AMD set to slash FX CPU pricing on September 1

Page 5 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
Status
Not open for further replies.

MiddleOfTheRoad

Golden Member
Aug 6, 2014
1,123
5
0
For low-threaded programs, FX is inferior hardware to Sandy/Ivy Bridge. And now with highly threaded programs, FX is inferior to Haswell.

I actually find it depressing that it took Intel multiple generations at 22 nm to pull ahead of AMD chips made on the ancient 32 nm process.

I'm very curious about this new FX 8320E chip that has been leaked -- I wonder how substantial the improvements were made to drop power
consumption while maintaining clock speed. I wonder if these are actually 9000 series FX chips that can't quite hit 4.7 or 5.0 Ghz?
 
Last edited:

Cerb

Elite Member
Aug 26, 2000
17,484
33
86
When Intel actually starts selling Ivy Bridges or Haswells that do 8 threads at the prices that AMD charges for their 8 thread chips -- that actually might mean something.

But since they don't, it's Apples to Oranges.
It's dollars to dollars.

Phoronix's benchmarks look better for the FX because their benchmarks are worse. It's exactly how the P4 looked decent back in the day, according to the benchmarks of the day (or worse, the old Celerons). In Windows, there are better benchmarks to choose, now, but not under Linux. Where Linux handles CMT well, it also handles HT well, too. DC is not representative of most uses, and basic synthetic benchmarks aren't worth much of anything..AMD has to price the CPUs low because for what most of us do most of the time, they are simply inferior. They are not like Intel's CPUs at all, and hve a few niches where they excel, but not many, and most are not relevant to most users, even power users.

The biggest problem they have with that is that there are points beyond which no cost reduction will yield sales, and they've been stuck in a rut there with servers, notebooks, and to a lesser degree, non-APU desktops. They need 6-8T CPUs like Kaveri, now, and something very different in the near future, with a focus on lower (and better hidden) latencies and higher IPC/thread.
 

mrmt

Diamond Member
Aug 18, 2012
3,974
0
76
Hmm. Well, whatever their reasons for the price cut, for the sake of AMD and the buying public, let us hope they have something interesting to take over the flagship spot from the aging FX CPUs.

They won't. If they had something remotely interesting for the FX line they would have launched it on the server market first, not on the desktop market.
 

jhu

Lifer
Oct 10, 1999
11,918
9
81
When Intel actually starts selling Ivy Bridges or Haswells that do 8 threads at the prices that AMD charges for their 8 thread chips -- that actually might mean something.

But since they don't, it's Apples to Oranges.

Now you're shifting the goalposts. But since we're doing that, electricity usage and costs have to be factored in then. Currently on Newegg, the FX 8350 is $180 and the i5 4670K is $235. I'm picking the i5 because, based on my testing with multithreaded programs, one Haswell core is nearly as fast as 2 threads on an FX module.

FX 8350 has a rated TDP of 125W and the 4670K has a rated TDP of 84W. In practice actual power draw will be higher due to inefficiencies in the power supply. My FX 8350 machine idles at 70W and pulls down 255W full loaded (this is at stock speeds, measured with a Kill-o-watt). That's a difference of 185W, which is at least 48% higher than expected. Assuming the theoretical 4670K machine has a similar increase in power draw it would have a fully loaded power draw of 124W. The difference is then about 40W. Now let's look at electricity costs around the country. Average is $0.0984/kWh (High: Hawaii at $0.34/kWh, Low: Louisiana at $0.0690/kWh). I live in California, whose average is $0.135/kWh.

Here's summary of data:
Price:
FX 8350: $180
4670K: $235
Difference: $55

TDP:
FX 8350: 125W (255W actual)
4670K: 84W (124W estimated)
Difference: 41W (131W estimated)

Electricity cost: $0.069/kWh - $0.34/kWh, average $0.0984/kWh

Since you're using World Community Grid, I'll assume the computer is 100% loaded at all the times finding cures for cancer and such.

Difference in energy usage between the processors in 1 year:
Rated TDP - 359.160 kWh
Estimated system TDP - 1147.56 kWh


Difference total costs by:
Rated TDP - $24.78 - $122.11, average $35.34
Estimated system TDP - $79.18 - $390.17, average $112.92


The rated TDP greatly underestimates the true power draw at the outlet.
 
Last edited:

MiddleOfTheRoad

Golden Member
Aug 6, 2014
1,123
5
0
Now you're shifting the goalposts. But since we're doing that, electricity usage and costs have to be factored in then.
Here's summary of data:

Price:
FX 8350: $180
4670K: $235
Difference: $55

TDP:
FX 8350: 125W (255W actual)
4670K: 84W (124W estimated)
Difference: 41W (131W estimated)

The rated TDP greatly underestimates the true power draw at the outlet.

I actually feel sad that you invested the time in writing that post.

Because the power consumption argument is pretty much a total wash for anyone that buys a K or Black Edition. When a user buys an unlocked processor from EITHER manufacturer, power consumption is entirely up to the user's preference. They can set their chips to whatever level they want. Considering every FX is unlocked -- the user can change power consumption based on their mood. Yet another reason why these price drops are awesome -- how many unlocked 4 or 6 core Intels can you buy for $120? Exactly.

Nobody I know runs on the Grid at stock settings. Everybody I know is undervolting by a substantial margin. So all your figures are wildly exaggerated -- we'd all go broke if we actually ran that much wattage every day.
 
Last edited:

monstercameron

Diamond Member
Feb 12, 2013
3,818
1
0
I actually feel sad that you invested the time in writing that post.

Because the power consumption argument is pretty much a total wash for anyone that buys a K or Black Edition. When a user buys an unlocked processor from EITHER manufacturer, power consumption is entirely up to the user's preference. They can set their chips to whatever level they want. Considering every FX is unlocked -- the user can change power consumption based on their mood. Yet another reason why these price drops are awesome -- how many unlocked 4 or 6 core Intels can you buy for $120? Exactly.

Nobody I know runs on the Grid at stock settings. Everybody I know is undervolting by a substantial margin. So all your figures are wildly exaggerated -- we'd all go broke if we actually ran that much wattage every day.
those seem to be estimated upper bounds. Average usage most definitely will vary.
 

jhu

Lifer
Oct 10, 1999
11,918
9
81
I actually find it depressing that it took Intel multiple generations at 22 nm to pull ahead of AMD chips made on the ancient 32 nm process.

I see your ninja edit. Man, I feel as if the ghost of galego has returned...

Prior to goalposts moving, the discussion was based on processor performance. Intel has been ahead of AMD since Core2 (see my link), and almost overnight, AMD's $1000 consumer chips disappeared. If you put a bunch of low IPC computing units on a chip, at some point it will be faster than the higher IPC chips on highly multithreaded workloads. That doesn't change the fact that highly-multithreaded workloads are still not the norm. Single-thread performance still matters. AMD has finally made Piledriver IPC about equivalent to Phenom II. Meanwhile, Haswell's single-thread performance is 65% higher than Piledriver based on my tests.
 

PPB

Golden Member
Jul 5, 2013
1,118
168
106
Also bear in mind that FXs are mych more conservative regarding binnings. Intel cpus tend to be binned pretty aggressively, even for being mobile rejects.
 

Lepton87

Platinum Member
Jul 28, 2009
2,544
9
81
Now you're shifting the goalposts. But since we're doing that, electricity usage and costs have to be factored in then. Currently on Newegg, the FX 8350 is $180 and the i5 4670K is $235. I'm picking the i5 because, based on my testing with multithreaded programs, one Haswell core is nearly as fast as 2 threads on an FX module.

FX 8350 has a rated TDP of 125W and the 4670K has a rated TDP of 84W. In practice actual power draw will be higher due to inefficiencies in the power supply. My FX 8350 machine idles at 70W and pulls down 255W full loaded (this is at stock speeds, measured with a Kill-o-watt). That's a difference of 185W, which is at least 48% higher than expected. Assuming the theoretical 4670K machine has a similar increase in power draw it would have a fully loaded power draw of 124W. The difference is then about 40W. Now let's look at electricity costs around the country. Average is $0.0984/kWh (High: Hawaii at $0.34/kWh, Low: Louisiana at $0.0690/kWh). I live in California, whose average is $0.135/kWh.
My computer idles at 112W but I've got all SATA ports populated.
2x256GB MX100 RAID 0
128GB Sandforce 2
CD-ROM
1TB Seagate 7200RPM
2TB WD-Caviar Green
2x Seagate 3TB RAID 0
1TB Seagate External USB 3.0
I don't have a slot left to plug in my 64GB SSD Sandforce 1
On top of that I have lots of fans. All of those things add up in power draw, no to mention 1200W PSU which isn't very efficient at such small loads like 100W.
 

jhu

Lifer
Oct 10, 1999
11,918
9
81
My computer idles at 112W but I've got all SATA ports populated.
2x256GB MX100 RAID 0
128GB Sandforce 2
CD-ROM
1TB Seagate 7200RPM
2TB WD-Caviar Green
2x Seagate 3TB RAID 0
1TB Seagate External USB 3.0
I don't have a slot left to plug in my 64GB SSD Sandforce 1
On top of that I have lots of fans. All of those things add up in power draw, no to mention 1200W PSU which isn't very efficient at such small loads like 100W.

When I was testing on my machine, it was CPU+motherboard+memory+USB stick running Ubuntu 14.04. It idled at 70W, which at least is lower than the 1090T it replaced (idled at 80W). When I added 2x 2TB drives (ZFS mirror), it was idling at 90W.

What's the power draw when you pull everything but the CPU and memory?
 

DrMrLordX

Lifer
Apr 27, 2000
23,112
13,215
136
They won't. If they had something remotely interesting for the FX line they would have launched it on the server market first, not on the desktop market.

Obviously. FX is dead, for now. They need an APU that can step up to lead their product lineup. The 7850 is not exactly doing that currently except in HSA-enabled apps.

edit: the price list from xbitlabs looks . . . interesting. That 8320E could be a winner, or it could be that AMD saw how many people were successfully undervolting FX chips and decided to try their hand at it.

And the 9590 for $220? It's like the Pentium D all over again.
 
Last edited:

bononos

Diamond Member
Aug 21, 2011
3,941
190
106
Also bear in mind that FXs are mych more conservative regarding binnings. Intel cpus tend to be binned pretty aggressively, even for being mobile rejects.

What are you talking about? The FX series is known to run hot, even above tdp at stock clocks like theres minimal binning going on, probably because the company is sliding under the waves. And because of that, theres the problem of motherboard throttling at stock settings because vrm's are stressed out.

Its the other way around with Intel. Very conservative and no problems with heat that they even had the cheek to substitute paste for solder underneath the heatspreader.
 

AtenRa

Lifer
Feb 2, 2009
14,003
3,362
136
What are you talking about? The FX series is known to run hot, even above tdp at stock clocks like theres minimal binning going on, probably because the company is sliding under the waves. And because of that, theres the problem of motherboard throttling at stock settings because vrm's are stressed out.

Its the other way around with Intel. Very conservative and no problems with heat that they even had the cheek to substitute paste for solder underneath the heatspreader.

FX operates at lower temperatures than Intel CPUs, you clearly have this the other way around. And TDP is not power or energy consumption.
Also the VRM throttling on some motherboards is not the FX fault but the Motherboards design.
They controlling the VRMs through temperature, other motherboards give you the choice to control the VRMs through Current(I).
 

RussianSensation

Elite Member
Sep 5, 2003
19,458
765
126
Well at $190-220 these processors are a lot more viable against an i5 for someone who wants a 'budget' workstation build without overclocking. For rendering, 3d work, cryptographic tasks, compression/decompression, and video encoding, the FX9590 offers superior performance to an i5. But because of the excessive power usage, the electricity costs over 3 years would make it more expensive to own than an i5 and actually make you question if you should just invest into an i7 for these tasks upfront. For the first time in a while the argument for an i7 for a stock system is really strong as i7 4790K boosts to 4.2-4.4Ghz in the real world.

3dsmax.png

premiere.png

premiere.png

lightroom.png

finereader.png

truecrypt.png

winrar.png

x264.png

freemake.png


This really would be a decent processor for this type of work if the power consumption wasn't sooooo bad. AMD needs to urgently shrink these to 16nm or at least 20nm.

power-3.png

Source

A stock FX-9590 system will probably use as much power as a 4.5Ghz 5960X.....

For gaming though, it's still not competitive even at 5Ghz.

If AMD can't afford to shrink this architecture until 2016, they should really invest more heavily into AMD gaming evolved to get developers to optimize game engines for 8 threads.
 
Last edited:

SAAA

Senior member
May 14, 2014
541
126
116
If Intel is "generally a better buy" than why do they score so poorly on the Passmark Value Chart:
http://www.cpubenchmark.net/cpu_value_available.html

That chart is misleading for a single reason, not because it uses synthetic benchmarks, but because the value it uses for performance is only multythread score.
Of course the module architecture of Bulldozer and successors is good for that being born as server chips but guess what do the same for single-few thread loads and any Sandy bridge CPU is suddenly a much better value/price.

An argument can be made if you need high multythread performance every day for cheap, but then why all servers don't use AMD chips?
 

DrMrLordX

Lifer
Apr 27, 2000
23,112
13,215
136
What are you talking about? The FX series is known to run hot, even above tdp at stock clocks like theres minimal binning going on, probably because the company is sliding under the waves. And because of that, theres the problem of motherboard throttling at stock settings because vrm's are stressed out.

Its the other way around with Intel. Very conservative and no problems with heat that they even had the cheek to substitute paste for solder underneath the heatspreader.

No problems with heat? There are some who complain that their Haswells aren't stable running Prime95 28.5 at stock. While hardly scandalous, it isn't entirely accurate to say that Intel chips are always running cooler thanks to their lower TDPs.

FX operates at lower temperatures than Intel CPUs, you clearly have this the other way around. And TDP is not power or energy consumption.
Also the VRM throttling on some motherboards is not the FX fault but the Motherboards design.
They controlling the VRMs through temperature, other motherboards give you the choice to control the VRMs through Current(I).

What you are saying sort of plays into one of my previous comments in this thread (and serves to highlight why FX price cuts alone may not be enough).

From an enthusiast's point-of-view, the only reason to get an FX is to overclock it to 9590 levels of performance so as to mitigate Piledriver's single-threaded performance problems. So, that means getting an 8320 or 8350 (or 8320SE/8370SE after Sept. 1) and pushing it to 4.5-4.7 ghz at least, depending on what is allowed by the silicon lottery.

Now you have to deal with the fact that only a few motherboards out there reliably support the kind of power draw that Vishera can muster at such speeds. Assuming your overclocked chip is pulling over 200W (which is not outside the realm of possibility), you do not want to be caught with a 4+1 phase board, period.

If you are going to stop at the 4.5-4.7 ghz range, the Gigabyte UD3 is probably the sweet spot with a 6+2 phase design. There are reportedly some folks that have run the 9590 (often with a bit of undervolting) on this board.

If you want your FX to shine in all its glory by shooting for 5 ghz or higher, then it's pretty much the Sabertooth. There are other options, but most of the good ones cost more.

So, just looking at the board, you can expect to pay at least ~$110 for the UD3, or $170 (or more) for the Sabertooth. And that doesn't even include cooling for the chip, which will almost certainly have to be a strong aftermarket HSF or better. Some of us are fortunate enough to already have big HSFs sitting around, but even those might not be enough for the 5 ghz club.

And since neither the UD3 nor Sabertooth have onboard video, you're going to need a vid card too.

Alternatively, if you are just looking for WCG points, you could probably get a cheap 4+1 board and run an 8320 undervolted (or maybe one of the new 8370SEs) for much, much less money.

But, I digress.

For those of us determined to make the most of the FX in a wide variety of tasks, overclocking is a must, and when overclocking comes into play, the need for excellent cooling and solid boards rears its ugly head.

I realize that Kaveri does not offer the kind of multithreaded power of an overclocked FX, but for my money, I would much rather see an A8-7600 on an Asus A88x Plus. Allegedly, the Asus Plus and Pro boards can be made stable at 129 mhz blck in IDE mode, allowing for the 7600 to hit 4257 mhz. You probably wouldn't need much of an HSF to pull that off, either, AND you wouldn't even need dGPU unless you really wanted one.

And all that for maybe $200-$220 for CPU + board.

FX ownership is expensive even before taking utility bills into account.

If board costs can come down somehow, then the prospect of AM3+ being a discount overclocking platform might seem more realistic. As it stands, there are too many other interesting budget overclock options going on right now (G3258, maybe the A8-7600, the 860k) for octal-core FX to fit in, even with a healthy price cut.
 

mrmt

Diamond Member
Aug 18, 2012
3,974
0
76
Also bear in mind that FXs are mych more conservative regarding binnings. Intel cpus tend to be binned pretty aggressively, even for being mobile rejects.

Yes, yes. So conservatively that they still refused to launch the thermal datasheet for the FX line. I'm sure they are doing this in order to not show the consumers how lower their power consumption is from stated TDP.
 

NTMBK

Lifer
Nov 14, 2011
10,491
5,924
136
An 8-core steamroller/excavator at 20-16nm would be exceptional. It is a loss for all of us that AMD didnt use 20nm process for its BigCore designs in 2014.

Meh, I'm happy with their decision to be honest. Instead of spending their limited resources on a processor based on a flawed architecture, they are putting it towards their next generation product. Hopefully the new x86 architecture will be a return to competitiveness.
 

bononos

Diamond Member
Aug 21, 2011
3,941
190
106
FX operates at lower temperatures than Intel CPUs, you clearly have this the other way around. And TDP is not power or energy consumption.
Also the VRM throttling on some motherboards is not the FX fault but the Motherboards design.
They controlling the VRMs through temperature, other motherboards give you the choice to control the VRMs through Current(I).

Do we even know that for a fact? AMD sensors haven't been reporting proper temps since the Phenoms. And its gotten more squirrelly since Bulldozer with odd temps under load (wide variation). Motherboard sensors might not even agree with reported temps like your own testing which shows max temp at 62C but the motherboard socket thermistor at 76C. And AMD strangely hasn't released the usual thermal/power/voltage guidelines.

All which makes sense when you consider the strange throttling issues that some users are reporting because AMD cpu's are probably not binned properly and are probably hitting thermal limits. Or motherboards are throttling the cpu because the VRMs are hitting power limits well in excess of tdp even at stock clocks.

You can't blame motherboards manufacturers when AMD hasn't released the specs for Bulldozer. Max power consumption was always close to the meaning of TDP for both Intel/AMD until Bulldozer. It still holds true for Intel.
 

mrmt

Diamond Member
Aug 18, 2012
3,974
0
76
Do we even know that for a fact? AMD sensors haven't been reporting proper temps since the Phenoms

AMD gave away some extra bits to the server/workstation folks in their website:

http://products.amd.com/en-us/Opter...=1000&f6=G34&f7=C0&f8=32nm&f9=&f10=6400&f11=&

Here, for example, they disclose the maximum temperature of the chip, so indeed FX works at lower temperatures compared to Intel chips. But yet they still refuse to provide the thermal datasheet even for opteron processors.
 

BSim500

Golden Member
Jun 5, 2013
1,480
216
106
FX 8350 has a rated TDP of 125W and the 4670K has a rated TDP of 84W. In practice actual power draw will be higher due to inefficiencies in the power supply. My FX 8350 machine idles at 70W and pulls down 255W full loaded (this is at stock speeds, measured with a Kill-o-watt). That's a difference of 185W, which is at least 48% higher than expected. Assuming the theoretical 4670K machine has a similar increase in power draw it would have a fully loaded power draw of 124W. The difference is then about 40W. Now let's look at electricity costs around the country. Average is $0.0984/kWh (High: Hawaii at $0.34/kWh, Low: Louisiana at $0.0690/kWh). I live in California, whose average is $0.135/kWh.

Good post in general, but just a correction to bold part : My experience with high-end desktop chips is that whilst AMD chips power consumption figures are typically under-estimated, Intel's are often over-estimated, and various benchmarks bear this out for "84w" Intel Haswell's. Eg:-

Anandtech : Stock 4670K - Idle 33w / Load 97w = 64w load delta
http://www.anandtech.com/show/7003/the-haswell-review-intel-core-i74770k-i54560k-tested/2

Hardwareluxx : Stock 4670K - Idle 45w/ Load 109w = 64w load delta
http://www.hardwareluxx.com/index.p...intel-core-i7-4770k-and-i5-4670k.html?start=6

Mine : Running Prime on my i5-3570 = 37w idle / 52w 1T / 66w 2T / 77w 3T / 88w 4T system load = 51w load delta. That's full system load (inc motherboard + SSD + optical drive, USB devices, etc), not just the CPU alone, measured at the wall, not calculated.

So the real difference is even greater, as a measured 97-109w (AC which includes PSU losses) is lower than a theoretical 124w, which widens the running cost gap by even more, ie, a 255w AMD vs a 97-109w Intel = 146-158w difference under load x 24hrs x 365 days on something like "World Community Grid" (essentially "constant folding") = 1,278-1,384 extra kwh's per year. Which at average $0.0984/kWh = $125-$136 high annual running costs. 2-3 years of that = $250-$408 at which point even an i7 looks like a bargain...

And all that's assuming you have just "average" electricity costs, and don't live in an expensive state or country (eg, in the UK, avg electricity is £0.105 which = $0.174/kwh) D: at which point FX chips for any constant 100% load are a catastrophic ripoff including 2-3 years electricity bills.
 

RussianSensation

Elite Member
Sep 5, 2003
19,458
765
126
Meh, I'm happy with their decision to be honest. Instead of spending their limited resources on a processor based on a flawed architecture, they are putting it towards their next generation product. Hopefully the new x86 architecture will be a return to competitiveness.

I doubt it. In a perfect world, even if AMD had caught up in IPC to Intel's next generation architecture, they would continue to be behind Intel's world leading manufacturing node process. It's really not possible to compete with Intel on performance or price/performance in the mid-range and high-end CPU space for anyone. For AMD to have any chance, TSMC/GloFo/Samsung/IBM would need to have a high performance class leading fabrication process -- they don't! But even the first part of AMD matching Intel in IPC is hardly possible given how far behind they are and how much less funds AMD has for R&D/world class engineers. AMD would need to pull a Pentium 4 --> Core 2 Duo type of architectural leap to overcome 65-70% IPC disadvantage, which will become almost 90-100% with Skylake/Cannonlake.

The only timeframe in the history of AMD vs. Intel when AMD was seriously competitive is mostly the result of Intel's failed architecture (Pentium 4/D), not because of anything extraordinary from AMD's CPU division. Had Intel launched Intel Pentium M Banias on the desktop, it would have owned Athlon 64/X2 as well.

Because AMD is going to be constantly behind Intel in the fabrication process, even if their CPUs were to be just as fast for a similar price, AMD's CPUs would use more power. Even if AMD's next architecture manages to match Skylake/Cannonlake in IPC and price/performance, Intel can just lower a 6-core Skylake-E to $299-340 and it's not going to be possible for AMD to compete on the high-end. I think it's only really possible to compete in the high-end with Intel if HSA takes over and the graphics component starts taking over various aspects of general purpose computing/PC tasks. Otherwise, AMD will compete in the lower segments where their APUs and FX-6300 currently reside.
 
Last edited:
Status
Not open for further replies.