ESRAM block... and is to suspect it could be quite cheaper than GDDR5, even the tiny cheap Wuii chip has it.

Intel eDRAM has a L4 is a mistake, it makes the chip too large and too expensive, only the L4 Flag File CAM structures are larger than any ESRAM so far, and this in spite the 22nm size.

Early leaked docs shown FM2+ supporting both DDR3 and GDDR5. The GDDR5 support was dropped in recent leaked docs and motherboards like this one

only have SO-DIMM DDR3 support, but AMD could release a high-end kaveri with new chipset supporting SO-DIMM GDDR5. The current high price of GDDR5 is directly related to low production volumes. Yes, high end dGPUs use GDDR5, but in very small quantities. Only a very small percentage of gamers have installed more than 2 GB GDDR5 VRAM. And gamers are a subset of PC users. A new generation of PCs with >= 4 GB GDDR5 for system RAM would drop prices considerably.

Moreover, AMD could release GDDR5 support only in an enthusiasts high-performance line, leaving cheap DDR3 for the rest.

Yes, Windows is poorly prepared for 6 cores. This is one of reasons why I doubt on if AMD will release a future 3 module kaveri version or directly will release the 2 module carizo APU with excavator cores.

The discussion ESRAM+DDR3 vs GDDR5 is interesting one and reproduces the XboxOne vs PS4 war. I would prefer the GDDR5 solution, because the ESRAM solution would obligate developers to tune software for the small size of the ESRAM in order to maximize performance

Microsoft tells developers that the ESRAM is designed for high-bandwidth graphics elements like shadowmaps, lightmaps, depth targets and render targets. But in a world where Killzone: Shadow Fall is utilising 800MB for render targets alone, how difficult will it be for developers to work with just 32MB of fast memory for similar functions?

http://www.eurogamer.net/articles/digitalfoundry-xbox-one-memory-better-in-production-hardware

In one sense, the ESRAM breaks the one-memory-pool purity of the GDDR5 approach, by introducing two pools: one small and fast and other slow and large. I think all game developers prefer the Sony approach of a single unified large pool of ultrafast memory.

My only doubt would be on what would be the performance gain of GDDR5 memory in pure CPU tasks. The Piledriver FX chips benefited from faster DDR3 memory, unlike intel Ivi Bridge.

 

[esrever]

http://www.fudzilla.com/home/item/32199-amd-hawaii-will-be-priced-below-us-$600

titan performance at $599? We will have to wait and see.