ExtremeTech: AMD Bulldozer FX pricing revealed: a lot cheaper than Sandy Bridge

[inf64]

I didn't think it was gonna be amazing, but I thought IPC near Nehalem was realistic. After all, all they'd only be bridging one generational gap. If they did that, the FX-8150 would be a bit faster than the Core i7-990X overall. But alas, now I see that's asking too much.

What you were thinking was not that far fetched after all. But IPC is a variable category,you may find Zambezi scoring awesome in one app and then scoring poor in some other. Some tradeoffs were being made with Bulldozer,let's hope overall we have a healthy boost over Thuban that at least puts it in shooting range of Westmere in some workloads.

 

[superccs]

Can they just release some BD chips already??!?!?? Cmon AMD seriously.

 

[LOL_Wut_Axel]

Ya, but rumors continued to hint that top parts would turbo to 4.0-4.2ghz. Then and there I became extremely suspicious regarding IPC.

If you have 90% of IPC of Nehalem, then a 4.2ghz 8 core FX = 3.78ghz Core i5 760. You know how fast that is?

If a 4.2ghz 8 core FX = 3.78ghz Lynnfield/Nehalem, then it's also roughly equal to an 8 core 2500k @ 3.3ghz because SB has a 15% increase in IPC over Nehalem.

So you'd end up with a 4.2ghz FX 8150 = 8 core 2500k @ 3.3ghz roughly. That's insanely fast.

I'm not quite sure if you're being serious. Given scaling, and given Nehalem IPC, an FX-8150 at 4GHz would equal roughly a Core i7-990X at the same clock speed. Remember that all of Intel's >$280 CPUs have HyperThreading.

 

[996GT2]

It seems AMD decided to focus on cores this generation.

 

[AtenRa]

BD doesn't come with a "free" ~$50 graphics card. Add in the price of a discrete card and the price seems right in line compared to llano.

Core i3 2120 Dual core + HT 3.3GHz @ $139.99
Llano A8-3850 quad core 2.9GHz @ $139.99

And now you want me to believe that,

Core i5 2300 Quad core 2.8GHz @ 184.99 cheapest SB Core i5 in Newegg
Phenom II X6 1100T 3.3GHz @ $189.99
BD FX-6100 6-core 3.3GHz @ $155.00

BD FX-8120 8-core 3.1GHz Turbo 4GHz @ $185.00
So it takes not only double the cores but higher frequency in order to be competitive vs cheapest Quad core SB (No HT) ??

 

[RussianSensation]

I'm not quite sure if you're being serious. Given scaling, and given Nehalem IPC, an FX-8150 at 4GHz would equal roughly a Core i7-990X at the same clock speed. Remember that all of Intel's >$280 CPUs have HyperThreading.

990X doesn't Turbo to 4.0-4.2ghz, so what is this comparison at the same clock speed? What you are saying is that you expected FX to have an IPC ~ Lynnfield/Nehalem. So in essence a 4.2ghz 8 core i5 760. You are telling me a 990X can beat an 8 core Lynnfield @ 4.2ghz, no chance!

I am not sure if you are being serious thinking AMD had a $300 CPU that would equal or beat a $999 Intel CPU?

There is no way AMD would be so worried about raising clocks beyond 4.0ghz if they really had an 8 core i5 750 @ 3.3ghz equivalent CPU in the wings considering they were selling X4 980 for X6 1075T for $169. An 8 core i5 750 @ 3.3ghz would be more than 2x faster than the X4 980.

Even the lowly FX-4100 series clocked at 4.0ghz with Nehalem-like IPC would easily = 2500k. You really expect me to believe AMD would go from being uncompetitive in the last 5 years to having a $300 CPU that is 2x faster than 2500k?

 

[LOL_Wut_Axel]

LOL. I've seen this image several times, and each time that passes it becomes truer.

 

[Vesku]

Ya, but rumors continued to hint that top parts would turbo to 4.0-4.2ghz. Then and there I became extremely suspicious regarding IPC.

If you have 90% of IPC of Nehalem, then a 4.2ghz 8 core FX = 3.78ghz Core i5 760. You know how fast that is?

If a 4.2ghz 8 core FX = 3.78ghz Lynnfield/Nehalem, then it's also roughly equal to an 8 core 2500k @ 3.3ghz because SB has a 15% increase in IPC over Nehalem.

So you'd end up with a 4.2ghz FX 8150 = 8 core 2500k @ 3.3ghz roughly. That's insanely fast.

That's not all 8 core turbo though, 8 core turbo would depends on workload versus TDP. Let's say it's 80% Nehalem (doesn't that put expectations close to just an 8 core Phenom II?), 3GHz 8 core i5 7xx 3.3GHz assuming some turbo room for the workload. Then how much headroom would we expect considering the delays. 10% OC headroom 3.2GHz 8 core i5 7xx, 20% OC headroom 3.4-3.5GHz 8 core i5 7xx?

I'm guessing BD is going to be workload dependent, it seems just from the official info that they planned for future workloads and compiled server workloads. Guessing it will be important to find the BD performance of all the applications a desktop enthusiast cares about before diving in. Even if it's a scrambled omelette of a product I'm starting to get excited about seeing how much I can squeeze out of it on my AM3+ system.

 

[LOL_Wut_Axel]

990X doesn't Turbo to 4.0-4.2ghz, so what is this comparison at the same clock speed? What you are saying is that you expected FX to have an IPC ~ Lynnfield/Nehalem. So in essence a 4.2ghz 8 core i5 760. You are telling me a 990X can beat an 8 core Lynnfield @ 4.2ghz, no chance!

You forgot the part where I said at the same clock speed. You're inflating the results. I don't care about Turbo. Stop putting words into my mouth.

I am not sure if you are being serious thinking AMD had a $300 CPU that would equal or beat a $999 Intel CPU?

Considering that the $1000 one was just a $600 one with an "Extreme Edition" moniker and an unlocked multiplier, yes. Especially since Gulftown is about to be phased out, and it would mean that it'd have higher multi-threaded performance and lower single-threaded performance. It'd be a worthy tradeoff for most.

There is no way AMD would be so worried about raising clocks beyond 4.0ghz if they really had an 8 core i5 750 @ 3.3ghz equivalent CPU in the wings considering they were selling X4 980 for X6 1075T for $169. An 8 core i5 750 @ 3.3ghz would be more than 2x faster than the X4 980.


To compete in single-threaded as well. And you're forgetting that scaling decreases as you add more cores, which is probably why Intel decided to go the route of HyperThreading.

^^

 

[RussianSensation]

Core i5 2300 Quad core 2.8GHz @ 184.99 cheapest SB Core i5 in Newegg
Phenom II X6 1100T 3.3GHz @ $189.99
BD FX-6100 6-core 3.3GHz @ $155.00

BD FX-8120 8-core 3.1GHz Turbo 4GHz @ $185.00
So it takes not only double the cores but higher frequency in order to be competitive vs cheapest Quad core SB (No HT) ??

1. I would say that 1100T is overpriced at $190 to begin with. So AMD should have lowered the price of this CPU anyway. FX-6100 with IPC of Phenom II at $155 sounds reasonable.

2. AMD may offer you more cores/more performance in multi-threaded apps and way less performance in 1-4 threaded apps. Even now 1100T is slower than 2500k despite 2 more cores. Would you buy an 1100T over an i5-2400/2500k? Even before SB launched, most people here chose i5/i7s over X6s. AMD's market share was not really moving, so the answer is clearly no. So AMD may need 8 cores to convince more people to buy it/cement its market positioning as the "multi-threaded" CPU.

 

[AtenRa]

What you are saying is that you expected FX to have an IPC ~ Lynnfield/Nehalem. So in essence a 4.2ghz 8 core i5 760. You are telling me a 990X can beat an 8 core Lynnfield @ 4.2ghz, no chance!

I am not sure if you are being serious thinking AMD had a $300 CPU that would equal or beat a $999 Intel CPU?

There is no way AMD would be so worried about raising clocks beyond 4.0ghz if they really had an 8 core i5 750 @ 3.3ghz equivalent CPU in the wings considering they were selling X4 980 for X6 1075T for $169. An 8 core i5 750 @ 3.3ghz would be more than 2x faster than the X4 980.

Even the lowly FX-4100 series clocked at 4.0ghz with Nehalem-like IPC would easily = 2500k. You really expect me to believe AMD would go from being uncompetitive in the last 5 years to having a $300 CPU that is 2x faster than 2500k?

8 Core BD will not turbo to 4.2GHz all cores but half, so your equation is wrong.

With Nehalem IPC an 8-core 3.6GHz BD will be same or marginally faster than Core i7 980-990X in Multithreaded apps (dont forget Core i7 has HT +15-20% more performance avg per core).

 

[sm625]

Core i3 2120 Dual core + HT 3.3GHz @ $139.99
Llano A8-3850 quad core 2.9GHz @ $139.99

And now you want me to believe that,

Core i5 2300 Quad core 2.8GHz @ 184.99 cheapest SB Core i5 in Newegg
Phenom II X6 1100T 3.3GHz @ $189.99
BD FX-6100 6-core 3.3GHz @ $155.00

BD FX-8120 8-core 3.1GHz Turbo 4GHz @ $185.00
So it takes not only double the cores but higher frequency in order to be competitive vs cheapest Quad core SB (No HT) ??

Why not? If 6 core BD really does price at $155, then that means it is probably 50% faster than an i3-2100 in multithreaded apps, and 30-50% slower in single threaded apps. It would be equivalent to 6 slightly overclocked bobcat cores.

 

[RussianSensation]

^^

I just said in my previous post, EVEN if I discounted Turbo by AMD, an FX-8100 with IPC of Nehalem = 8 core i5 750 @ 3.3ghz. You expect me to believe that AMD would release such a processor at only $300? You know it would be up to 2x faster than 2500k right and barely slower in 1-4 threaded apps? Not sure if serious. But I can't discount Turbo, can I? FX processor will have Turbo and it will go upwards of 4.0ghz. You know this, which implies you expected an 8 core 4.0ghz equivalent i5 750, for $300.

Also, like I said, if they had Nehalem like IPC, then they wouldn't have been selling X4/X6 processors for the last 9 months. They would have released 3.2-3.3ghz "Nehalem" like BD chips in 4-6-8 core formats and then refreshed the entire line-up in the fall with higher clocked variants. They didn't do this because BD cannot compete at low clock speeds, which means its IPC =!= Nehalem.

8 Core BD will not turbo to 4.2GHz all cores but half, so your equation is wrong.

Read above. Highlighted in blue for you.

With Nehalem IPC an 8-core 3.6GHz BD will be same or marginally faster than Core i7 980-990X in Multithreaded apps (dont forget Core i7 has HT +15-20% more performance avg per core).

5 years of no show from AMD, 1.5 years delay with Phenom I, Phenom II barely matching 65nm quads from 2007, and then out of the blue, they are going to release a $300 processor that beats a $1000 flagship from 6 months ago. And Dirk was fired for such an amazing CPU?

Also HT on avg only adds 4-5%. So your 15-20% performance increase is on the high-end, not average.

 

[Majic 7]

Here are some prices from BLT. Usually higher than most.http://www.shopblt.com/cgi-bin/s.cgi?order_id=!ORDERID!&s_lastpos=0&t_all=1&s_max=25&s_all=AMD+FX+FD 188, 222, 266$

 

[LOL_Wut_Axel]

I just said in my previous post, EVEN if I discounted Turbo by AMD, an FX-8100 with IPC of Nehalem = 8 core i5 750 @ 3.3ghz. You expect me to believe that AMD would release such a processor? You know it would be up to 2x faster than 2500k right? Not sure if serious.

I'm not sure where you're pulling those performance numbers from. Because of decreasing scaling, given the same IPC, roughly: 8C/8T=6C/12T.

Is the Core i7-990X 2x faster than the Core i5-2500K? Not sure if serious.

 

[Idontcare]

LMFAO that's hilarious

 

[AtenRa]

You really expect me to believe AMD would go from being uncompetitive in the last 5 years to having a $300 CPU that is 2x faster than 2500k?

I would say that it is possible that an 8-Core 3.6GHz BD will be 50% faster than Core i5 2500K in multithreaded apps.
BD has 50% more transistors, more Cache and more executable units (8 cores), if they can’t produce 50% more performance than 4 core SB (with 10-15% added frequency) then they better stop designing CPUs.

So AMD should have lowered the price of this CPU anyway. FX-6100 with IPC of Phenom II at $155 sounds reasonable.

You cant be serious, BD is not Llano

 

[996GT2]

You cant be serious, BD is not Llano

Well since we have no solid benchmarks to go on, it could very well be that slow. Nobody knows and AMD sure doesn't want us to know.

 

[LOL_Wut_Axel]

Well since we have no solid benchmarks to go on, it could very well be that slow. Nobody knows and AMD sure doesn't want us to know.

Looking at the price I wouldn't be surprised if it was slower than Llano, LOL.

 

[inf64]

Why does it need to have IPC of Nehalem? How about 85% of Nehalem's IPC(same core count of course),on average, and 4.2Ghz Turbo on 4 cores?
Look here. Take as an example i7 950 and i7 960. Use in between overall score for hypothetical 4C Zambezi and apply 0.85x - you get 154pts. 50% more cores on desktop,with intel,brings you 16% more performance in that chart.Hence 100% cores (4C Zambezi => 8C Zambezi) brings you 32% more performance: 154x1.32=204pts in the chart. Now this is for 3.1Ghz/4Ghz. 3.6Ghz/4.2Ghz should therefore score roughly 10% more (accounting for high turbo on 8120) : 204*1.1=224pts. That's a hair slower than 990x and a bit faster than 2600K.
Remember this is possible performance of Zambezi that assumes it is 17% slower than Nehalem at the same clock.

 

[AtenRa]

Well since we have no solid benchmarks to go on, it could very well be that slow. Nobody knows and AMD sure doesn't want us to know.

If a 6-core BD has lower-same performance as 6-Core Thuban, then i will personally give my fishing boat to Mr, Mike Butler for his retirement, after the Bulldozer Release fiasco.

 

[Vesku]

Well since we have no solid benchmarks to go on, it could very well be that slow. Nobody knows and AMD sure doesn't want us to know.

Well, Cray is upgrading 12 core MagnyCours chips with 16 core Interlagos. I'm not in charge of large deployment but I'd imagine they'd want at least a ~25-30% performance bump for the expense and effort of an upgrade. Which would mean the output of a BD module would at minimum need to be equal to 2 previous cores. If they couldn't offer that in the same TDP not sure they would have been able to sell them. A bit worrying it's taken several delays to get there though.

 

[RussianSensation]

I'm not sure where you're pulling those performance numbers from. Because of decreasing scaling, given the same IPC, roughly: 8C/8T=6C/12T.

Where in the world did you come up with 8C/8T = 6C/12T? Based on what data? Scaling doesn't change with more threads if the program actually uses 8 threads. An 8 core i5 760 would be 90-100% faster in a program that uses 8 threads instead of 4 vs. a i5 760. If you argue that few programs use 8 threads, then I agree, which is why BD being slower in 1-4 threads is critical.

Is the Core i7-990X 2x faster than the Core i5-2500K?

i7-990X is a not a SB architecture and it only has 6 cores. So how would it be 2x faster than a 4 core SB? Your comparison makes no sense.

I would say that it is possible that an 8-Core 3.6GHz BD will be 50% faster than Core i5 2500K in multithreaded apps.

You are contradicting yourself. An 8 core Nehalem would be way faster than just 50% faster than a 4 core 2500k.

Under your scenario (~50% faster), it would have worse IPC than Nehalem/Lynnfield (since if it was = 3.6ghz 8-core Nehalem, it would be almost ~100% faster). Remember 8 divided by 4 is not 50% greater, but 100% greater mathematically:

3.6ghz Nehalem = 3.13ghz SB (since SB is 15% faster per clock).
Then 3.6ghz 8 core Nehalem would be be nearly nearly 90-95% faster than a 4 core 2500k @ 3.3ghz in an 8 threaded app. Even if we discount perfect scaling, it would still be 70-80% faster.

I never said it's not possible to have 50% faster performance in multi-threaded apps since it's obvious AMD is focusing on multi-threaded performance. What I did say, however, was that there is no way FX-8100 = 8 core i5 760 @ 3.3ghz based on simple math.

Well, Cray is upgrading 12 core MagnyCours chips with 16 core Interlagos. I'm not in charge of large deployment but I'd imagine they'd want at least a ~25-30% performance bump for the expense and effort of an upgrade. Which would mean the output of a BD module would at minimum need to be equal to 2 previous cores. If they couldn't offer that in the same TDP not sure they would have been able to sell them. A bit worrying it's taken several delays to get there though.

16 / 12 = 33% more.

So in that case 1 BD module = Phenom dual core. That implies pretty much identical IPC to Phenom II.

 

[busydude]

AT just posted this news on the main page.. and the comments are not encouraging:

Sort of. Without performance numbers (which we still know almost nothing about, though I expect AMD would have told us by now if the news was good), it's hard to say how one of AMD's Bulldozer cores compares to a more traditional dual-core CPU.

Unfortunately, all signs point to Zambezi being less than stellar; I suspect that clock for clock, a single BD core will be slower than current K10.5 stuff, but you'll get more cores. It will also be interesting to see how Turbo Core plays out; if it's as cautious as some of the other chips, the eight core chips will only run at the base (or base + 1) frequency for anything more than dual-core workloads.

 

[996GT2]

AT just posted this news on the main page.. and the comments are not encouraging:

Not good indeed, especially since the comments are written by AT's own editors.