AMD CPU speculation... and expert conjecture

[Cazalan]

It could also mean they used the 6 cores from the XBox as a threading target, haha.

Cheers! 😛

Anand did some server benchmarks today and some surprising results still with turning HT/CMT off. Some 20% gains.

http://www.anandtech.com/show/6508/the-new-opteron-6300-finally-tested/9

 

[noob2222]

This is wrong. The 20 degree increase in IVB temps was due to using a thermal interface material between the heat spreader and the CPU die which causes a gap to form between the cpu die and heat spreader (on sandy bridge the heat spreader was soldered on which replaces this gap with solder). Removing the head spreader and reducing the distance between the head spreader and cpu die and replacing the TIM decreases temperatures by.... guess how much? 20 degrees.

The lower overclockability of IVB has to do with 22nm finFET device characteristics, not heat dissipation.

Check out this thread over at Anandtech for a detailed discussion.

Delidded my i7-3770K, loaded temperatures drop by 20°C at 4.7GHz

http://i272.photobucket.com/albums/...l Core i7-3770K/Delidded3770k47GHzresults.png

Rofl, the infamous tim blame. http://www.theinquirer.net/inquirer/news/2171299/intel-admits-ivy-bridge-chips-run-hotter

They all know more than intel about this.

First off, the warranty is 100% void for removing the lid, 2nd, approx 50% of the attempts to recreate this scenario resulted in dead cpus, and others saw gains of maybe 7c. One extreme case by an obviously extremist individual does not mean all attempts will be the same at the end.

But it does show that you can reduce temps by using extreme measures such as reducing the gap between the silicon and the lid (where he ground it with a dremel tool). Others have directly mounted the heatsink on the silicon, removing the lid alltogether.

 

[jdwii]

Off subject but this made me laugh

by game creators for game creators" and that its architecture is "like a PC, but supercharged." PS4 uses the X86 CPU and has 8GB of memory and a local hard drive. It uses APU technology and GDDR5 memory, which is typically reserved for "top of the line, high end graphics cards

 

[griptwister]

Off subject but this made me laugh

by game creators for game creators" and that its architecture is "like a PC, but supercharged." PS4 uses the X86 CPU and has 8GB of memory and a local hard drive. It uses APU technology and GDDR5 memory, which is typically reserved for "top of the line, high end graphics cards

Lol. 😀 :lol: Sony is so funny, aren't they? Haha!

 

[anxiousinfusion]

"like a PC, but supercharged."

Folks who don't follow electronics are going to get the wrong perception because of this statement. Expect many "PC Killer!" articles to follow.

 

[JAYDEEJOHN]

Hopefully soon, PCs will do some of this

 

[griptwister]

Hopefully soon, PCs will do some of this

What if AMD used DDR5 with Excavator? LOL! Wouldn't that be a trip! (if they even get that far, hopefully they will) :lol:

Any news on PD refresh??? What I wanna know is, can AMD make minor adjustments to the archistructure and still call it a Refresh? Because that's the only way how I see them dropping Latency times and getting better FPS.

 

[de5_Roy]

AMD to Showcase Fusion “Temash”-Based Tablets, Convertibles with Turbo Dock Technology.
AMD to Demo Temash-Based Devices at MWC Next Week
http://www.xbitlabs.com/news/mobile/display/20130220142436_AMD_to_Showcase_Fusion_Temash_Based_Tablets_Convertibles_with_Turbo_Dock_Technology.html

Sony confirms PlayStation 4's PC guts
http://techreport.com/news/24387/sony-confirms-playstation-4-pc-guts

Sony Announces PlayStation 4: PC Hardware Inside
http://www.anandtech.com/show/6770/sony-announces-playstation-4-pc-hardware-inside
^^ i smells hsa. hopefully sony will push hsa into mainstream like they did with blu-ray.

AMD "Jaguar" Micro-architecture Takes the Fight to Atom with AVX, SSE4, Quad-Core
http://www.techpowerup.com/180394/AMD-quot-Jaguar-quot-Micro-architecture-Takes-the-Fight-to-Atom-with-AVX-SSE4-Quad-Core.html
^^ the instruction set support will put some serious hurt on atom. internal bw seems to have improved a lot. but, 15% typical ipc improvement from 40nm to 28nm sounds underwhelming.

JPR: PC graphics shipments fell in Q4
http://techreport.com/news/24383/jpr-pc-graphics-shipments-fell-in-q4
wtf. despite having the best apus and great gfx cards, intel Gained on amd. intel even spectacularly failed ivy bridge demos. 🙁

 

[mayankleoboy1]

^^ the instruction set support will put some serious hurt on atom. internal bw seems to have improved a lot. but, 15% typical ipc improvement from 40nm to 28nm sounds underwhelming.

Instruction set dont mean anything on their own. They need to be actively kept in mind during coding. A major part of SSE3/4 AVX and FMA is to use parallelism in the code. Where again, coders are the bottleneck.
Piledriver has advanced instructions like BMI and FMA. And no consumer software uses them. And i doubt any scientific software uses them either.

Compiling with -bdver2 gives little improvement in the code.
Incidentall, compiling with -core-avx-i (which uses AVX and Bull Mountain) also offers little benefit. Most softwares see best improvement with SSE3.

 

[de5_Roy]

Instruction set dont mean anything on their own. They need to be actively kept in mind during coding. A major part of SSE3/4 AVX and FMA is to use parallelism in the code. Where again, coders are the bottleneck.
Piledriver has advanced instructions like BMI and FMA. And no consumer software uses them. And i doubt any scientific software uses them either.

Compiling with -bdver2 gives little improvement in the code.
Incidentall, compiling with -core-avx-i (which uses AVX and Bull Mountain) also offers little benefit. Most softwares see best improvement with SSE3.

does that include server softwares as well? i heard that jaguar might make its way into microservers.

 

[kettu]

Don't just look at the FX-8350, check out the FX-6300, its trading blows with the i5 2500k, and right behind the 3570k!

Indeed. Also worth noting is the poor performance of Intel i3-3220 which I've often seen recommended even over the FX-6300. Even hyperthreading won't save i3 in this game. The wave of the future I guess.

 

[de5_Roy]

the wave of the future is 720p gaming (the crysis 3 bench) with no af/aa? i guess it's futuristic and innovative only when amd 'wins' the benchmark..... ;-D
people looking for more gaming benches featuring this kind of innovation should check out some of the ocaholic cpu benches, amd wins a lot of 'em.

 

[kettu]

the wave of the future is 720p gaming (the crysis 3 bench) with no af/aa? i guess it's futuristic and innovative only when amd 'wins' the benchmark..... ;-D
people looking for more gaming benches featuring this kind of innovation should check out some of the ocaholic cpu benches, amd wins a lot of 'em.

Well, GameGPU.ru tested at 1080p with similar results. Besides do you expect the i3 to gain performance at higher resolutions/settings? We're not talking about some meaningles 80 fps vs 100+ fps difference between CPUs at 720p. i3 min fps is quite pitiful 22fps and averages only ~32 fps vs 41/46 for fx-6300 (for similar pricing by the way). If your CPU isn't capable even at 720p then it certainly won't cut the mustard at higher resolutions and settings.

I'll stand by my statement, wave of the future, man, wave of the future...

 

[gamerk316]

Thats the point, it isnt easy, and it isnt the way weve always done it, but it can be done.
Leave it to crytek to innovate

Counter-argument: Performance more or less matches clockspeed [3960x being the exception] until you hit the older CPU's, where IPC decreases become a performance factor. Test an i5-3770k @ 4GHz (or even the 3570k), and I would expect it to leapfrog the FX-8350 based on the scaling I'm seeing.

In short: I see higher clocked FX processors beating lower clocked i5s. At equal clocks, I would expect the i5s to win. Hence, Cysis 3 looks sensitive to clock speed.

Anyone care to bench?

 

[gamerk316]

Instruction set dont mean anything on their own. They need to be actively kept in mind during coding. A major part of SSE3/4 AVX and FMA is to use parallelism in the code. Where again, coders are the bottleneck.
Piledriver has advanced instructions like BMI and FMA. And no consumer software uses them. And i doubt any scientific software uses them either.

Compiling with -bdver2 gives little improvement in the code.
Incidentall, compiling with -core-avx-i (which uses AVX and Bull Mountain) also offers little benefit. Most softwares see best improvement with SSE3.

Which makes sense: Most compilers don't support AVX yet. Nevermind there aren't a lot of situations where you can replace in AVX calls...

 

[gamerk316]

I wont think it will be much sooner , AMD reports 32% decline in Q4 revenue, net loss widens

we need to be awaited and can expect it late 2014.... Agree ??

for more infromation about AMD decline

http://servers.cbronline.com/news/amd-reports-32-decline-in-q4-revenue-net-loss-widens-230113

eek. With only about $1.2 Billion Cash on Hand, increasing losses like this are NOT sustainable. AMD simply can NOT afford ~$400 Million quarterly losses right now.

 

[gamerk316]

(Quad posting, I know)

On the PS4 hardware: CPU wise, looks like Jaguar, as suspected. 8-cores, undetermined clocks (probably about 1.6 GHz or so). IPC will determine if this is a net gain over the Cell or not (would have to have about 2x the IPC of PCC, which is about what X86 gets), though the ease of coding over Cell will likely net a bit extra increase. Its a better CPU, but not a huge leap in performance.

On the GPU, thats where 90% of the performance is. Looks like performance roughly equivalent to a 7850/7970. So at least 4x the power of the PS3 GPU (modified 7800GTX).

The 8GB GDDR5 RAM is shared between the system and GPU. This is a good thing in some regards (CPU has direct access to the GPU framebuffer this way), but I worry about CPU latency compared to DDR3 (GDDR tends to have high bandwidth, low latency. DDR is the opposite). This could end up being a significant limiting factor, as memory reads/writes are going to be costly. As it doesn't appear the PS4 will support task switching like the 720, its likely all that RAM (or at least a good sized chunk of it) will be free for developer use, so the per-application RAM limit should be higher then the 720.

My major concern, again, is that the closer to PC hardware will lead devs to not take advantage of the fact they can code very close to metal, and lose a lot of performance as a result.

 

[JAYDEEJOHN]

Maybe at first, as always, but looking at the old consoles, and what they can do with them, add in the improvements of the new consoles, it will only get better

 

[gamerk316]

In all honesty, I'm convinced we are fast approaching a 1983 style reduction of the gaming industry. Smartphones have changed EVERYTHING. The casual market that made the old style consoles successful is gone to the smartphone market. Developers are few and focused on make or break projects (Homefront), and several are loosing money (EA). As a result, rather then innovate, they keep doing the same thing, because it works. Until it doesn't (consumer fatigue). On top of that, Sony is loosing money, Nintendo has serious financial problems, and even MSFT isn't makeing that much on its gaming unit.

I fully expect this to be the last major console generation, at least for a while.

 

[Cazalan]

So the 8xJaguar is official now. I'm a bit surprised but it seems the console makers were more confident in pushing multiple weak cores, than depending on faster strong cores. Another ding for the Bulldozer modules even with Steamroller enhancements.

It will be interesting to see what those cores can do with a relatively high bandwidth memory interface. That has been a weakness for these low end APUs with just single or dual channel DDR3 interface.

 

[kettu]

Counter-argument: Performance more or less matches clockspeed [3960x being the exception] until you hit the older CPU's, where IPC decreases become a performance factor. Test an i5-3770k @ 4GHz (or even the 3570k), and I would expect it to leapfrog the FX-8350 based on the scaling I'm seeing.

In short: I see higher clocked FX processors beating lower clocked i5s. At equal clocks, I would expect the i5s to win. Hence, Cysis 3 looks sensitive to clock speed.

Anyone care to bench?

Let's look at the numbers within CPU families:
i3-3220: Two cores, with hyperthreading 3.3Ghz clockspeed
i5-3570k: Four cores, no hyperthreadin, 3.4Ghz clockspeed with turbo upto 3.8Ghz

Let's be generous and assume that the i5 can hit the maximum turbo frequency constantly. That's going to be ~15% higher clockspeed compared to i3. Yet it performs on average ~61% faster while the minimum fps is 100% higher.

Also compare the quad-core i5-760 with it's older architecture and lower clockspeed (max turbo at 4 cores 2.933Ghz) it still outperforms the i3. Then there's even older C2Q @ 2.83Ghz but still almost equals the dual-core i3.

Clearly the core count is much more significant than clockspeed.

We can do the same comparison with AMD:
X4 965: Four cores, 3.4Ghz
X6 1100t: Six cores, 3.3Ghz, turbo upto 3.7Ghz

In this case it's clear the X6 can't utilize the turbo since more than 3 cores are active according to gamegpu.ru. Despite the (insignificant) clockspeed disadvantage the X6 is on average ~22% faster and the minimum FPS is 30% higher. The difference between X6 turbo and X4 stock clock is ~9% so even that wouldn't explain the difference.

Again we see that core count is more significant than clockspeed.

 

[Blandge]

First off, the warranty is 100% void for removing the lid, 2nd, approx 50% of the attempts to recreate this scenario resulted in dead cpus,

This is undoubtedly true.

and others saw gains of maybe 7c. One extreme case by an obviously extremist individual does not mean all attempts will be the same at the end.

20 degrees isn't even the best results.

3770K De-lidded and lapped With Liquid Pro - Awesome results.

Max Core 1 = 53c (-23c)

Max Core 2 = 56c (-32c)

Max Core 3 = 58c (-25c)

Max Core 4 = 54c (-25c)
An incredible max improvement of 32 Degrees C!

But it does show that you can reduce temps by using extreme measures such as reducing the gap between the silicon and the lid (where he ground it with a dremel tool). Others have directly mounted the heatsink on the silicon, removing the lid alltogether.

Exactly, my point is that Intel has more options for better cooling. Just like there are seemingly insurmountable problems for transistor tech, there are similar problems for heat dissipation. At the same time there's solutions to both problems. Intel clearly recognizes the issue, and it will be addressed with each new process.

To address one of the things I said:

The lower overclockability of IVB has to do with 22nm finFET device characteristics, not heat dissipation.

I retract this statement as nonsensical as obviously heat dissipation is the device characteristic that gates overclocking (except in the case of extreme cooling). However, the difference in overclockability between 32 and 22nm can be severely reduced with better packaging technology. Why Intel decided to use TIM instead of soldering is anybody's guess. Was the incentive financial or technical?

In any case, heat density does not cause problems for IVB's 22nm process at stock frequencies and voltages, and it shouldn't cause any major problems at 14nm either (at stock).

 

[JAYDEEJOHN]

http://blogs.amd.com/fusion/2013/02/21/amd/

 

[Blandge]

Counter-argument: Performance more or less matches clockspeed [3960x being the exception] until you hit the older CPU's, where IPC decreases become a performance factor. Test an i5-3770k @ 4GHz (or even the 3570k), and I would expect it to leapfrog the FX-8350 based on the scaling I'm seeing.

In short: I see higher clocked FX processors beating lower clocked i5s. At equal clocks, I would expect the i5s to win. Hence, Cysis 3 looks sensitive to clock speed.

Anyone care to bench?

IPC IS a factor when comparing a FX-8350 to a i5 3570K such that the frequency advantage of the FX-8350 should not be able to overcome its IPC disadvantage (to the degree that we see in Crysis 3 benchmarks). This included with the performance gains seen with the i7-3960X points to the idea the Crysis 3 sees reasonable scaling with core count.

 

[griptwister]

http://blogs.amd.com/fusion/2013/02/21/amd/

So does this mean that AMD isn't releasing a 8 core APU??? I was actually thinking of purchasing an APU if they were actually going to make an 8 core version!