AMD unleashes first ever commercial “5GHz” CPU, the FX-9590

[Chiropteran]

Thats not what I said. But AMDs new desktop line for a foreseeable future will be 4T.

What are you even talking about?

Did you read this thread? The FX-9590 and it's lesser sibling will indeed be 8 core. Are you trying to say that the release of these CPU won't occur in the "foreseeable future"?

 

[BallaTheFeared]

Where we're going, we don't need IPC.

I laughed. :thumbsup:

 

[inf64]

There is no 3M APU & 4M/6M SR in AMD server roadmaps. They wont release server versions of these imaginary chips (in 2014), so chances of an enthusiast focused desktop release are pretty much non-existent . They might be able to do it, but they probably dont want to (same as the LGA1150 desktop Broadwell drama). Please accept that before it gets embarassing, your Steamroller FX 6M/12T @ 4GHz beating Haswell-EP 8C/16T in 2014 predictions were already fun enough.

I never said they will launch any of these. I said it is possible from engineering stand point . Get your facts straight please. It's all "what if" speculation and we had no data at all on what AMD planned to do. The roadmap they released just few days ago is newest one and reflects the reality of what they want to do.

 

[thilanliyan]

Screw 220w. Give me a more palatable (perf/watt wise) CPU (Steamroller maybe?) for my AMD system.

 

[Idontcare]

Where we're going, we don't need IPC.

This is true on many different levels.

IPC is like horsepower, and in the age of cheap gas and machoism that fed the muscle car genre the product engineering was all about horsepower.

But what followed was that management found out there was a ridiculously larger TAM waiting to be tapped for vehicles that catered to the everyday life of the soccer mom and commuting worker.

I'd wager Toyota has made more in terms of absolute profits on their Prius lineup than Lamborghini has in its entire existence.

Where compute is going is away from IPC (driven by instruction-level parallelism or ILP) and towards thread-parallelism (TLP).

If you (the customer) really need more performance then you ought to be willing to pay more for it...and paying more for it doesn't mean buying a higher clocked SKU or a higher IPC core design; rather it will mean paying more for a software package that takes advantage of the availability of more cores.

The AMD's and Intel's of the world aren't going to custom build a Dodge Viper to feed a small niche market of muscle-car minded buyers. (they'll leave that to IBM) But they will build designs like Xeon Phi and Seamicro products that deliver a ton of flops for customers who are willing to invest in the creation of software products that can extract TLP.

Itanium is basically the proof, the silly ridiculous academic experiment, into determining whether or not the market is interested in pursuing ILP. The pain that comes from pursuing ILP at nearly any cost turned out to be a bust. The market isn't interested. So TLP is the next best thing.

Soccer moms and commuters everywhere would rather drive a fleet of Priuses than pay for the thrill of their 21st century flying car ala the Jetsons. Practicality trumps all, CPUs are not immune.

 

[Tsavo]

I'd wager Toyota has made more in terms of absolute profits on their Prius lineup than Lamborghini has in its entire existence.

ONE Prius, not their whole production run thus far. I don't know how well they've been doing since Audi bought them, but prior to that, they limped penniless from investor group to investor group.

 

[sushiwarrior]

And what chip and socket should this 8T steamroller FX come from? Servers will be 4T in the 1P segment. So why would the desktop get 8T? Seems counterintuitive.

That's an APU server, "Berlin CPU/APU ... HSA application support (APU)". The CPU model is just the APU without the IGP. So they will not have fully scaled up Steamroller (and had it certified, which is lengthy at best) by the time that product is launching.

But this is besides the point, you're using the SERVER roadmap as some kind of proof it's common knowledge that server processors come later due to the stringent certification process.

 

[SlowSpyder]

That's an APU server, "Berlin CPU/APU ... HSA application support (APU)". The CPU model is just the APU without the IGP. So they will not have fully scaled up Steamroller (and had it certified, which is lengthy at best) by the time that product is launching.

But this is besides the point, you're using the SERVER roadmap as some kind of proof it's common knowledge that server processors come later due to the stringent certification process.

I'm not saying AMD won't have an eight core SR processor, who knows. Its anyone's guess at this point. I doubt anyone thought a 220 watt TDP 4.7/5GHz 8 core Piledriver based part was coming either. But, I do want to point out that if I remember correctly, AMD's Barcelona did indeed launch before Phenom did.

 

[sushiwarrior]

I'm not saying AMD won't have an eight core SR processor, who knows. Its anyone's guess at this point. I doubt anyone thought a 220 watt TDP 4.7/5GHz 8 core Piledriver based part was coming either. But, I do want to point out that if I remember correctly, AMD's Barcelona did indeed launch before Phenom did.

And Barcelona had a TLB glitch, so they probably learned their lesson about launching early after that

 

[guskline]

As usual IDC a great post! Some on this forum "bash" a poster when they use a car analogy but I think it's a great device. In my own instance, we own a Solstice GXP as our fun car. Tuning it cost @$400 and I was able to get 281HP at the wheels from a 2 liter turbo charged Direct injection engine making it a hot little car. HOWEVER a Corvette ZR-1 it is not. You pay @$135,000 for that priviledge! THAT is HOT!!!!!!!!

 

[mrmt]

And Barcelona had a TLB glitch, so they probably learned their lesson about launching early after that

Bulldozer was launched more or less simultaneously in desktop and servers, and the launch management was a success from a quality point of view.

 

[Charles Kozierok]

I don't really see the issue as ILP versus TLP. I see TLP as something that they are pursuing simply because they have no other choice -- ILP's been mostly maxed out and CPU speeds have stagnated, so all that's left is duplicating the hardware and telling the software people "you make it work!".

People choose a Prius over a Lamborghini for a number of reasons, but if they could get something with the advantages of both, they'd probably buy it. The chip companies may want us to think that clock speeds are languishing because "nobody needs anything faster", but we all know that if they could put out a 10 GHz chip, they'd do it post-haste.

 

[Zucker2k]

IDC's car analogy is spot on. Only, these chips are expensive and will guzzle gas, just like a lamborghini. On the other hand, relative performance will be disappointing; it'll struggle to beat the 4-seater Prius while only carrying two passengers (Desktop i7 has an igp, right? Right?) ;-)

 

[Makaveli]

Where we're going, we don't need IPC.

Win!!

 

[Piotrsama]

Per your request I ran POV Ray ver 3.7 RC7 on the 8350 stock and at 4.7 Ghz and the 3770k stock and at my OC of 4.4Ghz.
For 8 threads
8350 stock 2m30s
8350 @4.7Ghz 2m31s
3770k stock 3m17s
3770k @4.4Ghz 2m 43s

What do you get on your 8350 at 4.7Ghz?

I already hypothesized you were running OCed both times.

My 8350 scores are:
4000Mhz 2m57s
4334Mhz 2m45s
4615Mhz 2m35s

 

[BallaTheFeared]

2m25s with i5-4570k @ 4.8GHz

 

[guskline]

2m25s with i5-4570k @ 4.8GHz

Impressive for a true four core. Shows the power of the Haswell core.

 

[Makaveli]

Impressive for a true four core. Shows the power of the Haswell core.

That is impressive I score

2m26s @ 4Ghz
2m18s @ 4.2Ghz

Will be interesting to see what kinda of numbers Haswell E will put down.

 

[BallaTheFeared]

I will fully offer up the possibility that I'm doing it wrong!

 

[guskline]

BTW: I just cranked the FX 8350 up to 5 Ghz on all 8 cores and ran Aida64/Cinbench 11.5/ Passmark 8 cpu only to give some perspective of what the proposed FX 9570 might perform like. From what I read it will be 4.7 Ghz on all 8 cores and turbo to 5 Ghz so I'll assume that a 8 core 5Ghz test result "might" be higher than the FX 9570 at stock.

Here are the results added to my sheet comparing a FX8350 stock vs a 3770k stock and at the designated OC pts. The memory read/copy/latency figures for the 8350 @ 4.6Ghz is slightly scewed due to my running the fsb at 219 while dropping the multiplier to 21. (21 x 219= 4.6Ghz)

 

[monstercameron]

BTW: I just cranked the FX 8350 up to 5 Ghz on all 8 cores and ran Aida64/Cinbench 11.5/ Passmark 8 cpu only to give some perspective of what the proposed FX 9570 might perform like. From what I read it will be 4.7 Ghz on all 8 cores and turbo to 5 Ghz so I'll assume that a 8 core 5Ghz test result "might" be higher than the FX 9570 at stock.

Here are the results added to my sheet comparing a FX8350 stock vs a 3770k stock and at the designated OC pts. The memory read/copy/latency figures for the 8350 @ 4.6Ghz is slightly scewed due to my running the fsb at 219 while dropping the multiplier to 21. (21 x 219= 4.6Ghz)

you posted in both topics...
what were the temps and powerdraw w/ oc intel and oc amd.
I am wondering does the intel part lose all its efficiency clocking that high.