Would you buy a Steamroller/Excavator based AMD FX CPU if it existed today?

[DrMrLordX]

......urban legends are still used to sustain flawed argumentation and continualy misleading realities about the FX line.

Urban legends? Seriously? Go back and look at what Flank3r did with this 8370E, or better yet, have a look at some of the lower-clockspeed results people are turning in with the 8320E.

http://www.overclock.net/t/1518055/fx-8320-e-there-is-life-in-my-old-dog/0_100
http://www.overclock.net/t/1519949/overclocking-the-fx-8320e-with-a-gigabyte-990fxa-ud5-rev3-0/0_100

That's just a few of them. Problem is that results like those won't turn up in typical 8370E or 8320E reviews since . . . reviewers don't care to tune the CPU to see if it can hit 4.5 ghz at less vcore than your typical 8350.

Bottom line: The Stilt told us we'd see results like this, Flank3r more-or-less confirmed it, and now people are discovering that the E chips tune pretty well (i.e. generally better than the 8350). If you think all those people are just blowing smoke, or aren't testing their CPUs well enough, you can take it up with them. There just aren't a lot of data points to go off right now.

 

[Abwx]

The urban legend moniker wasnt directed to you, i thought you would had understood.

While we are on the topic of CPUs for the AM3+ eventualy i ll point for thoses who tweaks their settings that, at least one year ago, all FXs were not born equals, and that 8350 are better than 8320, 6350 and 4350 being the same compared to the 6300 and 4300 respectively.

One doesnt simply buy a 8320/6300/4300 and get the 8350/6350/4350 by just changing the settings, the former named are inherently better according to hardware.fr overclocking/downclocking/undervolting/overvolting made on their samples, this has also been confirmed during the 8350/8370E comparison and the 8370 should follow the trend by being noticeably better than the 8350 as well.

 

[Chicken76]

Would you buy a Steamroller/Excavator based AMD FX CPU if it existed today?

Yes I would buy a 4-module (even a 3-module) Excavator to upgrade my Zosma. But only if it were manufactured using a process suitable for CPUs. The current Steamroller APUs are a bit of a letdown. I understand that the process used is a CPU-GPU tradeoff, but HEDT needs some attention too.

 

[AtenRa]

It wasn't a decision made in a vacuum to AMD's interests though.

It was a business decision reached by both parties once GloFo had enough data to determine how much it would cost to develop the process tech that AMD wanted, and AMD told GloFo they were not interested in paying for it if it was going to cost that much.

Ultimately it was AMD who made the business decision to cancel their own SKUs and for GloFo to cancel the development of the high performance 20nm process.

No different than what happened to 45nm and 32nm at TSMC (the customers told TSMC not to spend R&D money developing those nodes, not the other way around).

In process tech, IDM and foundries alike, it is not a "build it and they will come" business. You only start building it (start developing the node) if at the same time you have customers who are willing to sign contracts (4 yrs out) or letters of intent to have their chips fabbed on that node.

If there isn't enough seed-interest at time T-4yrs, then come time T the node will have never been funded for development in the first place. But it is the customers who ultimately direct the foundries in terms of what to build, when to build it, and how fancy (expensive, performance, etc) to make the node.

It really isn't correct to lay blame of AMD's product SKU situation at the feet of their foundry when it was AMD themselves who made the business decision to direct GloFo to go cheap on its 20nm process flow, knowing well in advance that such a decision had the unavoidable consequence of undermining the viability and competitiveness of AMD's planned CPU designs.

Maybe i wasn't clear, what i meant was that the initial planing was to use 32nm Gate First SOI for Bulldozer and Vishera (2011 to 2013), then they would use IBMs 22nm Gate First SOI for SteamRoller and Excavator (2014-2016).
I wasnt trying to say who/why canceled the 22nm Process, i was only trying to tell why both Steamroller and Excavator (server/desktop) parts were canceled due to the shifting of priorities and the restructuring of AMD from 2011 onwards.

 

[Chicken76]

... all FXs were not born equals, and that 8350 are better than 8320, 6350 and 4350 being the same compared to the 6300 and 4300 respectively.

One doesnt simply buy a 8320/6300/4300 and get the 8350/6350/4350 by just changing the settings, the former named are inherently better according to hardware.fr overclocking/downclocking/undervolting/overvolting made on their samples...

I don't understand this. Which ones are better in your oppinion?

 

[Alatar]

I would buy a SR/EX FX chip if it clocked like Vishera did, I always have fun with LN2 and Vishera.

However for normal use I wouldn't. I've got 3 platforms + CPUs that would perform better anyway.

8370E has better perf/watt than a 4670K in many apps

I can't see how this could be true (in any meaningful amount of applications, I'm sure you can find cherrypicked examples of almost anything) considering that an 8370E still pulls more than a 4790K but also performs about on par with an 8150 on average.

Vishera is old. Mobos are outdated, it only performs well in extremely niche scenarios and even then usually only against i5s.

I own three but they're still old and slow and when cooling needs, mobo features, often bad performance in games etc. is taken into account the value proposition is also quite shaky.

 

[Lepton87]

It's really sad that AMD can't quite catch up to the 980X a CPU that was released in March of 2010 and here we are over 4 years later and AMD still doesn't have a comparable CPU, while they have come close for a stock performance that was only possible at the expense of a very high power consumption and no overclocking headroom.(9590 220W TDP). Even still an overwhelming majority of games runs better on the i7. And an overclocked i7 980X is just untouchable for AMD's CPUs. How much more time will have to pass for AMD to release a CPU that would be competitive with an overclocked i7 980X/970(if you don't count those crazy-priced extreme fleece edition CPUs)? And I'm not talking about some cherry-picked i7 Galftown samples(at 4.8-5GHz), just an average OC of 4.2-4.3GHz.

 

[Idontcare]

Maybe i wasn't clear, what i meant was that the initial planing was to use 32nm Gate First SOI for Bulldozer and Vishera (2011 to 2013), then they would use IBMs 22nm Gate First SOI for SteamRoller and Excavator (2014-2016).
I wasnt trying to say who/why canceled the 22nm Process, i was only trying to tell why both Steamroller and Excavator (server/desktop) parts were canceled due to the shifting of priorities and the restructuring of AMD from 2011 onwards.

Oh yeah, oops sorry I misread the conversation then Carry on, don't mind me!

 

[jpiniero]

It's really sad that AMD can't quite catch up to the 980X a CPU that was released in March of 2010 and here we are over 4 years later and AMD still doesn't have a comparable CPU

I imagine the 8M Opterons are faster in multithreaded apps. Well, maybe.

AMD is done competing against Intel on high end desktops. That's just the reality of things. They don't have the resources to chase every little market; they have to be choosy.

 

[AtenRa]

Oh yeah, oops sorry I misread the conversation then Carry on, don't mind me!

No worries, it is always a good thing to read your opinions especially on fabrication subjects

One more thing,

AMD choice to ditched 22nm in favor for 14nm FF had a tremendous negative affect on the company. They effectively ditched the Bulldozer architecture and node development within half the life time of the original plan. That changed the entire company's planing from mobile to desktop to server roadmaps creating chaos and products never mend to exist just to feel the gaps created.
It also negatively affected the APU line not only by using a half node (28nm) instead of a much better all around 22nm SOI. But also because it made Big-die Quad Module dies be directly competitive against smaller APU SKUs.

But in the long run i believe it could be beneficial for them, if they will be able to sustain their current status and finish 2015 without any significant losses.
A new mArch designed for perf/watt and a new 14nm FF process in 2016 could give them a tremendous advantage even against Intel in the APU department.

 

[DrMrLordX]

The urban legend moniker wasnt directed to you, i thought you would had understood.

No, but it is directed at the people whom I am citing (I understand it is also directed at those who claim unrealistic results for older 8320 chips or tout their golden chips as the norm, but it should be noted that I am not discussing those results. They are outliers, or just plain liars).

While we are on the topic of CPUs for the AM3+ eventualy i ll point for thoses who tweaks their settings that, at least one year ago, all FXs were not born equals, and that 8350 are better than 8320, 6350 and 4350 being the same compared to the 6300 and 4300 respectively.

One doesnt simply buy a 8320/6300/4300 and get the 8350/6350/4350 by just changing the settings, the former named are inherently better according to hardware.fr overclocking/downclocking/undervolting/overvolting made on their samples, this has also been confirmed during the 8350/8370E comparison and the 8370 should follow the trend by being noticeably better than the 8350 as well.

Correct. From the original FX lineup, the 8350 is better than the 8320. Same wafer, with binning to differentiate product.

hardware.fr notwithstanding, every result I've seen thus far indicates that the 8370E is now eating the 8350's lunch, and the 8320E is basically doing the same. Yeah there's some good 8350s out there (and at least one bum 8320E chip that I've seen), but overall I'm seeing 4.5 ghz @ 1.36v or lower from the E chips, which is better than what most 8350s can put up.

Originally I had thought the 8320E and 8370E had come from the same wafers as the other chips and just had different stock/turbo speeds and VIDs to fit into the 95W power envelope. At first glance, it appeared to be so. Then people tried overclocking them, and now it is apparent that things are quite different.

At ~$140-$150, the 8320E is now the go-to FX chip, in my opinion. Sure, it's binned lower than the 8370E and (by logical extension) the 8350, but it does not come from the same wafer as the 8350. Process refinements really did make a difference, enough so that the 8320E deserves a nod. It is possible that the 8310 is capable of similar feats and may amount to an even-better deal than the 8320E, but most people buying them seem to be pairing the with motherboards that do not permit much overclocking, so there are no available data points.

 

[SlowSpyder]

I haven't torture tested, but my FX 9370 seemed pretty happy at ~1.225v, 4.4GHz. The FX 8350 is set at 1.35v for 4.0GHz (4.2GHz turbo) from the factory, I believe.

 

[Abwx]

Correct. From the original FX lineup, the 8350 is better than the 8320. Same wafer, with binning to differentiate product.

Very likely, the 8320 use more voltage than the 8350 to get to 4.0, in principle the latter will dissipate 10W less at said frequency, this is at stock of course.

hardware.fr notwithstanding, every result I've seen thus far indicates that the 8370E is now eating the 8350's lunch, and the 8320E is basically doing the same. Yeah there's some good 8350s out there (and at least one bum 8320E chip that I've seen), but overall I'm seeing 4.5 ghz @ 1.36v or lower from the E chips, which is better than what most 8350s can put up.

Originally I had thought the 8320E and 8370E had come from the same wafers as the other chips and just had different stock/turbo speeds and VIDs to fit into the 95W power envelope. At first glance, it appeared to be so. Then people tried overclocking them, and now it is apparent that things are quite different.

Hfr got 4.5 at 1.32V and 160W TDP with Prime 95, this should translate to 128W with Fritzchess and perhaps even less with CB 11.5, it is obvious from their tests that the E versions are significantly better than the regular 8350.

At ~$140-$150, the 8320E is now the go-to FX chip, in my opinion. Sure, it's binned lower than the 8370E and (by logical extension) the 8350, but it does not come from the same wafer as the 8350. Process refinements really did make a difference, enough so that the 8320E deserves a nod. It is possible that the 8310 is capable of similar feats and may amount to an even-better deal than the 8320E, but most people buying them seem to be pairing the with motherboards that do not permit much overclocking, so there are no available data points.

They possibly could have improved their process the same way that when in 45nm with the Phenom X4 reaching 3.7 and the X6 getting 3.3 at the same TDP.

That said i dont think that both the 8300 and 8310 are as good as the E iterations, same as with the 8320/8350, they need more voltage than the recently released 95W SKUs which would had deserved a new chipset, even a basic one with just 4 USB3, 4 SATA 3 would be welcomed, it s not much about thoses features and PCIe 3.0, set apart the former this can be soved with some chips, it s rather about cutting the plateform power draw by 15-16W at iddle and 18-20W when the plateform is fully stressed, fortunately the ones with ancient chipsets a la 760G/785/790/8XX that were normaly dedicated to the Phenom X6 have less of this problem and the E FXs are dedicated to thoses MBs according to AMDs statements.


About the 6C-4C i ll clarify my point, the 4350/6350 are better binned parts that need less voltage at the higher frequencies of interest than the 4300/6300 respectively, if one is to get a FX6300 it s better to take a 6350 and downgrade it at 6300 settings, they also overclock better according to Hfr.

 

[mrmt]

Sucks being AMD right now, but big risks can bring big rewards or utter failure.

I think the Q414 CC will be very interesting in terms of direction. I think Rory leave is indicating further bumps in the R&D pipeline, bumps that will have to be addressed with the proverbial axe.

 

[SlowSpyder]

Hfr got 4.5 at 1.32V and 160W TDP with Prime 95, this should translate to 128W with Fritzchess and perhaps even less with CB 11.5, it is obvious from their tests that the E versions are significantly better than the regular 8350.

Do you have a link to that? I have a spare AM3+ board laying around that's going to get build. I'd consider getting an 8320E (or 8370E) for it, or swapping out my CPU for a new one if I can get more out of it and/or undervolt more. Just curious to see what they were able to do.

 

[Abwx]

Do you have a link to that? I have a spare AM3+ board laying around that's going to get build. I'd consider getting an 8320E (or 8370E) for it, or swapping out my CPU for a new one if I can get more out of it and/or undervolt more. Just curious to see what they were able to do.

The ocking numbers are at the bottom of the page :

http://www.hardware.fr/focus/99/amd-fx-8370e-fx-8-coeurs-95-watts-test.html

The right column number is the power at the 12V rail level, it is to be factored by 0.9 to get the CPU TDP, notice the difference between VID and the actual voltage measured by a probe, the stress soft used is Prime 95.

Anyway if your numbers are those :

I haven't torture tested, but my FX 9370 seemed pretty happy at ~1.225v, 4.4GHz.

Then your 9370 is significantly better than their 8370E, assuming your VID is accurate then the TDP should be 126-134W under Prime 95 and 100-108W with Fritzchess or CB 11.5, in comparison THG needed 1.26V at 4.5 with apparently a better sample than HFR.

http://www.tomshardware.com/reviews/amd-fx-8370e-cpu,3929.html

IIRC you posted the power delta for your set up using CB 11.5, if your delta at the main is in the 130-140W ballpark then your VID is accurate.

 

[DrMrLordX]

I haven't torture tested, but my FX 9370 seemed pretty happy at ~1.225v, 4.4GHz. The FX 8350 is set at 1.35v for 4.0GHz (4.2GHz turbo) from the factory, I believe.

That's quite extraordinary, making your 9370 better than even the E chips (best I've seen at 4.4 so far is 1.33v, though that was only an 8320E).

Hfr got 4.5 at 1.32V and 160W TDP with Prime 95, this should translate to 128W with Fritzchess and perhaps even less with CB 11.5, it is obvious from their tests that the E versions are significantly better than the regular 8350.

Then I think we are mostly in agreement here. The FX paradigm has shifted somewhat.

They possibly could have improved their process the same way that when in 45nm with the Phenom X4 reaching 3.7 and the X6 getting 3.3 at the same TDP.

I was thinking the same thing. The improvement from C3 to R0 was pretty big. Some people are reporting better memory controller performance on their E chips which was what happened with R0 as well.

That said i dont think that both the 8300 and 8310 are as good as the E iterations,

Probably not, though it would still be interesting to see what the 8310 can do.

Do you have a link to that? I have a spare AM3+ board laying around that's going to get build. I'd consider getting an 8320E (or 8370E) for it, or swapping out my CPU for a new one if I can get more out of it and/or undervolt more. Just curious to see what they were able to do.

If you look at the prices, there are 9370s for around $209 right now, and I've seen them dip down to sub-$200 prices on a few occasions. The 8370E isn't much cheaper. If all 9370s are as good as the one you have right now, you might be better off just getting another one of those, assuming you have a HSF for it.

edit: there's also 9590s for $219:

http://www.ncixus.com/products/?usa...pn=FD9590FHHKWOF&manufacture=AMD&promoid=1353

Only lasts until the 12th.

 

[Ramses]

Yes I would if it's inline with the price/performance of the last few chips.
I was happy with my 955BE, the 1090T after it, the 8350 after it, and the 9590
now. I've got zero reason to upgrade other than boredom and I don't expect that
to change in the next year at least. I've got a couple of heavy business users
at work that just in the last few months upgraded to 1090t based boxes from ancient stuff and they get long fine for cheap, and am about to build a database/order management software "server" most likely with an FX cpu.

 

[SlowSpyder]

The ocking numbers are at the bottom of the page :

http://www.hardware.fr/focus/99/amd-fx-8370e-fx-8-coeurs-95-watts-test.html

The right column number is the power at the 12V rail level, it is to be factored by 0.9 to get the CPU TDP, notice the difference between VID and the actual voltage measured by a probe, the stress soft used is Prime 95.

Anyway if your numbers are those :



Then your 9370 is significantly better than their 8370E, assuming your VID is accurate then the TDP should be 126-134W under Prime 95 and 100-108W with Fritzchess or CB 11.5, in comparison THG needed 1.26V at 4.5 with apparently a better sample than HFR.

http://www.tomshardware.com/reviews/amd-fx-8370e-cpu,3929.html

IIRC you posted the power delta for your set up using CB 11.5, if your delta at the main is in the 130-140W ballpark then your VID is accurate.

Thank you for the link, heading in to work, will check it when I get in.

I've used my FX for some short benches that way and for some day to day use and gaming. But never really pushed it with anything I would consider a tougher test than Cinebench 15, I can't vouch for it being completely stable at 4.4 / 1.225v, but it hasn't given me any problems either, so it could be. But before I'd consider it anywhere close to that, I'd have to run it through the paces of course. I'll mess around tonight a bit, I should have time. I have a kill-o-watt, the power use under load went up ~120 watts running Cinebench.

 

[SlowSpyder]

If you look at the prices, there are 9370s for around $209 right now, and I've seen them dip down to sub-$200 prices on a few occasions. The 8370E isn't much cheaper. If all 9370s are as good as the one you have right now, you might be better off just getting another one of those, assuming you have a HSF for it.

edit: there's also 9590s for $219:

http://www.ncixus.com/products/?usa...pn=FD9590FHHKWOF&manufacture=AMD&promoid=1353

Only lasts until the 12th.

Last year I remember the FX9370 going for $199.99 with Far Cry 3 included. Hoping for some great deals like that (that FX9590 price isn't bad at all either) as we get closer to the holidays. A rather low priced FX8320E would be perfect.

 

[Abwx]

Slowspyder, what was your power delta with Cinebench 11.5 or R15 at those settings.?

This should be enough to have an idea, no need to molest the thing with Prime 95, we know that this soft will use 20-25% more power than CB or Fritz if the CPU has AVX.

 

[rancherlee]

Maybe Microcenter will have 99$ 8320E chips this year. I'm happy with my "Old style" 99$ 8320 I got this summer, 4.5 @ stock voltage (cool n' quiet still on, turbo off). I had it down to 1.1v at the stock 3.5 and didn't try lower. Pretty happy with the performance right now, don't think I'd upgrade to different chip if they released one for the AM3+

 

[SlowSpyder]

Slowspyder, what was your power delta with Cinebench 11.5 or R15 at those settings.?

This should be enough to have an idea, no need to molest the thing with Prime 95, we know that this soft will use 20-25% more power than CB or Fritz if the CPU has AVX.

My machine idles at about 120 watts on the Kill-o-Watt. Running Cinebench R15 the highest peak draw I saw was 242 watts during the bench, a couple watts below that during most of it.

 

[Abwx]

My machine idles at about 120 watts on the Kill-o-Watt. Running Cinebench R15 the highest peak draw I saw was 242 watts during the bench, a couple watts below that during most of it.

Thanks for the numbers, that s very valuable info, power under CB R15 is comparable to CB 11.5, the VID seems quite accurate in respect of real voltage, your CPU TDP was barely 98W at peak...

 

[WT]

Would I ?? Hell yea !! I'm TRYING to buy one, but finding the real issue is finding a good 970 board that offers 8+2 phase power. The crap boards that seem to be everywhere are offering 4+1 which means you can't OC any chip worth a damn.

Using a 965BE, and while it is fine for what I use it for, the urge to upgrade is getting stronger.