AMD CPU speculation... and expert conjecture

[jdwii]

core i3s don't have staying power. they're okay as starter gaming cpus but lose performance edge soon when demanding games come out. amd's starter gaming cpu is 750k, performs more or less lower core i3's equivalent, except those crazy 3.8GHz ones.
due to performance and upgrade reasons, neither intel nor amd have "upgradable" platforms. intel's lga1150 will soon be replaced by lga1151 and at the high end lga2011 (haswell-e's). meanwhile amd's entry level is isolated with socket fm2/+ and at it's high end, am3+. cpus like core i3, fx6300 (due to present pricing) occupy that gap.

The difference is with the Intel you can replace the cheap pentium or I3 with a I7 with a A10 7850K you can replace it with nothing better(that we know of). With AM3+ the 8350fx is the only realistic option and in a lot of cases the cheaper boards(anything under 150$) don't support anything better

 

[palladin9479]

core i3s don't have staying power. they're okay as starter gaming cpus but lose performance edge soon when demanding games come out. amd's starter gaming cpu is 750k, performs more or less lower core i3's equivalent, except those crazy 3.8GHz ones.
due to performance and upgrade reasons, neither intel nor amd have "upgradable" platforms. intel's lga1150 will soon be replaced by lga1151 and at the high end lga2011 (haswell-e's). meanwhile amd's entry level is isolated with socket fm2/+ and at it's high end, am3+. cpus like core i3, fx6300 (due to present pricing) occupy that gap.

The difference is with the Intel you can replace the cheap pentium or I3 with a I7 with a A10 7850K you can replace it with nothing better(that we know of). With AM3+ the 8350fx is the only realistic option and in a lot of cases the cheaper boards(anything under 150$) don't support anything better

That is nonsense, mostly in the sense that it's too expensive to replace parts now. That $110 USD i3 (or $80 USD Pentium) cost is added to the costs of the later purchased i5/i7 as part of the system total. So unless your using extremely expensive parts, it's simply not worth it in the long run. Better to upgrade the platform as a whole every 3 to 4 years. Anything in the "budget" sector is disposable by design and the reason why "upgrade path" is such utter bullsh!t. It's also why "future proof" is also bullsh1t, technology simply changes too fast.

Or are you seriously suggesting that someone building a budget system today will buy a 7750 three to four years from now?

 

[-Fran-]

That is nonsense, mostly in the sense that it's too expensive to replace parts now. That $110 USD i3 (or $80 USD Pentium) cost is added to the costs of the later purchased i5/i7 as part of the system total. So unless your using extremely expensive parts, it's simply not worth it in the long run. Better to upgrade the platform as a whole every 3 to 4 years. Anything in the "budget" sector is disposable by design and the reason why "upgrade path" is such utter bullsh!t. It's also why "future proof" is also bullsh1t, technology simply changes too fast.

Or are you seriously suggesting that someone building a budget system today will buy a 7750 three to four years from now?

Although I usually suggest the same thing you're saying (buy as expensive as one can and live with it until the time comes for another build), there are exceptions to the rule when building a system in a tight budget, but with the chance to change components down the road. As inefficient as it sounds and is, some people just have that choice.

For instance, AM3 had a great upgrade path from the Phenom II X4 in 65nm (I think) until Bulldozer (with some boards accepting PD -> Crosshair IV with a modded BIOS). And you can always swap video cards anytime you want and RAM, but I think you're focusing the discussion in CPUs. The other "upgrade path" could be to keep the CPU and swap a cheap 760G for a 990FX and stuff. That's a good 3 gen jump in stuff for your MoBo (audio, USB3 support and eSATA, for example) keeping the same CPU and GPU, but not much of a performance upgrade, haha.

A long lasting platform has its advantages, although I understand why Intel keeps swapping stuff around ever 1 or 2 gens of CPUs. Although sockets don't have merit (like we already discussed at some point).

And in regards to the GPU and CPU bottleneck, you're right, but you're just moving the "bottleneck" from one to the other by changing the balance inside the game engine. Point was that, as obvious as it sounds, there is a bottleneck when you look hard enough

Cheers!

EDIT: Typo.

 

[palladin9479]

Although I usually suggest the same thing you're saying (buy as expensive as one can and live with it until the time comes for another build), there are exceptions to the rule when building a system in a tight budget, but with the chance to change components down the road. As inefficient as it sounds and is, some people just have that choice.

In that scenario your just causing the entire thing to be even more expensive. It would be cheaper to just use credit to purchase what you need right up front rather then buying low under the expectation to upgrade "later". Same with GPU's and motherboards. Whatever you spend on that 760G is now wasted sunk costs when you purchase a 990FX, cheaper to just buy the 990FX upfront and pay the credit off. The interest from the higher initial purchase would be an order of magnitude lower then sunk costs of repurchasing hardware. And since we can't predict the future accurately enough to determine computing requirements it's a fools errand to try to plan today's purchase's around an upgrade in three to four years time. The only parts that can realistically be reused are non-performance parts like chassis's, monitor's and possibly PSU's.

Just imagine, a 750K or i3 purchased today with 8GB of memory and a 750ti GPU (mid range). In three to four years time, would it be economically viable to "upgrade" to a 950ti GPU (imaginary mid range) or try to reuse the same CPU / MB / memory? The wise financial decision by far is to build a new platform (CPU/MB/memory) and chose the right GPU for you at that time. This way the costs of computing can be amortized across the entire life span of the system which tends to result in lower total costs.

As far as APU's are concerned, they are disposable and should be treated as such. No APU should be priced over $140~150 USD as that's the point where it lose value. Under that amount it becomes the best solution for SFF / low profile / cost appliance devices. Take something like an M350 with 120~150w pico-PSU,8GB of DDR3 memory, an A8-7600 and a 2.5 inch HDD. That is an incredibly small yet versatile platform that consumes little power, produces little heat and can be tasked to do any number of things. Light / low end mobile is also where APU's work out really well due to their better space vs power vs performance vs cost ratios'. After that they kind of become a waste, no point in building a desktop around an APU as you have enough space, power and budget to put a GDDR5 dGPU inside.

 

[gamerk316]

you guys know that once mantle goes wide and dx12 comes out, core i3 and fx will become interchangeable. i3 will still have worse non-gaming app performance compared to hex cores and higher. fx will still have higher power consumption and higher app performance. reduction of cpu bottleneck doesn't help amd alone.

To be fair, the i3 can spit out good FPS numbers because over a long time period, any flatlines get averaged out. But minimum FPS wise, and latency wise, its going to lag behind FX, even if it puts up similar FPS numbers. I'd still be interested to see what one could go at about 4.5 though...

 

[anxiousinfusion]

No APU should be priced over $140~150 USD as that's the point where it lose value. Under that amount it becomes the best solution for SFF / low profile / cost appliance devices.

Funny you say that, given the A10-7800 is being introduced http://www.techpowerup.com/202592/amd-unveils-a10-7800-quad-core-socket-fm2-apu.html in the ballpark of $140. It's basically a 7850K.

 

[gamerk316]

That is nonsense, mostly in the sense that it's too expensive to replace parts now. That $110 USD i3 (or $80 USD Pentium) cost is added to the costs of the later purchased i5/i7 as part of the system total. So unless your using extremely expensive parts, it's simply not worth it in the long run. Better to upgrade the platform as a whole every 3 to 4 years. Anything in the "budget" sector is disposable by design and the reason why "upgrade path" is such utter bullsh!t. It's also why "future proof" is also bullsh1t, technology simply changes too fast.

Or are you seriously suggesting that someone building a budget system today will buy a 7750 three to four years from now?

Cosigned. It costs more money over the long haul to do small upgrades, rather then doing it right the first time. The latter is easier to fit in a budget though.

I generally build a platform to last 4-5 years. I go overkill on the CPU, and do a GPU upgrade after about 2 years or so. I literally went: Pentium 4 @ 1.6 -> Pentium D -> QX9650 -> 2600k. And the 2600k was totally unplanned, due in large part of making the mistake of using the nforce 790i as my platform, which was VERY unstable.

 

[jdwii]

core i3s don't have staying power. they're okay as starter gaming cpus but lose performance edge soon when demanding games come out. amd's starter gaming cpu is 750k, performs more or less lower core i3's equivalent, except those crazy 3.8GHz ones.
due to performance and upgrade reasons, neither intel nor amd have "upgradable" platforms. intel's lga1150 will soon be replaced by lga1151 and at the high end lga2011 (haswell-e's). meanwhile amd's entry level is isolated with socket fm2/+ and at it's high end, am3+. cpus like core i3, fx6300 (due to present pricing) occupy that gap.

The difference is with the Intel you can replace the cheap pentium or I3 with a I7 with a A10 7850K you can replace it with nothing better(that we know of). With AM3+ the 8350fx is the only realistic option and in a lot of cases the cheaper boards(anything under 150$) don't support anything better

That is nonsense, mostly in the sense that it's too expensive to replace parts now. That $110 USD i3 (or $80 USD Pentium) cost is added to the costs of the later purchased i5/i7 as part of the system total. So unless your using extremely expensive parts, it's simply not worth it in the long run. Better to upgrade the platform as a whole every 3 to 4 years. Anything in the "budget" sector is disposable by design and the reason why "upgrade path" is such utter bullsh!t. It's also why "future proof" is also bullsh1t, technology simply changes too fast.

Or are you seriously suggesting that someone building a budget system today will buy a 7750 three to four years from now?

Well for me i upgrade when i can money wise not just buy and leave it, Only a 8350fx-I5 with a 280X-770GTX would be good enough to do that for 3 years. Some people only have 400-500$ to spend now and then upgrade. For that much you either have a 60$ Pentium and a 100$ 250X and it will be upgradable in the future when they can afford more.

Edit also technology moves to fast? If anything its slow right now. A I7 920 from 2008 still keeps up in all the latest games that is 6 years old. The best video card back in 2008, However in that year we where still in the 9800GTX(4870) day's.

 

[jdwii]

That is nonsense, mostly in the sense that it's too expensive to replace parts now. That $110 USD i3 (or $80 USD Pentium) cost is added to the costs of the later purchased i5/i7 as part of the system total. So unless your using extremely expensive parts, it's simply not worth it in the long run. Better to upgrade the platform as a whole every 3 to 4 years. Anything in the "budget" sector is disposable by design and the reason why "upgrade path" is such utter bullsh!t. It's also why "future proof" is also bullsh1t, technology simply changes too fast.

Or are you seriously suggesting that someone building a budget system today will buy a 7750 three to four years from now?

Although I usually suggest the same thing you're saying (buy as expensive as one can and live with it until the time comes for another build), there are exceptions to the rule when building a system in a tight budget, but with the chance to change components down the road. As inefficient as it sounds and is, some people just have that choice.

For instance, AM3 had a great upgrade path from the Phenom II X4 in 65nm (I think) until Bulldozer (with some boards accepting PD -> Crosshair IV with a modded BIOS). And you can always swap video cards anytime you want and RAM, but I think you're focusing the discussion in CPUs. The other "upgrade path" could be to keep the CPU and swap a cheap 760G for a 990FX and stuff. That's a good 3 gen jump in stuff for your MoBo (audio, USB3 support and eSATA, for example) keeping the same CPU and GPU, but not much of a performance upgrade, haha.

A long lasting platform has its advantages, although I understand why Intel keeps swapping stuff around ever 1 or 2 gens of CPUs. Although sockets don't have merit (like we already discussed at some point).

And in regards to the GPU and CPU bottleneck, you're right, but you're just moving the "bottleneck" from one to the other by changing the balance inside the game engine. Point was that, as obvious as it sounds, there is a bottleneck when you look hard enough

Cheers!

EDIT: Typo.

Agreed when you have the money however a nice high-end CPU and GPU will work but a lot of people buy gaming rigs for under 800$ and upgrade later when they have money. Again with Intel a 50$ board can handle a 330$ CPU just fine and a 1500$ GPU without any bottlenecks at all. Plus plenty of ram for gamers. I like having a upgrade path and not being stuck my AM3+ board lasted quite a long time actually and it will still make it longer as long as the 8350fx stay's decent during gaming and the PCI-E 2.0 isn't much of a bottleneck.

another thing i heard was when he claimed you pay more at the end well yeah? People can't always shed that much cash at once that is why we have loans for houses and cars sure you pay more but you get it now. Many people would love to spend 4000$ on a PC. Another thing to remember when you buy a 1000$ video card it will probably become obsolete in 3 years anyways like a 6990 did with directx 12 and mantle.

 

[palladin9479]

Spoiler

core i3s don't have staying power. they're okay as starter gaming cpus but lose performance edge soon when demanding games come out. amd's starter gaming cpu is 750k, performs more or less lower core i3's equivalent, except those crazy 3.8GHz ones.
due to performance and upgrade reasons, neither intel nor amd have "upgradable" platforms. intel's lga1150 will soon be replaced by lga1151 and at the high end lga2011 (haswell-e's). meanwhile amd's entry level is isolated with socket fm2/+ and at it's high end, am3+. cpus like core i3, fx6300 (due to present pricing) occupy that gap.

The difference is with the Intel you can replace the cheap pentium or I3 with a I7 with a A10 7850K you can replace it with nothing better(that we know of). With AM3+ the 8350fx is the only realistic option and in a lot of cases the cheaper boards(anything under 150$) don't support anything better

That is nonsense, mostly in the sense that it's too expensive to replace parts now. That $110 USD i3 (or $80 USD Pentium) cost is added to the costs of the later purchased i5/i7 as part of the system total. So unless your using extremely expensive parts, it's simply not worth it in the long run. Better to upgrade the platform as a whole every 3 to 4 years. Anything in the "budget" sector is disposable by design and the reason why "upgrade path" is such utter bullsh!t. It's also why "future proof" is also bullsh1t, technology simply changes too fast.

Or are you seriously suggesting that someone building a budget system today will buy a 7750 three to four years from now?

Well for me i upgrade when i can money wise not just buy and leave it, Only a 8350fx-I5 with a 280X-770GTX would be good enough to do that for 3 years. Some people only have 400-500$ to spend now and then upgrade. For that much you either have a 60$ Pentium and a 100$ 250X and it will be upgradable in the future when they can afford more.

Edit also technology moves to fast? If anything its slow right now. A I7 920 from 2008 still keeps up in all the latest games that is 6 years old. The best video card back in 2008, However in that year we where still in the 9800GTX(4870) day's.

You are making a foolish long term economical decision. Spending $200 to upgrade a $400 build gives it the cost of a $600 build but with the capability of a $500 build. It would of been cheaper to use an additional $100 in credit on top of the original $400 to achieve that $500 build, then pay off the $100 credit. You get an extra $95 (interest on the credit) in your pocket. This is why small incremental upgrades are bad economic decisions. Only major upgrades every three to four (or five for some people) years make long term sense. Constantly chasing product releases is how enthusiasts end up wasting tons of money. I challenge anyone to come up with a budget "upgrade" scenario and I'll demonstrate, with numbers, how it's financially a bad decision and how it would be better to use a small personal loan or credit card to just start with what you need.

Funny you say that, given the A10-7800 is being introduced http://www.techpowerup.com/202592/amd-unveils-a10-7800-quad-core-socket-fm2-apu.html in the ballpark of $140. It's basically a 7850K.

The 7850K is not good because it's priced well outside of it's value range. I used the range because the iGPU on those is about the strength of a sub $100 DDR3 dGPU along with their CPU being the strength of a sub $100 USD dCPU. The venerated Athlon 750K is just an APU with the iGPU removed. So the value range must be below the cost of a budget dCPU and a GDDR5 dGPU since the dGPU will always give superior performance. That number ends up being right at $140 USD though the lower the better. It's why I was so excited about the A8-7600 priced at ~$125 USD. At that range there is nothing that comes close to competing with it for value. Unfortunately AMD deliberately pushed back the release date in order to give room for early adopters to buy the higher margin 7850K and 7700K.

 

[gamerk316]

You are making a foolish long term economical decision. Spending $200 to upgrade a $400 build gives it the cost of a $600 build but with the capability of a $500 build. It would of been cheaper to use an additional $100 in credit on top of the original $400 to achieve that $500 build, then pay off the $100 credit. You get an extra $95 (interest on the credit) in your pocket. This is why small incremental upgrades are bad economic decisions. Only major upgrades every three to four (or five for some people) years make long term sense. Constantly chasing product releases is how enthusiasts end up wasting tons of money. I challenge anyone to come up with a budget "upgrade" scenario and I'll demonstrate, with numbers, how it's financially a bad decision and how it would be better to use a small personal loan or credit card to just start with what you need.

But people only see the NOW, and thus make poor long-term financial decisions.

 

[jdwii]

You are making a foolish long term economical decision. Spending $200 to upgrade a $400 build gives it the cost of a $600 build but with the capability of a $500 build. It would of been cheaper to use an additional $100 in credit on top of the original $400 to achieve that $500 build, then pay off the $100 credit. You get an extra $95 (interest on the credit) in your pocket. This is why small incremental upgrades are bad economic decisions. Only major upgrades every three to four (or five for some people) years make long term sense. Constantly chasing product releases is how enthusiasts end up wasting tons of money. I challenge anyone to come up with a budget "upgrade" scenario and I'll demonstrate, with numbers, how it's financially a bad decision and how it would be better to use a small personal loan or credit card to just start with what you need.

But people only see the NOW, and thus make poor long-term financial decisions.

Agree a 100% i wasn't saying he was wrong i was just telling him how it is. These people who buy gaming rigs are not always like console guys and expect a 5-7 year upgrade cycle. A lot jump between gen's like buying a 5870 and then upgrad to a 7970. What makes PC's great is the choice to upgrading instead of starting all over.

Also how can you convince a console gamer to spend 800$ instead of telling them you can save money on games with steam and put that money towards better hardware down the line. People do not want to always use credit for this or that i hate using credit for anything.(plus you can sell your stuff easier then as well i sold my 1100T for 100$ in less then a week giving me a extra 80$ investment on my newer CPU)
For 500$ You can either get this

Spoiler

Case: 29.99$ (Rosewill QN100 Dual Fans ATX Mid Tower Computer Case, come with 1x Front 120mm Fan, 1x Rear 120mm Fan)
http://www.newegg.com/Product/Product.aspx?Item=N82E16811147199
Power Supply: 49.99$ (CORSAIR CX series CX500 500W ATX12V v2.3 80 PLUS BRONZE Certified Active PFC Power Supply)
http://www.newegg.com/Product/Product.aspx?Item=N82E16817139027
Motherboard: 44.99$ (ASRock H81M-DGS R2.0 LGA 1150 Intel H81 SATA 6Gb/s USB 3.0 Micro ATX Intel Motherboard)
http://www.newegg.com/Product/Product.aspx?Item=N82E16813157483
Ram: 74.99$ (G.SKILL Value 8GB (2 x 4GB) 240-Pin DDR3 SDRAM DDR3 1600 (PC3 12800) Desktop Memory Model F3-1600C11D-8GNT)
http://www.newegg.com/Product/Product.aspx?Item=N82E16820231635&ignorebbr=1
Hard Drive: 57.99$ (Western Digital WD Blue WD10EZEX 1TB 7200 RPM 64MB Cache SATA 6.0Gb/s 3.5" Internal Hard Drive Bare Drive - OEM)
http://www.newegg.com/Product/Product.aspx?Item=N82E16822236339
CPU: 74.99$ (Intel Pentium G3258 Haswell Dual-Core 3.2GHz LGA 1150 Desktop Processor BX80646G3258)
http://www.newegg.com/Product/Product.aspx?Item=N82E16819117374
GPU: 159.99$ (SAPPHIRE DUAL-X 100365L Radeon R9 270 2GB 256-Bit GDDR5 PCI Express 3.0 Video Card With BOOST & OC)
http://www.newegg.com/Product/Product.aspx?Item=N82E16814202090
Total: 493$

Or this from Amd

Spoiler

Case: 29.99$ (Rosewill QN100 Dual Fans ATX Mid Tower Computer Case, come with 1x Front 120mm Fan, 1x Rear 120mm Fan)
http://www.newegg.com/Product/Product.aspx?Item=N82E16811147199
Power Supply: 44.99$ (CORSAIR CX series CX430 430W ATX12V v2.3 80 PLUS BRONZE Certified Active PFC Power Supply)
http://www.newegg.com/Product/Product.aspx?Item=N82E16817139026
Motherboard: 59.99$ (MSI 970A-G43 AM3+ AMD 970 + SB950 SATA 6Gb/s USB 3.0 ATX AMD Motherboard)
http://www.newegg.com/Product/Product.aspx?Item=N82E16813157483
Ram: 74.99$ (G.SKILL Value 8GB (2 x 4GB) 240-Pin DDR3 SDRAM DDR3 1600 (PC3 12800) Desktop Memory Model F3-1600C11D-8GNT)
http://www.newegg.com/Product/Product.aspx?Item=N82E16820231635&ignorebbr=1
Hard Drive: 57.99$ (Western Digital WD Blue WD10EZEX 1TB 7200 RPM 64MB Cache SATA 6.0Gb/s 3.5" Internal Hard Drive Bare Drive - OEM)
http://www.newegg.com/Product/Product.aspx?Item=N82E16813130679
CPU: 119.99$ (AMD FX-6300 Vishera 6-Core 3.5GHz (4.1GHz Turbo) Socket AM3+ 95W Desktop Processor FD6300WMHKBOX)
http://www.newegg.com/Product/Product.aspx?Item=N82E16819113286
GPU: 109.99$ (SAPPHIRE 100366-2L Radeon R7 260X 2GB 128-Bit GDDR5 PCI Express 3.0 CrossFireX Support Video Card)
http://www.newegg.com/Product/Product.aspx?Item=N82E16814202081
Total: 498$

Picking the Intel one gives you a 32% more powerful GPU

Spoiler

Then we can conclude that the CPU will be less of an bottleneck with Directx 12 and mantle. Personally i would never want a gaming rig worth less then 800$ anyways.

 

[szatkus]

I hope AMD does catch up soon. I want to see competition. I don't really care what xxnm they are on. I just want performance and I don't want an extra space heater in the summer. If AMD could at least match Haswell, I would be pretty damn impressed.

Forget haswell, on par with skylake is coming...

Maybe they showed performance numbers of next-gen core, but I doubt that you know anything about Skylake performance.

 

[vmN]

There was some rumors that AMDs next x86 architecture would feature 6 ALUs per core (Haswell have 4 ALUs per core. Piledriver have 2 ALUs per ALU cluster)
I doubt it would feature AVX512 (skylake will be the first to feature it, and AMD is normally a generation behind with AVX support), so it might still be stuck with 256bit SIMD clusters.

They surely would improve the frontend alot (as It will now be dedicated to a single backend) so we could suspect seeing 4 decoders per core. They would certainly also need to improve the fetch so it can support AVX2 and SMT.

They would need a faster cache system to keep the SIMD cluster fed with resources when working with heavier AVX workloads, so I suspect much faster L1 and L2 cache.

I somehow still doubt AMD manage to get 6 ALUs per core, I believe we would instead see 4 ALUs per core.

 

[juanrga]

There was some rumors that AMDs next x86 architecture would feature 6 ALUs per core (Haswell have 4 ALUs per core. Piledriver have 2 ALUs per ALU cluster)
I doubt it would feature AVX512 (skylake will be the first to feature it, and AMD is normally a generation behind with AVX support), so it might still be stuck with 256bit SIMD clusters.

They surely would improve the frontend alot (as It will now be dedicated to a single backend) so we could suspect seeing 4 decoders per core. They would certainly also need to improve the fetch so it can support AVX2 and SMT.

They would need a faster cache system to keep the SIMD cluster fed with resources when working with heavier AVX workloads, so I suspect much faster L1 and L2 cache.

I somehow still doubt AMD manage to get 6 ALUs per core, I believe we would instead see 4 ALUs per core.

I expect 3 ALUs, 2 FMAC units, and 4 decoders per core. I also expect a stack cache.

 

[juanrga]

I used to believe that claim up until October last year. The consoles appear to be demonstrably lower than even that considering they can barely handle games despite compromising resolution and/or framerate.

First gen games are never representative of consoles hardware. Consider that gaming PC evolution is based in hardware upgrades (some people here is discussing how and when to upgrade), but console gaming is not based in that, because the hardware is fixed along the lifetime of the console.

Console gaming evolution is based in subsequent hardware optimization by devs. Newer games use more of the hardware. Thus expect future games to show the real potential of the hardware in consoles.

 

[palladin9479]

There was some rumors that AMDs next x86 architecture would feature 6 ALUs per core (Haswell have 4 ALUs per core. Piledriver have 2 ALUs per ALU cluster)
I doubt it would feature AVX512 (skylake will be the first to feature it, and AMD is normally a generation behind with AVX support), so it might still be stuck with 256bit SIMD clusters.

They surely would improve the frontend alot (as It will now be dedicated to a single backend) so we could suspect seeing 4 decoders per core. They would certainly also need to improve the fetch so it can support AVX2 and SMT.

They would need a faster cache system to keep the SIMD cluster fed with resources when working with heavier AVX workloads, so I suspect much faster L1 and L2 cache.

I somehow still doubt AMD manage to get 6 ALUs per core, I believe we would instead see 4 ALUs per core.

Adding ALU's is easy, they are very small and don't take much to run. The hard part is keeping those ALU's busy doing something useful. That is what requires complicated front end decoding / scheduling / predicting and an fast intelligent back end cache subsystem.

This actually brings up a very good point. We've been doing "parallel execution" in hardware for well over a decade now. A single stream of code is broken into multiple streams that are then executed simultaneously inside the CPU with the results being returned to the OS, a form of hardware multi-threading. Intel relies on the hardware to do all this thread work while AMD relied on the software. The software failed to properly utilize the underlying hardware while Intel's hardware didn't. And thus the "bulldozer is sh!t !!1010101" line was born.

 

[colinp]

The Bulldozer is sh1t line was born because Bulldozer was sh1t. I can't wait for AMD to ditch it and focus all their resources on post BD lines.

To expect them to reach Intel current gen levels is wishful thinking. For them to reach parity with Skylake (simultaneously) from their current position would require a miracle that would leave Jesus scratching his head, wondering how they'd managed it.

 

[palladin9479]

The Bulldozer is sh1t line was born because Bulldozer was sh1t. I can't wait for AMD to ditch it and focus all their resources on post BD lines.

Yet it wasn't "sh!t", it was quite good for what it was designed for. Unfortunately server style design's don't translate well into consumer space and thus the quality of BD's design was judged by what one module could do vs what all four could do. The total performance you got from BD was very good considering it only costs $180~200 USD. It's single biggest flaw was that it relied on software to make efficient use of available processing resources and that software simple didn't exist.

 

[colinp]

You can't blame the programmer for a) AMD's design choices; b) the terrible cache, FPU and predictor performance.

 

[vmN]

the SIMD cluster isn't the biggest issue and I wonder when people started blaming the bad performance on it. Of course a standalone SIMD cluster for each core is a better solution, but it is not the end of the world. (atleast for the general user. Different people different needs)

 

[logainofhades]

AMD is relying in a modular design and automated tools because lacks the money for a monolithic hand-crafted design. Automated design will reduce performance by about 20%. I don't know the penalty of using a modular approach.

I believe the modular design of Faildozer is going away, and will not be used in the new arch AMD is working on. It might stick around for APU's for awhile longer though.

I generally build a platform to last 4-5 years. I go overkill on the CPU, and do a GPU upgrade after about 2 years or so. I literally went: Pentium 4 @ 1.6 -> Pentium D -> QX9650 -> 2600k. And the 2600k was totally unplanned, due in large part of making the mistake of using the nforce 790i as my platform, which was VERY unstable.

I tend to upgrade every couple of years or so, CPU wise, unless I am forced to upgrade due to a hardware failure. That is how I ended up with a 3570k. My i5 750 motherboard was going out, in my file server, so I swapped it with my i5 2400 and P67 pro4 after a trip to Microcenter had me coming home with an i5 3570k and z77 extreme4, when I had originally planned on an i3 or pentium for the file server. :lol: It is nearly impossible to resist those motherboard bundles though. That and I was lazy. Didn't have to swap heatsinks around from my i5 2400 rig. I just used the Hyper 212 plus that was on the i5 750.

You are making a foolish long term economical decision. Spending $200 to upgrade a $400 build gives it the cost of a $600 build but with the capability of a $500 build. It would of been cheaper to use an additional $100 in credit on top of the original $400 to achieve that $500 build, then pay off the $100 credit. You get an extra $95 (interest on the credit) in your pocket. This is why small incremental upgrades are bad economic decisions. Only major upgrades every three to four (or five for some people) years make long term sense. Constantly chasing product releases is how enthusiasts end up wasting tons of money. I challenge anyone to come up with a budget "upgrade" scenario and I'll demonstrate, with numbers, how it's financially a bad decision and how it would be better to use a small personal loan or credit card to just start with what you need.

It isn't a foolish long term decision, for me anyway. Anything I replace in my gaming rig ends up finding its way into my other systems, which I am up to 5 of now. :lol: (One of them is going to my sister soon, so she has a PC for Sims 4. I imagine my old core 2 e8190 with a GT 240, maybe a GTS 450, will be sufficient for it.) So not only am I upgrading my main rig with a $200-$250 upgrade, on of my other systems gets an upgrade as well. Like how I replaced the GT 240's that were in my file server and one AMD rig with HD 5850's after I went from CF HD 5850's to a single HD 7970.

 

[szatkus]

There was some rumors that AMDs next x86 architecture would feature 6 ALUs per core (Haswell have 4 ALUs per core. Piledriver have 2 ALUs per ALU cluster)
I doubt it would feature AVX512 (skylake will be the first to feature it, and AMD is normally a generation behind with AVX support), so it might still be stuck with 256bit SIMD clusters.

They surely would improve the frontend alot (as It will now be dedicated to a single backend) so we could suspect seeing 4 decoders per core. They would certainly also need to improve the fetch so it can support AVX2 and SMT.

They would need a faster cache system to keep the SIMD cluster fed with resources when working with heavier AVX workloads, so I suspect much faster L1 and L2 cache.

I somehow still doubt AMD manage to get 6 ALUs per core, I believe we would instead see 4 ALUs per core.

Maybe 6 ALUs per module?

Ok, any source for this rumor?

 

[juanrga]

We've been doing "parallel execution" in hardware for well over a decade now. A single stream of code is broken into multiple streams that are then executed simultaneously inside the CPU with the results being returned to the OS, a form of hardware multi-threading. Intel relies on the hardware to do all this thread work while AMD relied on the software. The software failed to properly utilize the underlying hardware while Intel's hardware didn't. And thus the "bulldozer is sh!t !!1010101" line was born.

You are here mixing ILP with SMT and CMT. Both Intel and AMD rely on hardware for ILP, whereas both rely on software for multi-threading parallelism: AMD uses CMT and Intel uses SMT.

The traditional problem for AMD in the desktop/laptop is that average software doesn't use eight or more threads, but this problem only affected to flagship FX-8000 series processors, which performed as FX-6000 or even as FX-4000 under that software. Dual-core and quad-core APUs sold very well.

The "bulldozer is sh!t !!1010101" is a correct characterization of the architecture.

Yet it wasn't "sh!t", it was quite good for what it was designed for. Unfortunately server style design's don't translate well into consumer space and thus the quality of BD's design was judged by what one module could do vs what all four could do. The total performance you got from BD was very good considering it only costs $180~200 USD. It's single biggest flaw was that it relied on software to make efficient use of available processing resources and that software simple didn't exist.

It is just the contrary. Precisely it is in servers where Bulldozer/Piledriver failed most. AMD lost the market share that won with previous designs and currently has about 3% of servers. The server software is well-threaded, Intel has been selling 24-threads Xeon CPUs. The failure is not in the software but in Bulldozer/Piledriver.

Fortunately for us, the previous plans to release Steamroller and Excavator Opteron CPUs were canceled because would hurt the company still more.

Bulldozer single biggest flaw is that it is an unbalanced and inconsistent design. It is good that AMD fired the people responsible of this fiasco. Also Bulldozer was cheap because AMD was loosing money with each sale. AMD went into the red numbers and currently the CPU division is still losing money.

AMD is relying in a modular design and automated tools because lacks the money for a monolithic hand-crafted design. Automated design will reduce performance by about 20%. I don't know the penalty of using a modular approach.

I believe the modular design of Faildozer is going away, and will not be used in the new arch AMD is working on. It might stick around for APU's for awhile longer though.

I said "modular design" not modules design. The new arch is modular and relies in interchangeable parts: cores, cache, SIMD units, stacked ram... This modular design is the base for AMD semicustom business. AMD is not going to design monolithic dies anymore.

The post-excavator arch is not based in modules. I was the first one in this thread that said that Keller is abandoning the CMT approach of Buldozer/Piledriver/Steamroller/Excavator for the new architecture.

 

[juanrga]

Maybe 6 ALUs per module?

There is no modules.