How much is AMD behind.

[alyarb]

Bulldozer's idle power is pretty close to Sandy or Ivy, so your system would have to be loaded for the difference to appreciably come into play.

If your systems are fully loaded 24/7 then yeah, the power saved by the intel machine will pay for the difference of the more expensive platform within 11,000 hours. That is at 14 cents per kw-h, assuming a 100w difference in load power (according to anand's ivy testing) and looking at pricing on newegg, $320 for ivy and $170 for BD.

 

[inf64]

Good post alyarb. I didn't go into details but ~10K hours seems about right. That's roughly 1 year and 3 months of constant load on all cores on both systems.

 

[taltamir]

Newegg is way overpriced and the chips you are looking at are on the expensive end.

http://www.microcenter.com/single_product_results.phtml?product_id=0359809
http://www.microcenter.com/single_product_results.phtml?product_id=0388579
http://www.microcenter.com/single_product_results.phtml?product_id=0354589
http://www.microcenter.com/single_product_results.phtml?product_id=0388577

even if you do go by newegg, intel has cheaper CPUs and AMD more expensive ones:
http://www.newegg.com/Product/Produc...ame=LGA%201155
http://www.newegg.com/Product/Produc...ocket%20AM3%2b

Comparing socket 1155 to AM3+ prices.

For example, here is the most expensive zambezi core:
http://www.newegg.com/Product/Produc...82E16819103960
at 200$.
Now lets see what kind of Ivy Bridge you can get for 200$ on newegg:
http://www.newegg.com/Product/Produc...82E16819116506

So how do they compare?
http://www.anandtech.com/show/5871/intel-core-i5-3470-review-hd-2500-graphics-tested/5
Here anandtech compares the 205$ i5-3470 to the 200$ FX-8150
the i5 in question http://www.newegg.com/Product/Product.aspx?Item=N82E16819115234

unfortunately they don't compare power consumption between the two.

 

[alyarb]

Microcenter consistently sells CPUs below market to get you into the store. Good to cite them in a hot deals thread, but not here where you have international readers or people who otherwise do not have access to below-market pricing.

Superbiz has a 3770 for $307, so you could say that cost would be recoupable in 10k hours rather than 11k? Either way, those 10,000 load-hours will take quite a long time to cover because most systems are only loaded for brief periods.

I am not advocating intel or AMD on any performance grounds, but I am saying the higher cost of a more efficient platform can be recouped before the platform is decommissioned. either way you've got two dead-end sockets so you are in for the long haul. may as well go with pcie 3.0, thunderbolt, etc

 

[nyker96]

I think the sad and interesting part about all this is that, amd no longer cares that it's way behind in cpu area, just puts all the energy into gpu side.

 

[taltamir]

Microcenter consistently sells CPUs below market to get you into the store. Good to cite them in a hot deals thread, but not here where you have international readers or people who otherwise do not have access to below-market pricing.

Superbiz has a 3770 for $307, so you could say that cost would be recoupable in 10k hours rather than 11k? Either way, those 10,000 load-hours will take quite a long time to cover because most systems are only loaded for brief periods.

1. Newegg is still overpriced, they are riding their fame and good name for extra short term profits.

2. I am discussing things as an american in this english speaking forum.
Interntional buyers don't have microcenter, they also have very different pricing due to marketing execs of various companies being crazy and then there are VAT and what have you. So yea, it doesn't apply to EVERYONE, but it does apply to tens of millions of people.

3. Even if you completely remove my argument on the overpricedness of newegg, you are ignoring my other points. The comparison of a lower end bulldozer part to a higher end intel part.
How does the 200$ bulldozer part hold up to the 200$ intel ivy bridge part?
the 100$ bulldozer part to the 100$ intel? The notion that intel costs 150$ more is unfounded.

 

[AtenRa]

1. Newegg is still overpriced, they are riding their fame and good name for extra short term profits.

2. I am discussing things as an american in this english speaking forum.
Interntional buyers don't have microcenter, they also have very different pricing due to marketing execs of various companies being crazy and then there are VAT and what have you. So yea, it doesn't apply to EVERYONE, but it does apply to tens of millions of people.

3. Even if you completely remove my argument on the overpricedness of newegg, you are ignoring my other points. The comparison of a lower end bulldozer part to a higher end intel part.
How does the 200$ bulldozer part hold up to the 200$ intel ivy bridge part?
the 100$ bulldozer part to the 100$ intel? The notion that intel costs 150$ more is unfounded.

Since we are using Microcenter,

AMD FX8120 at $149,99
http://www.microcenter.com/single_product_results.phtml?product_id=0375765

Intel Core i5 3570K at $189,99
http://www.microcenter.com/single_product_results.phtml?product_id=0388577

But at MC we can also have an AMD FX8120 + ASUS M5A97 AM3+ ATX for $194,98. That is $5 more than the 3750K and you also get the Motherboard.

http://www.microcenter.com/specials/promotions/AMDbundlePROMO.html

 

[inf64]

I think the sad and interesting part about all this is that, amd no longer cares that it's way behind in cpu area, just puts all the energy into gpu side.

Again what is "way behind"? We actually demonstrated that in case of top of the line 8T vs 8T ,IB on desktop is 22% faster. When has 22% faster become way ahead? Also for 22% faster performance users are going to pay 1.75x more money. We debunked the "lower wattage savings" since it takes 1+ year at constant full load on all cores for power bill savings to justify the price difference.

And yes,AMD is focusing on heterogeneous computing since this is the future. In 4-5 years GPU will be 1st class citizen in CPU cores and will be able to do a lot of stuff it can't do today. And speedups won't be 10-100% ,it will be an order of magnitude higher. Note that not all workloads will be able to utilize/offload to GPU cores. Some will be still single thread or very hard to extract parallelism from. Here intel will still have considerable advantage but as time goes by, these kind of worklaods will become more irrelevant.

3. Even if you completely remove my argument on the overpricedness of newegg, you are ignoring my other points. The comparison of a lower end bulldozer part to a higher end intel part.
How does the 200$ bulldozer part hold up to the 200$ intel ivy bridge part?
the 100$ bulldozer part to the 100$ intel? The notion that intel costs 150$ more is unfounded.

That's easy answer. AMD's top priced part is FX8150 and in application workloads(stock vs stock) it's slightly faster than 3570K. Intel's 200$ part on newegg (using neweggg for FX8150 too) is 3450 . 3450 is 13% slower than 3570K,so ~15% slower than FX8150 in application workloads. Still worse perf./dollar.

 

[taltamir]

Since we are using Microcenter,

AMD FX8120 at $149,99
http://www.microcenter.com/single_product_results.phtml?product_id=0375765

Intel i5-3450 at 149.99
http://www.microcenter.com/single_product_results.phtml?product_id=0388579

Intel Core i5 3570K at $189,99
http://www.microcenter.com/single_product_results.phtml?product_id=0388577

AMD FX8150 at 199.99
http://www.microcenter.com/single_product_results.phtml?product_id=0375764&ekw=amd_fx-8150

But at MC we can also have an AMD FX8120 + ASUS M5A97 AM3+ ATX for $194,98. That is $5 more than the 3750K and you also get the Motherboard.

http://www.microcenter.com/specials/promotions/AMDbundlePROMO.html

That is a good deal. but they have similar ones for intel. IIRC I got my intel i5-3570K with a mobo combo that discounted the mobo by 70$

 

[Charles Kozierok]

Again what is "way behind"?

Sufficiently behind that people don't have a compelling reason to choose AMD platforms over Intel, and AMD publicly announces that it is throwing in the towel when it comes to the performance race?

 

[inf64]

22% on desktop is peanuts gap. 100% is sufficient gap. First one AMD can close up pretty fast(and you will see it) with Vishera. If it was 100% gap then AMD would probably never catch intel.
AMD will have a product soon that in applications will be able to measure up to 3770K in real world applications(not sysmark and not CPU limited games mind you). For some this may not be a big feat but looking how good SB/IB generation is on both desktop and server it will be quite an accomplishment.

edit: also as was previously stated you pay much more than premium price for 22% more performance. 1.75x more. It should be much less expensive (~30%) for it to be better deal for people who actually use their CPUs for productive work.

 

[taltamir]

AMD will have a product soon that in applications will be able to measure up to 3770K in real world applications

We have had countless empty promises from AMD, I will believe it when I see it.

Also, intel significantly hobbles their current chips due to lack of competition. Not only is OC headroom on their chips massive while still maintaining good power consumption, but it could have been better if they didn't just change to an inferior but cheaper heat transfer mechanism on the integrated heat spreader.

 

[inf64]

As always intel is "hindering" their chips... I don't buy that at all. Why should they do that? The problem is that die area is smaller and heat dissipation becomes a problem. This will be a great problem on smaller nodes. CPUs will generate similar amount of heat while being ever smaller. You can do some things to combat this but to blame poor thermal grease for all the problems related to heat/clocking is not logical. Sure you can get better results with better heat conducting materials but it would be reflected on price also. Nothing is free in today's world.

 

[ShintaiDK]

22% on desktop is peanuts gap. 100% is sufficient gap. First one AMD can close up pretty fast(and you will see it) with Vishera. If it was 100% gap then AMD would probably never catch intel.
AMD will have a product soon that in applications will be able to measure up to 3770K in real world applications(not sysmark and not CPU limited games mind you). For some this may not be a big feat but looking how good SB/IB generation is on both desktop and server it will be quite an accomplishment.

edit: also as was previously stated you pay much more than premium price for 22% more performance. 1.75x more. It should be much less expensive (~30%) for it to be better deal for people who actually use their CPUs for productive work.

Where did your 22% come from btw? Also the Derpdozer still uses ~2x the power as well. Not to mention without 8 threads, it quickly goes extremely wrong for it.

Vishera is just another dissaster waiting to happen. Im not sure why you put any faith into it. AMD themselves dont, its the last of the line.

Serverwise nobody buys AMD because it really suck even more there. Xeons just utterly trash it. APUs is all they got left.

So not sure what kind of code we need to run to satisfy you.

 

[taltamir]

As always intel is "hindering" their chips... I don't buy that at all. Why should they do that?

So that that you will buy a next gen chip in a year or two that is as fast as their current chips could have been if not hobbled.

The problem is that die area is smaller and heat dissipation becomes a problem. This will be a great problem on smaller nodes. CPUs will generate similar amount of heat while being ever smaller. You can do some things to combat this but to blame poor thermal grease for all the problems related to heat/clocking is not logical. Sure you can get better results with better heat conducting materials but it would be reflected on price also. Nothing is free in today's world.

Intel switched their integrated heat spreader to a cheaper less effective model starting with ivy bridge. This is not a trade-off relating to die size (which is a REAL issue), this is a fact as reported by the tech news media such as anandtech. The statement they did so in order to hobble their chip's over-clockability is my opinion based on that fact.
The fact that smaller die are harder to spread the heat of is irrelevant to the claim that intel has switched to a cheaper and less effective method concurrently.

If there is falsehood there then it lies in the tech news media which has falsified the facts, but I do not see a reason to doubt this.

And regardless of that one thing, intel has been hobbling their chips in a plethora of ways for many years.
From ever decreasing multipliers, ever decreasing die size (in GPU density AND die size increase to maximize benefit and as a result you get such impressive improvements per generation), ever more features like HT being disabled on cheaper processors...

 

[inf64]

Where did your 22% come from btw? Also the Derpdozer still uses ~2x the power as well. Not to mention without 8 threads, it quickly goes extremely wrong for it.

Hardware.fr and THG ST and MT runtimes. In hardware.fr chart, 3770K is 22% faster than FX8150 (183.7 vs 150.7). In THG ST suite , 2600K is 27%(491s/673s=0.729) faster than FX8150 and 11.4%(879s/991s=0.886) faster in MT suite.
Power draw is much higher on FX though and this is a problem. But performance gap is not as huge as some want to make it look like.

 

[taltamir]

Power draw is much higher on FX though and this is a problem. But performance gap is not as huge as some want to make it look like.

Fair enough.

Although I am still amazed not a single person would comment about my take on the underlying causes of the performance gap and AMD's inability to bridge it.
http://forums.anandtech.com/showpost.php?p=33665185&postcount=46

To paraphrase it:
AMD (mostly) has feature parity (turbo mode, cool and quiet, independent power planes, power gating, etc) with Intel. But every time AMD copies an intel performance improving or power saving feature they implement a "budget" designed version of it that is inferior but cheaper and easier to design.

This is a problem because:
1. Intel and AMD have both spent years refining their design. If AMD switches their inherently inferior design to intel like design they are years behind in refining it.
2. AMD's investment into their current design is a sunk cost. So if intel spent 500m on a feature and AMD spent 100m, AMD will now have to spend 500m as well and end up spending a total of 600m, more then intel, just to match up intel... and they have to do that for every single component. (that is assuming AMD has equally robust corporate culture and management which can execute at the same cost as intel, if not it will cost them even more!)

There is a single plus side for AMD. They can compare the effectiveness of individual performance/power improving features in current intel CPUs to their own and go after the ones that make the most sense in terms of cost of redesign, benefit of redesign, and speed of redesign.

 

[Piano Man]

I say 2 years behind. BD was a monster flop compared to SB/IVB. I was hoping for another K7, but it didn't happen. I really do wish they would get their stuff together. Ivy Bridge was a pretty "meh" chip for me, and its not really gonna hurt Intel at all because AMD still has nothing to compete with it.

I fear that AMD is going to abandon the high end, and focus on their APU's for the mainstream, which means intel will be able to crank up costs on their enthusiast chips.

 

[taltamir]

I say 2 years behind. BD was a monster flop compared to SB/IVB. I was hoping for another K7, but it didn't happen. I really do wish they would get their stuff together. Ivy Bridge was a pretty "meh" chip for me, and its not really gonna hurt Intel at all because AMD still has nothing to compete with it.

What? Ivy bridge is a huge a success!
It provides a huge decrease in power consumption and manufacturing cost while slightly improving performance (normally you need to lose performance to improve costs and power)

These benefits are hugely tangible in the mobile & server market.

 

[ShintaiDK]

Hardware.fr and THG ST and MT runtimes. In hardware.fr chart, 3770K is 22% faster than FX8150 (183.7 vs 150.7). In THG ST suite , 2600K is 27%(491s/673s=0.729) faster than FX8150 and 11.4%(879s/991s=0.886) faster in MT suite.
Power draw is much higher on FX though and this is a problem. But performance gap is not as huge as some want to make it look like.

You do know that in Toms MT. The 2500K performs just a tiny bit under? And a 3570K would be faster?

And in the Hardware.fr, the 3570K is just 1% slower than the FX8150 in MT?

Thats basicly the best case scenario for AMD, and its not going well. The FX8150 actually uses close to 3 times of the 3570K in power.

If we turn to gaming the Hardware.fr basicly shows that the FX8150 needs to perform 70% more to reach IB. Intel could easily be twice as fast here if they increased the TDP a tad, tho still quite lower than Bulldozers TDP. And thats with Intel hosting the PCIe hub and an iGPU.

AMD got nothing, never will have again. Its game over on that area. Last chance is APUs.

 

[TuxDave]

So that that you will buy a next gen chip in a year or two that is as fast as their current chips could have been if not hobbled.

This is so hilariously wrong. There is no crippling of chips on purpose. (but there is crippling of chips by accident. )

 

[Piano Man]

What? Ivy bridge is a huge a success!
It provides a huge decrease in power consumption and manufacturing cost while slightly improving performance (normally you need to lose performance to improve costs and power)

These benefits are hugely tangible in the mobile & server market.

Yea, for that market, I guess so. I guess I"m talking about the Anandtech-I-like-to-OC-my-CPU-to-the-limit" market

 

[inf64]

@Shintai

Gamin,seriously? Who cares about gaming ,I hope people buy their PCs to be productive,not to waste time playing games. And in their selection the games they used are ridiculously CPU bound. In most cases you will be GPU bound with higher end gfx card anyway. Sure faster gaming CPU will give you more fps in some titles but overall you should not base your buying decision on games. As mention before, desktop workloads are not that well threaded. This is actually amplifying how well FX performs. In non-perfect parallel desktop world it achieves 88% of top of the line IB chip that is 1.75x more expensive and which will pay out it's price premium thru power bill in the period of 1+ year of constant full load (ridiculous scenario huh?).
As for THG suite ,of course MT runtime is not linearly decreasing with core count increase since not all workloads are scaling that well with more cores.
3570K is still sower in Hardwre.fr chart and it's more than 24% slower than 3770K. This is what SMT brings you. It's not a miracle feature you know...

 

[ShintaiDK]

@Shintai

Gamin,seriously? Who cares about gaming ,I hope people buy their PCs to be productive,not to waste time playing games. And in their selection the games they used are ridiculously CPU bound. In most cases you will be GPU bound with higher end gfx card anyway. Sure faster gaming CPU will give you more fps in some titles but overall you should not base your buying decision on games. As mention before, desktop workloads are not that well threaded. This is actually amplifying how well FX performs. In non-perfect parallel desktop world it achieves 88% of top of the line IB chip that is 1.75x more expensive and which will pay out it's price premium thru power bill in the period of 1+ year of constant full load (ridiculous scenario huh?).
As for THG suite ,of course MT runtime is not linearly decreasing with core count increase since not all workloads are scaling that well with more cores.
3570K is still sower in Hardwre.fr chart and it's more than 24% slower than 3770K. This is what SMT brings you. It's not a miracle feature you know...

So your entire case is now trapped into a corner where only using 8 threads is valid?

And if you pay 0.35$ per Kw/h. How long do you need a full load then before it becomes another issue there? Not to mention noise? Or heat removal in the datacenters?

What happens when I need double the worktime time and reduced server avaliability because the singlethreaded performance of Derpdozer is so horrible? You know Microsoft for example loves .net precompiles and such with their server applications. Singlethreaded all the way.

Its all fun that you love and fight for AMD. Its just a lost cause when the patient is on life support.

 

[inf64]

So your entire case is now trapped into a corner where only using 8 threads is valid?

And if you pay 0.35$ per Kw/h. How long do you need a full load then before it becomes another issue there? Not to mention noise? Or heat removal in the datacenters?

What happens when I need double the worktime time and reduced server avaliability because the singlethreaded performance of Derpdozer is so horrible? You know Microsoft for example loves .net precompiles and such with their server applications. Singlethreaded all the way.

Its all fun that you love and fight for AMD. Its just a lost cause when the patient is on life support.

Desktop shintai. The results are for desktop. Workloads are not scaling perfectly to multiple threads. So it's not the best case for Derpdozer as you call it.

I didn't discuss server space,but it has it's place there also. It's not rosy situation though and they know it. For desktop the gap is what those 3 links show. You can't turn it around,those are the facts. And yes,it consumes more power but as mentioned it's not going to be much of a problem unless you do full load most of your time on PC. Heat can be an issue but AMD's CPUs are actually not heating up that much,max temp is ~62C ,unlike SB/IB product range.