Sign in with
Sign up | Sign in
Your question
Closed

AMD Piledriver release date?

Tags:
  • CPUs
  • AMD
Last response: in CPUs
Share
September 29, 2012 3:53:16 AM

Hi guys,

Has this actually been confirmed yet? To my knowledge it hasn't but they said sometime in Q3 so I'm expecting it to be pretty soon..

I'm going to end up buying the 8350 to replace my current CPU (sig).

Anyone heard anything?

More about : amd piledriver release date

a b à CPUs
September 29, 2012 4:32:29 AM

Since you overclock (assumption based on your overclocked CPU in sig, of course), just be aware that chances are that the 8350 won't be any better-binned than any of the cheaper FX-83xx models and the cheaper models would probably overclock to the same average frequency at the same average voltage and power consumption.

http://www.tomshardware.com/news/Vishera-Bulldozer-Pile...

We'll probably have these CPUs in October.
a c 152 à CPUs
a b À AMD
September 29, 2012 4:40:44 AM

It will be in the next week or two but as luciferano said the Piledriver won't be much better than what you have now.
Related resources
a b à CPUs
September 29, 2012 5:18:00 AM

rds1220 said:
It will be in the next week or two but as luciferano said the Piledriver won't be much better than what you have now.


I didn't say anything like that and I wouldn't because that'd be lying. Piledriver CPUs on AM3+ (Vishera) will be far more power efficient than Phenom II and will have greater performance in most metrics, perhaps all, depending on what CPUs you compare.
a c 152 à CPUs
a b À AMD
September 29, 2012 5:29:20 AM

If it has the performance AMD is claiming it will still put it at the level of a first generation I core processors. I would still rather have an overclocked Phenom II 955 or 965 BE.
a b à CPUs
September 29, 2012 5:39:33 AM

rds1220 said:
If it has the performance AMD is claiming it will still put it at the level of a first generation I core processors. I would still rather have an overclocked Phenom II 955 or 965 BE.


Take an FX-8120. Either disable the second core of each module (easier) or cut down its P states to very low levels (more difficult, less of a highly threaded performance hit). This alleviates the front-end bottle-neck of a Bulldozer module and can increase performance per primary core of each module by about 20-30% per Hz of the CPU frequency. Also, you're cutting cores that games wouldn't be using, so you shut them down and save power that would otherwise be wasted, meaning a two-pronged power efficiency boost that makes a great impact on power efficiency of the CPU.

Overclock the CPU/NB frequency to around 2.8GHz to 3GHz. The FX-8120 is now between Sandy Bridge and Nehalem/Westmere in performance per Hz per primary core of each module. Basically, you're treating it like a quad-core CPU and it gets highly improved performance per Hz out of this configuration. This helps gaming performance greatly while also leaving huge overclocking headroom. I'm not talking just 5GHz, I'm talking like 6GHz on affordable air cooling. It could literally compete with the i5-2500K in overclocking versus overclocking performance and power consumption comparisons, although the newer i5-3570K would win in power efficiency.

Piledriver as used in Trinity has no L3 cache, but can still already best Bulldozer by about 15% at the same frequency and core configuration while using less power to reach that frequency. With L3 cache, Vishera could easily have a more than 20-25% increase in performance per Hz of the CPU frequency while having even greater overclocking headroom.

Phenom II doesn't have a chance of keeping up with Bulldozer, let alone Piledriver, when they're configured properly for gaming. You can have your preferences, but I'd spend the literal two to five minutes that it'd take to configure these newer CPUs properly (it is easier and quicker to do than the overclocking of the CPU frequency is). Phenom II has no such core configuration optimization trick, although it does have the CPU/NB trick, to a lesser extent. However, this, although helpful, is of much lower significance than the core configuration altering for modern FX CPUs and simply isn't enough for Phenom II to keep up.
a c 152 à CPUs
a b À AMD
September 29, 2012 5:58:52 AM

Everything you've said about Piledriver is hope's and dreams right now until it's actually released. If Piledriver can manage a 10% performance increase over Phenom II / FX, then it should be equivalent to the 1st generation Core i3/i5/i7 CPUs. It would need to boost performance by about 23% - 25% to get to Sandy Bridge. To equal Ivy Bridge, it would have to be an increase of 30%. Also where are you seeing that Bulldozer beats out Phenom II. Phenom II beats out Bulldozer in all but a very few games. The benchmarks have been out for a year now that show that, that's fact not fuzzy feelings of hopes and dreams.

a b à CPUs
September 29, 2012 6:03:20 AM

rds1220 said:
Everything you've said about Piledriver is hope's and dreams right now until it's actually released. If Piledriver can manage a 10% performance increase over Phenom II / FX, then it should be equivalent to the 1st generation Core i3/i5/i7 CPUs. It would need to boost performance by about 23% - 25% to get to Sandy Bridge. To equal Ivy Bridge, it would have to be an increase of 30%. Also where are you seeing that Bulldozer beats out Phenom II. Phenom II beats out Bulldozer in all but a very few games. The benchmarks have been out for a year now that show that, that's fact not fuzzy feelings of hopes and dreams.


It's not hopes and dreams, it's simple math. We've already seen a review with many CPU benches of Trinity from Tom's and Trinity has Piledriverwithout any L3 cache. Unless 8MiB of L3 makes a less than 10% performance difference, Vishera has a more than 20-25% gain in performance per Hz. Trinity has also already proven the lower power aspect quite well too, although the very Vishera stock frequencies compared to their TDPs and then compared to similar Bulldozer models were enough to prove it.

You failed to understand what I have told you if you're going to bring up irrelevant info such as most Bulldozer reviews given their improper and amateurish methods. Only two of them that I've read actually tested CPU core configuration altering and its performance advantage and even then, they did so poorly, although the few results that they gave were great.

You can't argue with someone who knows the exact architectural and non-architectural problems and how to circumnavigate them about this unless you want a serious lesson about these CPUs. Hopes and dreams are unecessary due to the cold, hard facts that I've compiled. If you'd like a more detailed explanation of what my above suggestions do and why they work, I'd be glad to explain.
a c 283 à CPUs
a b À AMD
September 29, 2012 6:16:02 AM

luciferano said:
Only two of them that I've read actually tested CPU core configuration altering and its performance advantage and even then, they did so poorly, although the few results that they gave were great.


Trying to polish a turd by making into something that it never was... There's a good reason why most reviews didn't test in that configuration: It shouldn't need to be done.

Does it help? Absolutely it does, but intentionally crippling a chip to make it a better product doesn't make any sense (except to those who own said chips...).
a b à CPUs
September 29, 2012 6:18:26 AM

DJDeCiBeL said:
Trying to polish a turd by making into something that it never was... There's a good reason why most reviews didn't test in that configuration: It shouldn't need to be done.

Does it help? Absolutely it does, but intentionally crippling a chip to make it a better product doesn't make any sense (except to those who own said chips...).


It doesn't matter if I should need to do it, although yes, it shouldn't need to be done. I can get an FX-8120 for under $150 when it's on sale at Newegg (Microcenter doesn't have a store near me, so they're simply not an option for their in-store deals) and make it perform as well as i5s that cost more than $200. Spending literally just a few minutes more to set it up is easily worth those savings. I don't know if it's still going, but AMD also had a promotion where buyers of their CPUs were eligible for a rebate from AMD that gave money dependent on core count and the eight-core models got like $25 to $30 back and if that's still going, I could almost halve my CPU cost for comparable performance.
a c 152 à CPUs
a b À AMD
September 29, 2012 6:20:34 AM

AMD expects a 15% or less improvement over BD with the PD release, that isn't going to change much. It will be a little better in gaming but not much and again it will put it at the level of Nehalmen from Intel. The whole module garabage is still going to be slow and ineffcient. All they are doing is shoving clock speeds down people's throats hoping people will buy into it just like they did with "more corez is better." So you're personal benchmarks show the module architecture of Bulldozer is great when the industry overwhelmingly finds the exact oppositre in all but a few programs?
a b à CPUs
September 29, 2012 6:23:00 AM

Trinity 2 October 2012 with Vishera around a week later. As to performance, I am not allowed to say.
a b à CPUs
September 29, 2012 6:23:58 AM

rds1220 said:
AMD expects a 15% or less improvement over BD with the PD release, that isn't going to change much. It will be a little better in gaming but not much and again it will put it at the level of Nehalmen from Intel. The whole module garabage is still going to be slow and ineffcient. All they are doing is shoving clock speeds down people's throats hoping people will buy into it just like they did with "more corez is better." So you're personal benchmarks show the module architecture of Bulldozer is great when the industry overwhelmingly finds the exact oppositre in all but a few programs?


They're not my personal benchmarks and you failed to comprehend what I've typed. I've said that optimizing the core configuration, not some stupid special benchmarks, helps real-world lightly threaded performance greatly. Basically, you can trade off highly threaded performance for lightly threaded performance which is generally more important nowadays, especially for gaming. Most reviewers fail to realize that this can be done, so they didn't account for it.
a c 152 à CPUs
a b À AMD
September 29, 2012 6:32:26 AM

luciferano said:
They're not my personal benchmarks and you failed to comprehend what I've typed. I've said that optimizing the core configuration, not some stupid special benchmarks, helps real-world lightly threaded performance greatly. Basically, you can trade off highly threaded performance for lightly threaded performance which is generally more important nowadays, especially for gaming. Most reviewers fail to realize that this can be done, so they didn't account for it.


All they did was deliberately cripple the CPU to make it perform better. That’s as phony as the 8 GHz overclock they did with Bulldozer. It’s a joke that you have to cripple a CPU just to get it to perform better. So pretty much all you did was cherry pick a few benchmark that you agree with to try and justify your buy. Benchmarks that are a complete joke because they had to skew the results by deliberately crippling the chip to make it look better.
a b à CPUs
September 29, 2012 6:41:19 AM

Architectural evolutions are not linear, a progression doesn't automatically make a new family operate at said point merely because its the logical step up, so to quantify PD's performance in that manner is a gross nativity.

FRAPS benchmark 64 Player Operation Metro Ultra presets, MSAA 4x, AF 8x, 1920x1080 Ambient occlusions HBAO.

Radeon HD 7950

(min/avg/max)

8150: 35/69/111
6200: 32/65/105
1100T: 34/68/106
4170: 35/69/114
4100: 30/65/100
965BE: 29/59/98

2500K: 35/69/112
3570K: 35/69/112
3770K: 38/71/119

a b à CPUs
September 29, 2012 6:48:04 AM

Relevant links:
http://techreport.com/articles.x/21865
Shows that core-optimization can and does help. They used a less effective method and didn't normalize the clock frequencies between the standard core configuration and the modded version, so performance data is skewed and you need to account for that to do direct comparisons. Performance was increased by about 10-20% (usually closer to 20%) by mere thread affinity rather than full disabling or better despite the additional disadvantage of having the lower clock frequency.

http://www.tomshardware.com/reviews/a10-5800k-a8-5600k-...
http://www.tomshardware.com/reviews/ivy-bridge-benchmar...
Even Trinity that has Piledriver without L3 cache shows a very consistent ~15% performance per Hz improvement over Bulldozer that has a large 8MiB L3 cache. Athlon II versus Phenom II prove that the L3 cache can and does matter, so we can easily use quite simple math to get an accurate estimate of how much the cache in Vishera will improve performance over even Trinity's Piledriver cores.

Of course, all of the things such as branch prediction, flip-flop power efficiency, and so much more, of course Piledriver is much more power efficient than /bulldozer. It was literally designed for that and it has delivered. We already know this from Trinity.

I couldn't find a set of benchmarks showing performance scaling with the CPU/NB frequency, do here's some from Phenom II:
http://www.anandtech.com/show/3877/asrock-890fx-deluxe-...

http://hwbot.org/news/hwbot_research_lab_amd_thuban_nb_...;jsessionid=2D144BF208DD438D121DADFFBF5EDB2E
As noted here, a higher performance CPU will see greater scaling from overclocking the CPU/NB frequency, so Bulldozer in the methods described in my above posts should scale better than Phenom II and Piledriver should scale better than Bulldozer. CPU/NB performance scaling is especially high in games that seem to like high cache performance as much as decent cache capacity.
a b à CPUs
September 29, 2012 6:49:41 AM

rds1220 said:
All they did was deliberately cripple the CPU to make it perform better. That’s as phony as the 8 GHz overclock they did with Bulldozer. It’s a joke that you have to cripple a CPU just to get it to perform better. So pretty much all you did was cherry pick a few benchmark that you agree with to try and justify your buy. Benchmarks that are a complete joke because they had to skew the results by deliberately crippling the chip to make it look better.


I DIDN'T CHERRY PICK ANY BENCHMARKS. I simply found ways to optimize the CPU for lightly threaded workloads instead of highly threaded workloads. This isn't crippling, it's refocusing the CPU's performance target. Calling it crippling is like calling turning off Hyper-Threading on a CPU when you play games that don't benefit from it so that you can reduce heat and get a higher CPU frequency overclock to improve performance where it matters for you. That's not crippling, it's using the CPU in ways that benefit what YOU are doing.
a b à CPUs
September 29, 2012 6:51:02 AM

sarinaide said:
Architectural evolutions are not linear, a progression doesn't automatically make a new family operate at said point merely because its the logical step up, so to quantify PD's performance in that manner is a gross nativity.

FRAPS benchmark 64 Player Operation Metro Ultra presets, MSAA 4x, AF 8x, 1920x1080 Ambient occlusions HBAO.

Radeon HD 7950

(min/avg/max)

8150: 35/69/111
6200: 32/65/105
1100T: 34/68/106
4170: 35/69/114
4100: 30/65/100
965BE: 29/59/98

2500K: 35/69/112
3570K: 35/69/112
3770K: 38/71/119


How linear it is is irrelevant. Tom's own benches say 15% from Bulldozer with L3 cache to Piledriver without L3 cache. In fact, if we go by that AMD was probably referring to only desktop models by saying increasing performance in each generation by 10-15%, we can already say that this idea was abandoned in face of much higher gains per generation. Piledriver brings a more than 20-25% gain and Steamroller after it will provide a similar gain over Pledirver with Excavator intended to have an even larger gain over Steamroller.
a c 283 à CPUs
a b À AMD
September 29, 2012 6:54:08 AM

luciferano said:
This isn't crippling, it's refocusing the CPU's performance target. Calling it crippling is like calling turning off Hyper-Threading on a CPU when you play games that don't benefit from it so that you can reduce heat and get a higher CPU frequency overclock to improve performance where it matters for you.


Then AMD should market it to gamers as a quad core (with your adjustments, of course). They would sell millions.

That's what I have a problem with. They market it as an 8 core (which it is), but it's a poorly performing 8 core CPU. It's a much better performing quad core...
a b à CPUs
September 29, 2012 6:56:13 AM

DJDeCiBeL said:
Then AMD should market it to gamers as a quad core (with your adjustments, of course). They would sell millions.

That's what I have a problem with. They market it as an 8 core (which it is), but it's a poorly performing 8 core CPU. It's a much better performing quad core...


You're correct. AMD should have made versions that have the second core of each module disabled to get a 4C/4M configuration. They would have probably sold greatly.
a c 152 à CPUs
a b À AMD
September 29, 2012 6:57:35 AM

DJDeCiBeL said:
Then AMD should market it to gamers as a quad core (with your adjustments, of course). They would sell millions.

That's what I have a problem with. Th ey market it as an 8 core (which it is), but it's a poorly performing 8 core CPU. It's a much better performing quad core...


Exactly. You can call it what you want but I think it's stuipid to have to disable half the CPU just to get better perfomance.
a b à CPUs
September 29, 2012 7:01:11 AM

It is true that in like 5% of benches run a 1100T beat a 8150 and in some games the 1100T and 980BE edge a 8150 by a frame or two, but the other 95% show that the FX chips comprehensively beat Phenom II. If you also factor that thubans more than aptly ran comparible with up to i7 950/960's it is rather moot trying to say PD will be at this or that level, sufficed to say that PD without disclosing will sit somewhere between nahelem and ivybridge, as to where....wait and see.

As to cherry picks, it is normally those so detesting of AMD that only focus on benches that show a gap, which are so few I can count them on 3 fingers, the rest show the fx 81XX parts between the i5 and i7 cuncurrent parts, the Fx 6XXX matching lower end i5 parts and the fX 4XXX running somewhere between i3 and low end i5's.

about 10% gaming and productivity benches really show up the difficiencies in module architecture, but the rest the penalties are minimal.

If I was to give BD on a whole a rating from 0-100 I would say about 86%, Intel about 91%
a b à CPUs
September 29, 2012 7:01:11 AM

rds1220 said:
Exactly. You, luci can call it what you want. I call it crippling but whatever helps you sleep at night.


It's crippling when I recommend doing it, but it's just fine when DJDeCiBeL says that the second core of each module should be disabled for gaming? Do you also call my Hyper-Threading example crippling?
a c 152 à CPUs
a b À AMD
September 29, 2012 7:08:43 AM

rds1220 said:
Exactly. You can call it what you want but I think it's stuipid to have to disable half the CPU just to get better perfomance.


Go back and see my edit. Like DJ said it should have been advertised as a quad core instead of the 8 core crap. If it's advertised as a eight core processor it should run like an 8 core processor.
a b à CPUs
September 29, 2012 7:10:09 AM

rds1220 said:
Go back and see my edit. Like DJ said it should have been advertised as a quad core instead of the 8 core crap. If it's advertised as a eight core processor it should run like an 8 core processor.


It does run like an eight-core CPU because it is one. FX-81xx CPUs can trade blows with the LGA 1155 i7s in highly threaded performance. There are eight cores (two per module, four modules). Our point is that there should have been some quad-core models with the second core of each module disabled instead of the second two full modules disabled so that we have models that are optimized for lightly threaded performance over highly threaded performance.
a b à CPUs
September 29, 2012 7:17:32 AM

It does run like a 8 core processor, similarly it runs as a hex and quadcore as well. The architecture with deeper pipelines and a single FPU per module along with shared L2 first increases branch mispredicts, second the baseclocks are to low to keep the pipes fed so you end up with loss of IPC, third latency due to shared L2 resources. You can mitigate the loss of IPC by increasing clock speed, at 4.2ghz its performance is rather impressive and the voltage and temps are still rather low.

Considering modular architecture was a step into the unknown considering AMD milked K2 since 2003, its not impressive yet but it most certainly is not crap as its put. Simply put Intel cannot pull away despite its deficiencies in all but a few benchmarks show any difference.

The successes of BD:

lower stock idle and load power consumption.
higher clockability vs K architecture at the cost of power consumption.
IMC performance 55% faster than Phenom II and 25% less latency.
Multi threading performance and CMT is around 1.75% of a dual core vs intels 1.15% of a dual core with HT.
a b à CPUs
September 29, 2012 10:14:31 AM

So what i want too know is, will Piledriver finaly give me a modern CPU worthy of replacing my aged 1100t OCd to 4ghz ?
a b à CPUs
September 29, 2012 1:37:10 PM

The problem was, nobody realized how much of a good thing the 1090/1100T where, mostly due to it being looked at as the last kick of a dying horse. They represent the absolute evolution end of K architecture running from Athlon64 to its final form. When BD came along, there was never going to be any reason for users of a 1090 or 1100 to make the move, but there is certainly a case if you have anything less.

While no FX part really makes a viable office machine for word processors or excel spreadsheets, along with pascal accounting and lite office apps, it does make feasible claims in CAD and engineering based environments.

As to whether you need to change, I don't think so on the fact that the Thuban is strong in that one, but there is always a cause to upgrade.
September 29, 2012 1:50:50 PM

Like sarinaide, I have a 8150. I have it OC to 4.5ghz and it is a great rig. Unlike sarinaide, I don't have access to the PileDriver. Listen to what he is saying carefully. BTW I had a 1100Thuban OC to 4 ghz and it was an outstanding chip also. Despite its "warts" overall the 8150 is better. See my list of rigs, I also have 2 i5 2500 k rigs so i'm hardly an AMD fanboy.
The 8150 is on sale at Newegg for $169.99 until 10/4.
a c 152 à CPUs
a b À AMD
September 29, 2012 5:00:24 PM

luciferano said:
It does run like an eight-core CPU because it is one. FX-81xx CPUs can trade blows with the LGA 1155 i7s in highly threaded performance. There are eight cores (two per module, four modules). Our point is that there should have been some quad-core models with the second core of each module disabled instead of the second two full modules disabled so that we have models that are optimized for lightly threaded performance over highly threaded performance.


Sure it is if you're dopey enough to buy into AMD's module scam. Really the 8100's are quad core with a sucky attempt at hyper threading. The same thing with the 4100's and 6100's. The 4100's are dual core and the 6100's are three core. As for the second part I wouldn't call that a victory. In most cases the I7 will beat out a Bulldozer even in highly threaded programs, when it actually does beat out the I7 it's JUST BARELY.

Also I don't think you understood DJ, he's saying that they should have advertisted the 8100 for what it is, a quad core CPU with an attempt at hyperthreading. He wasn't saying they should have had a seperate quad core with "cores" disabled.
a c 283 à CPUs
a b À AMD
September 29, 2012 5:17:43 PM

rds1220 said:
In most cases the I7 will beat out a Bulldozer even in highly threaded programs, when it actually does beat out the I7 it's JUST BARELY.


Yep, and it's not even close (Civ V bench notwithstanding): 3770K vs. 8150 http://www.anandtech.com/bench/Product/434?vs=551

As was mentioned in another thread (rds knows what I'm talking about), of course you can't expect a sub $200 CPU to perform like a $300+ CPU, but facts are facts. The i7's dominate a 8150.

Quote:
Also I don't think you understood DJ, he's saying that they should have advertisted the 8100 for what it is, a quad core CPU with an attempt at hyperthreading, not that they should have had a seperate quad core with "cores" disabled.


Correct, although if AMD were smart, they would actually have a version with cores disabled.
a b à CPUs
September 29, 2012 7:14:07 PM

rds1220 said:
Sure it is if you're dopey enough to buy into AMD's module scam. Really the 8100's are quad core with a sucky attempt at hyper threading. The same thing with the 4100's and 6100's. The 4100's are dual core and the 6100's are three core. As for the second part I wouldn't call that a victory. In most cases the I7 will beat out a Bulldozer even in highly threaded programs, when it actually does beat out the I7 it's JUST BARELY.

Also I don't think you understood DJ, he's saying that they should have advertisted the 8100 for what it is, a quad core CPU with an attempt at hyperthreading. He wasn't saying they should have had a seperate quad core with "cores" disabled.


It's not Hyper-Threading nor is it anything like it and it's not supposed to be anything like it. Hyper-Threading is a way to more efficiently use a core by allowing a second thread to work with unused resources of a core. A module is a way to merge two distinct cores in order to use die area more efficiently. It doesnt' scale perfectly because AMD failed to realize the front end needs of the two cores. Steamroller will fix this and performance scaling within each module will be 100% instead of 75-80%.

The FX-41xx CPUs are quad-core, the 61xx/62xx are six cores, and the 81xx are eight cores. There are literally two cores in each module. AMD failed to anticipate the front end needs (especially the x86 decoders, four per module is simply not enough for two cores). Steamroller will fix this by giving each core of each module a dedicated four x86 decoders (there will be eight per module instead of four per module, but instead of sharing them between each core, each core will have its own dedicated set of four x86 decoders).

Calling a module a sucky attempt at Hyper-Threading is like calling a semi truck a sucky attempt at making a car. They're not the same nor are they trying to be the same. They both have the same goal of getting something to another place just as both Hyper-Threading and the modular approach are both intended to improve performance efficiency in a given aspect, but they do so differently and with different characteristics.
a c 152 à CPUs
a b À AMD
September 29, 2012 9:07:09 PM

luciferano said:
It's not Hyper-Threading nor is it anything like it and it's not supposed to be anything like it. Hyper-Threading is a way to more efficiently use a core by allowing a second thread to work with unused resources of a core. A module is a way to merge two distinct cores together to use die area more efficiently. It doesnt' scale perfectly because AMD failed to realize the front end needs of the two cores. Steamroller will fix this and performance scaling within each module will be 100% instead of 75-80%.


More warm fuzzy feelings with hopes and dreams about the future. Steam Roller is still a ways off and even if it does fix the problems it will still only perform at the level of Sandy Bridges still behind Intel. Everything you said at the end is all speculation and compleatly irrelevant until Steam Roller comes out. Like I said all it is, is warm fuzzy feeling of hopes and dreams. I know what hyper threading is and how it works. So first you said it is like hyper threading
Quote:
It's crippling when I recommend doing it, but it's just fine when DJDeCiBeL says that the second core of each module should be disabled for gaming? Do you also call my Hyper-Threading example crippling
now it's nothing like hyper threading. So which is it, I don't think you really know r really have an understanding of Bulldozers micro architecture. Your first comment would imply that turning on and off cores is like turning on or off hyper threading on an I7. That would me the 8 core processor is really a quad core with hyperthreading.

luciferano said:
The FX-41xx CPUs are quad-core, the 61xx/62xx are six cores, and the 81xx are eight cores. There are literally two cores in each module. AMD failed to anticipate the front end needs (especially the x86 decoders, four per module is simply not enough for two cores). Steamroller will fix this by giving each core of each module a dedicated four x86 decoders (there will be eight per module instead of four per module, but instead of sharing them between each core, each core will have its own dedicated set of four x86 decoders).


You can keep trying to sell me on what a core is but unlike you I'm not foolish enough to fall for the scam. All it is, is a bunch of half cores mashed together to form a full core. So the so called 8 core is four half cores smashed together. Last time I checked 1/2+1/2+1/2+1/2 equals four not eight. Making the so called 8 core processor really four cores. Same story with the so called four core and six core CPU. So there are not two cores in each module it's two half cores in each module. You keep talking about steam roller and your warm fuzzy feeling for it but it's still a ways off. By the time we hit that point, if they even continue with desktop CPU's with non integrated graphics Intel will be on Haswell getting ready for the next architecture. If the performance continues on track steam roller will perform on the level of Sandy Bridgesd still far behind Intel.

luciferano said:
Calling a module a sucky attempt at Hyper-Threading is like calling a semi truck a sucky attempt at making a car. They're not the same nor are they trying to be the same. They both have the same goal of getting something to another place just as both Hyper-Threading and the modular approach are both intended to improve performance efficiency in a given aspect, but they do so differently and with different characteristics.


You can call it what you want but that's pretty much what it is. It's four cores with four virtual cores just like an I7 with hyperthreading. Difference is the I7 is faster and more effcient than the Bulldozer. It's not any different really SMT really isn't that much different than hyperthreading.

http://scalibq.wordpress.com/2012/02/14/the-myth-of-cmt...
a b à CPUs
September 30, 2012 1:35:35 AM

rds1220 said:
More warm fuzzy feelings with hopes and dreams about the future. Steam Roller is still a ways off and even if it does fix the problems it will still only perform at the level of Sandy Bridges still behind Intel. Everything you said at the end is all speculation and compleatly irrelevant until Steam Roller comes out. Like I said all it is, is warm fuzzy feeling of hopes and dreams. I know what hyper threading is and how it works. So first you said it is like hyper threading
Quote:
It's crippling when I recommend doing it, but it's just fine when DJDeCiBeL says that the second core of each module should be disabled for gaming? Do you also call my Hyper-Threading example crippling
now it's nothing like hyper threading. So which is it, I don't think you really know r really have an understanding of Bulldozers micro architecture. Your first comment would imply that turning on and off cores is like turning on or off hyper threading on an I7. That would me the 8 core processor is really a quad core with hyperthreading.



You can keep trying to sell me on what a core is but unlike you I'm not foolish enough to fall for the scam. All it is, is a bunch of half cores mashed together to form a full core. So the so called 8 core is four half cores smashed together. Last time I checked 1/2+1/2+1/2+1/2 equals four not eight. Making the so called 8 core processor really four cores. Same story with the so called four core and six core CPU. So there are not two cores in each module it's two half cores in each module. You keep talking about steam roller and your warm fuzzy feeling for it but it's still a ways off. By the time we hit that point, if they even continue with desktop CPU's with non integrated graphics Intel will be on Haswell getting ready for the next architecture. If the performance continues on track steam roller will perform on the level of Sandy Bridgesd still far behind Intel.



You can call it what you want but that's pretty much what it is. It's four cores with four virtual cores just like an I7 with hyperthreading. Difference is the I7 is faster and more effcient than the Bulldozer. It's not any different really SMT really isn't that much different than hyperthreading.

http://scalibq.wordpress.com/2012/02/14/the-myth-of-cmt...


I never said that it is like Hyper-Threading.

These aren't warm and fuzzy feelings that you keep going on about. I never said anything about Steamroller only making the one improvement. It will make more.

There are no virtual threads in a Bulldozer CPU. There are eight cores on every Bulldozer CPU die. Eight physical cores.

There is no scam and I'm not falling for anything. This isn't a bunch of half-cores thrown together. There are eight physical cores and they've been merged into modules to make them use less die area and be able to share data better. Things like the x86 decoders, the cache, and more is able to be made more compact when you merge two cores because some components of all of this are redundant and wasteful.

My Hyper-Threading example applies not because Hyper-Threading is similar, but because it too focuses on highly threaded performance and disabling it allows the CPU to be focused on lightly threaded performance. Hyper-Threading increases heat generation significantly, but if a game can't use more than four threads, it doesn't improve performance by much, so it's wasting thermal headroom that could instead be used by increasing the clock frequency, thus increasing lightly threaded performance.

Disabling the second core of each module too allows for a greater focus on lightly threaded performance. However, the modular approach is nothing like Hyper-Threading. The problem with Bulldozer's thread scaling is quite simple, really. There are only four x86 decoders per module. This is not enough. For example, Phenom II had three of them per core and Every Intel arch since Core 2 has four of them per core. Obviously, having two cores share four of them is almost bound to causes issues.

Reading a link doesn't make you a CPU engineering expert. Furthermore, your link was obviously written by a journalist, not a CPU expert. Much of it is plain wrong and/or blatant ignorance. AMD's modular design has almost nothing to do with why they use more transistors than Intel, why Intel has more performance per core, and much more. AMD's greater transistor needs are caused by AMD's die masks poorly implementing the micro-architecture (AMD's design methods used in the Bulldozer die were rushed and need about 20% more transistors than they would have if AMD had done it right like they did with Phenom II after the same design crap of the original Phenoms), AMD's excessively large caches (Intel's LGA 1155 quad-core CPUs have a total of about 10MiB of cache, AMD's Bulldozer die has a total of about 17MiB of cache), and there's also the process technology differences (you can't just compare very different CPUs that are built on entirely different processes as if they're on the same process and get any useful results). That's just some basics of how messed up that link is.

The writer thinks that Hyper-Threading splits the ALUs of a core between two threads like 1.5 per thread. That's like saying that three people can be split into two sets of 1.5 people.

Even better is how these aren't half-cores in Bulldozer. They are literally full cores. If you take two cores and combine their x86 decoders into one block of x86 decoders, their L2 cache into one block of cache, FPUs into a single FPU, and more, they're still two cores, they're just tightly linked and share resources. It's quite literally nothing like Hyper-Threading where there is just one core that has a front end that is altered to handle two threads. That one core's execution resources can still only execute one thread at a time, they can simply switch quickly to execute instructions from a second thread on execution resources that the first thread isn't using (such as when it stalls to wait for a DRAM access). Bulldozer shares it's front end between cores to more efficiently use die area than it would as two separate cores. It also lets a lot of these resources be more efficiently utilized, but AMD skimped on some (such as the x86 decoders), so people who are ignorant of any of this get the wrong idea. Yes, AMD screwed up. HOWEVER, it was not in any way what you seem to have thought it was and it's not because the modular approach is bad nor because it is anything like SMT (neither of which are true about it, just rumors made by people who don't know what they're talking about). AMD's biggest mistake was simply not "feeding the cores" well enough.

Also, if you got your idea of how Hyper-Threading works from your link, then you do not know how it works.

The modular approach is like the concept of shared cache, just implemented in other aspects of the CPU core hardware. I don't see people complaining about how Core 2 Quads had two dual-core dies that also had shared cache within each dual-core die, but when AMD does something similar although more effective, it's now an issue. AMD's concepts of Bulldozer are an extreme improvement over current CPUs, their implementation is what was flawed. AMD should have had better branch prediction to account for the deeper pipe lines (this is improved in Piledriver), AMD should have had more x86 decoders (this is improved in Steamroller), AMD should have done a lot of other things that are fixed in Piledriver, Steamroller, and excavator.

However, that doesn't make the BS in your link true (about both AMD and Intel) and it doesn't make your claims correct either. Heck, your link actually shows that these aren't virtual threads in Bulldozer, one of the few correct things about it.
a c 186 à CPUs
a b À AMD
September 30, 2012 1:38:55 AM

luciferano said:

Disabling the second core of each module too allows for a greater focus on lightly threaded performance.

Then, what's the point?

Why buy an 8 core CPU and not use 8 Cores? :heink: 
a b à CPUs
September 30, 2012 1:41:53 AM

amuffin said:
Then, what's the point?

Why buy an 8 core CPU and not use 8 Cores? :heink: 


AMD didn't make them available as they should be with the second core already disabled. It's the same reason for why we overclock, to get more out of the CPU. That it's core configuration being changed instead of frequency makes no difference in the purpose of it.
a c 283 à CPUs
a b À AMD
September 30, 2012 2:22:49 AM

luciferano said:
AMD didn't make them available as they should be with the second core already disabled. It's the same reason for why we overclock, to get more out of the CPU. That it's core configuration being changed instead of frequency makes no difference in the purpose of it.


You should be a politician, you know that? :lol: 

Spin...
a c 152 à CPUs
a b À AMD
September 30, 2012 5:45:05 AM

Yep. Must of worked for AMD's marketing team. I'm not even going to bother responding to your post lucif it's just a waste of time.There's no point in arguing with a fanboy you're going to keep buying AMD crap no matter how bad it is and then try to spin it to make it sound postive. Maybe you can get a job with Mitt Rommney you both are great flip flopping spin artist and full of crap.
a b à CPUs
September 30, 2012 5:54:24 AM

You two are beyond ridiculous. You mock and accuse me of working for AMD PR after I insult the company for their stupidity? I'f be fired for what I just said if I did work for them. I told you how things are with the chips and if you two are too ignorant of the technology to understand that, then that's your problem, not mine.
a c 283 à CPUs
a b À AMD
September 30, 2012 5:57:10 AM

luciferano said:
You two are beyond ridiculous. You mock and accuse me of working for AMD PR after I insult the company for their stupidity? I'f be fired for what I just said if I did work for them. I told you how things are with the chips and if you two are too ignorant of the technology to understand that, then that's your problem, not mine.


For what it's worth, my words were to be taken at face value. I'm serious when I say that you spin things very well. Finding the positives in a universally accepted negative.

That's not a bad thing, mind you, you just do it well.
a b à CPUs
September 30, 2012 6:00:25 AM

DJDeCiBeL said:
For what it's worth, my words were to be taken at face value. I'm serious when I say that you spin things very well. Finding the positives in a universally accepted negative.

That's not a bad thing, mind you, you just do it well.


I explained the truth of the matter when others failed to do so. Believe what you want, but what I've said is true and accurate. I'd suck as a politician because I'd quickly get fed up with the other politicians BS and probably do something very bad to them.
a c 283 à CPUs
a b À AMD
September 30, 2012 6:05:16 AM

luciferano said:
I'm not spinning anything off. I explained the truth of the matter when others failed to do so. Believe what you want, but what I've said is true and accurate.


I never said what you've said isn't true (or at the very least possible), but you're having to defend something that shouldn't need to be defended and honestly isn't easy to defend. I'm just impressed by how well you do it.
a b à CPUs
September 30, 2012 6:15:31 AM

DJDeCiBeL said:
I never said what you've said isn't true (or at the very least possible), but you're having to defend something that shouldn't need to be defended and honestly isn't easy to defend. I'm just impressed by how well you do it.


I just don't like to see people spreading lies. AMD has made mistakes (many of which were stupid screw-ups that shouldn't have happened), but they should be blamed for what mistakes that they made, not crap made up by people who don't know what they're talking about, especially very ignorant of CPU engineering journalists such as the writer of the article in rds1220's link. Thanks; I try to be thorough.
a c 152 à CPUs
a b À AMD
September 30, 2012 6:40:31 AM

luciferano said:
I just don't like to see people spreading lies. AMD has made mistakes (many of which were stupid screw-ups that shouldn't have happened), but they should be blamed for what mistakes that they made, not crap made up by people who don't know what they're talking about, especially very ignorant of CPU engineering journalists such as the writer of the article in rds1220's link. Thanks; I try to be thorough.


The only one spreading lies is you with your fanboy spin. I see nothing wrong with that article and the only ones who have cried about it are the same people who fanatically defend AMD like you Sarinade and Noob. Could it be fan boy touchiness, sounds like it to me.
a b à CPUs
September 30, 2012 6:44:37 AM

rds1220 said:
The only one spreading lies is you with your fanboy spin. I see nothing wrong with that article and the only ones who have cried about it are the same people who fanatically defend AMD like you Sarinade and Noob. Could it be fan boy touchiness, sounds like it to me.


The article is full of ignorance of both Intel and AMD technology. Almost everything about it is wrong except some of the specifications and even worse, it's main picture is seven years old. So, it's outdated and inaccurate anyway in a huge amount of ways.
a b à CPUs
September 30, 2012 6:46:04 AM

I'll go into further detail about the link. Quoted from the end of it:

Quote:
But what does CMT bring, effectively? Nothing. Their chips are much larger than the competition’s, or even their own previous generation. And since the Xeon is so much better with single-threaded performance, it can stay ahead in heavy multithreaded scenario’s, despite the fact that SMT does not scale as well as CMT or SMP. But the real advantage that SMT brings is that it is a very efficient solution: it takes up very little die-space. Intel could do the same as AMD does, and put two dies in a single package. But that would result in a chip with 12 cores, running 24 threads, and it would absolutely devour AMD’s CMT in terms of performance.

So I’m not sure where AMD thinks that CMT is ‘more efficient’, since they need a much larger chip, which also consumes more power, to get the same performance as a Xeon, which is not even a high-end model. The Opteron 6276 tested by Anandtech is the top of the line. The Xeon X5650 on the other hand is a midrange model clocked at 2.66 GHz. The top model of that series is the X5690, clocked at 3.46 GHz. Which shows another advantage of smaller chips: better clockspeed scaling.

So, let’s not pretend that CMT is a valid technology, comparable to SMT. Let’s just treat it as what it is: a hollow marketing term. I don’t take CMT seriously, or people who try to use the term in a serious context, for that matter.


There are so many things wrong with this that it's ridiculous and I'm amazed at to how this person could write such blatant ignorance. The over-generalizations alone are enough to discredit this writer completely.

This writer confuses CMT's impact with AMD's process problems (technically Global Foundry's problems) and die design implementation problems. This writer has little understanding of how Bulldozer modules and Hyper-Threading work. The writer doesn't even know what clock speed scaling is. The writer comes to a completely wrong conclusion just because the writer doesn't know what he's talking about. My above post explains just how little CMT has to do with Bulldozer's transistor needs (if AMD had made those chips with the same 10h micro-architecture of Phenom II and its harvested versions such as Athlon II and Sempron, their new FX line would use more transistors if the dies were designed with similar methodology), so I won't restate it here. The single and lightly performance problems of AMD are also not related to the modular concept. AMD's architecture could have had some improvements such as those made in Piledriver and Steamroller, but that doesn't change that the base performance issues are not at all related to this "CMT".

Even with AMD's CPUs as they are, our views on them would be greatly different if AMD had simply set them up better. AMD should have gotten proper Windows scheduling support with MS done before the launch. AMD should have made their Turbo implementation better. Due to how the cores in each module having to share an insufficient set of x86 decoders, AMD should have done something like Zero Core to disable the second core of each module when running single/lightly threaded work. Unlike frequency boosts, this would increase single and lightly threaded performance while decreasing power consumption during those tasks instead of increasing it. Of course, increasing frequency too is also still possible, but that's beside the point.

Point is that AMD should be held accountable for their real mistakes, not imagined ones.
September 30, 2012 7:04:19 AM

I like the idea of buying a four wheel drive car and having the option of switching to a two wheel drive to go faster , saying that if there is a option to use in any product why not use it ? I mean having a four wheel drive car with no option for two wheel drive is cutting corners .....

I really don't think AMD made much of a mistake if there is a option , i read a forum with someone stating , they have one machine running virtual box times 8 , i have also read people disabling cores to get better performance when running games , plus other trick you can do within windows ...

Before i5 etc and AMD bulldozer , i am sure there are some that would implement many procedures to make their CPU go faster ..

I mean whats the point of having a roast dinner with gravy and putting tomato sauce on it if it already has gravy , i guess the answer would be because i can , or i like it like that .....
a b à CPUs
September 30, 2012 7:40:58 AM

Obviously there is reason to wait if the OP is on a AM3+ setup or supporting socket 942. It will still offer a very viable option at a good price point without needing to buy a new platform.

When BD first released it was a dog to get working properly, a number of patches, hot fixes, updates and the performance is a lot more symmetrical, it is still a viable option depending on user circumstances.
September 30, 2012 4:04:59 PM

luciferano said:
I'll go into further detail about the link. Quoted from the end of it:

Quote:
But what does CMT bring, effectively? Nothing. Their chips are much larger than the competition’s, or even their own previous generation. And since the Xeon is so much better with single-threaded performance, it can stay ahead in heavy multithreaded scenario’s, despite the fact that SMT does not scale as well as CMT or SMP. But the real advantage that SMT brings is that it is a very efficient solution: it takes up very little die-space. Intel could do the same as AMD does, and put two dies in a single package. But that would result in a chip with 12 cores, running 24 threads, and it would absolutely devour AMD’s CMT in terms of performance.

So I’m not sure where AMD thinks that CMT is ‘more efficient’, since they need a much larger chip, which also consumes more power, to get the same performance as a Xeon, which is not even a high-end model. The Opteron 6276 tested by Anandtech is the top of the line. The Xeon X5650 on the other hand is a midrange model clocked at 2.66 GHz. The top model of that series is the X5690, clocked at 3.46 GHz. Which shows another advantage of smaller chips: better clockspeed scaling.

So, let’s not pretend that CMT is a valid technology, comparable to SMT. Let’s just treat it as what it is: a hollow marketing term. I don’t take CMT seriously, or people who try to use the term in a serious context, for that matter.


There are so many things wrong with this that it's ridiculous and I'm amazed at to how this person could write such blatant ignorance. The over-generalizations alone are enough to discredit this writer completely.

This writer confuses CMT's impact with AMD's process problems (technically Global Foundry's problems) and die design implementation problems. This writer has little understanding of how Bulldozer modules and Hyper-Threading work. The writer doesn't even know what clock speed scaling is. The writer comes to a completely wrong conclusion just because the writer doesn't know what he's talking about. My above post explains just how little CMT has to do with Bulldozer's transistor needs (if AMD had made those chips with the same 10h micro-architecture of Phenom II and its harvested versions such as Athlon II and Sempron, their new FX line would use more transistors if the dies were designed with similar methodology), so I won't restate it here. The single and lightly performance problems of AMD are also not related to the modular concept. AMD's architecture could have had some improvements such as those made in Piledriver and Steamroller, but that doesn't change that the base performance issues are not at all related to this "CMT".

Even with AMD's CPUs as they are, our views on them would be greatly different if AMD had simply set them up better. AMD should have gotten proper Windows scheduling support with MS done before the launch. AMD should have made their Turbo implementation better. Due to how the cores in each module having to share an insufficient set of x86 decoders, AMD should have done something like Zero Core to disable the second core of each module when running single/lightly threaded work. Unlike frequency boosts, this would increase single and lightly threaded performance while decreasing power consumption during those tasks instead of increasing it. Of course, increasing frequency too is also still possible, but that's beside the point.

Point is that AMD should be held accountable for their real mistakes, not imagined ones.


Why don't you comment on my article then?
Oh right, I see, you are just using ad hominems, rather than proper technical arguments. You wouldn't last long in a proper technical debate.
The TL;DR you have is "it's GF's fault, not AMD's". Which is rather silly, since the die size is a big factor in manufacturing issues, and that is exactly what my article is about. But I guess that demonstrates your superior understanding of SMT/CMT...?
September 30, 2012 4:15:48 PM

luciferano said:
not crap made up by people who don't know what they're talking about, especially very ignorant of CPU engineering journalists such as the writer of the article in rds1220's link.


You mean me? I'm not a journalist, do your research.
I also don't think 'ignorant' describes me very accurately. Again, do your research.
      • 1 / 2
      • 2
      • Newest
!