- Status
With my modest knowledge, i don't think HT have massive multithreaded gains. As i understand, only when one thread stalls, the second executes. In that regard, if you have two threads, 2 cores are always much better than 1 core with HT, but you already know that.
As most of you know, the real gain in HT rarely exceedes 15-20% in unoptimazed software, and in some cases it is known to acctualy hurts performance. Better the software, less are the HT gains. So, i find it funny when someone compare 4C/4T with 2C/4T. It will never be even close.
But, what i don't understand is does improving branch prediction acctualy helps HT or hurt it? Intel is already awesome on that field.
Meh... Phenom (I) and C2Q were priced similar as well when Phenom launched. It's hard to make heads or tails on performance based on those prices even if the information is accurate. Thanks for the info all the same, though.
frostedflakes
Diamond Member
- Mar 1, 2005
- 7,925
- 1
- 81
Saw those posted on another forum, and that was my initial reaction as well. The FX-8130P being priced similarly to the 2600K probably means performance is going to be similar. I'm expecting it to be maybe 5% faster than the 2600K.
Not sure what AMD is going to do when Intel releases their 6C/12T SB-E in Q4, though, don't see how 4M/8T Bulldozer will be able to compete with that (assuming performance is in fact comparable to the mainstream 4C/8T Sandy Bridge). AMD doesn't have any plans to release 6M/12T Bulldozer in 2011 do they? AFAIK they don't.
videoclone
Golden Member
- Jun 5, 2003
- 1,465
- 0
- 0
AMD will fight intel's Sandy E with Komodo
Komodo has already taped out and is being tested, so i think if they want they could always release it at the end of 2011
At the moment in on the AMD roadmap its stated for 2012
http://blogs.amd.com/fusion/2010/11...-all-about-velocity/dt-roadmap_with-footnote/
"Komodo": a 32nm CPU featuring up to 10 "Bulldozer" cores, designed for high-performance and enthusiast desktops
Komodo has already taped out and is being tested, so i think if they want they could always release it at the end of 2011
At the moment in on the AMD roadmap its stated for 2012
http://blogs.amd.com/fusion/2010/11...-all-about-velocity/dt-roadmap_with-footnote/
"Komodo": a 32nm CPU featuring up to 10 "Bulldozer" cores, designed for high-performance and enthusiast desktops
Last edited:
Even though phenom I was crap it had a longer life than just a few months, didn't it?
podspi
Golden Member
- Jan 11, 2011
- 1,982
- 102
- 106
In the server market, AMD usually prices comparing throughput, not singlethreaded performance. In the desktop market, this doesn't make much sense (since almost everything most people do is lightly threaded), but AMD hasn't had much choice lately (X6's are within spitting distance of Intel's CPUs in terms of throughput, but are no match for singlethread performance). I think this is great news, because it looks to me like BD is back to being priced off of relative single thread performance. I think for throughput an 8-core BD will be competitive with 6C/12T SB-E.
Not sure how AMD is going to react to IB, it might be hard for them to catch back up again in singlethread performance, but hopefully AMD gets to 22nm before Haswell. Honestly I was hoping for CPUs in the ~ $500 range at least, but if this is accurate I guess not.
Lol phenom I had a lifespan of about 1 year while AMD struggled to make a viable competitor to Core 2 Duo/Core 2 Quad.
Sad thing was, by the time Phenom II came out, Nehalem had already been released, and so Phenom II was only competitive with the already old Core 2 architecture.
One more thing. Let's not forget that even the latest Phenom II X4s are slower clock for clock compared to a 4.5 year old Intel Core 2 Quad Q6600. A 2.6 GHz Phenom II X4 910 gets beaten by a slower-clocked 2.4 GHz Q6600 in most benchmarks:
Last edited:
Cerb
Elite Member
- Aug 26, 2000
- 17,484
- 33
- 86
Hypertransport, not Hyperthreading. Both AMD and Intel used a shared Front Side Bus (FSB), until the K8 for AMD, and Nehalem for Intel (though, they began addressing the problems with Yonah and Conroe). With the FSB, if a CPU needs to communicate to another CPU (memory address changes, potential cache updates, and the like), it needed to go on the bus, tell the North Bridge, and the north bridge would go tell the other CPU. All CPUs shared one giant memory interface. As more CPUs needed to be added, the bus had to be slowed down to accommodate them, as well, on top of the resource contention issues of sharing the bus in the first place.
Hypertransport, one of several Alpha fruits ripened by AMD (didn't Microway have a lot to do with HT 1.0, as well?), works on a point-to-point basis. Starting with the K8, each CPU could communicate directly to one another, or, in configurations with many CPUs, make a couple hops over other CPU links, to get to the CPU they needed. Each CPU socket had its own memory. It was faster for AMD CPUs to make a couple jumps to get to another CPU's memory, than it was for Intel's to hit their shared memory interface, there was more bandwidth available v. Intel's FSB, and the systems could have a great deal of memory bandwidth, too (2 channels per socket).
Intel's QPI is pretty much equivalent to HT, with a few extra features, and is an expression of Intel's famously severe NIH syndrome
.If those are real prices, then I'm actually happy. Sure, it doesn't destroy Sandy Bridge, but at LEAST they have seemingly caught up to Sandy Bridge. It's a HUGE stepup from the Phenom II, which struggled to beat even Socket 775.
Who knows? If it happens to be a "blunder" like Phenom I was, then Bulldozer still has potential to improve a lot.
Who knows? If it happens to be a "blunder" like Phenom I was, then Bulldozer still has potential to improve a lot.
Yeah I know, but well...I don't see why they would price something in a competitor's range if it couldn't keep up with it. Also, that is why I said "seemingly".
AMDs pricing is almost always commensurate in performance with similarly priced Intel chips. Thats why AMD keeps mum on pricing till the last minute, so as to not let the cat out of the bag. So if the rumored prices are real, we have an indication how BD will perform.
But there are other clues. I am quite sure that if they had a big hit, we would have seen leaked benches a long time ago. Intel gave a performance preview of Conroe 6 months before it was released, they knew how valuable early impressions of a yet to be released product can be. But AMD here, like with Barcelona, seems intent on keeping tight lips with BD. I think just based on that, even without pricing clues, hint at a product just barely able to keep up with SB.
But there are other clues. I am quite sure that if they had a big hit, we would have seen leaked benches a long time ago. Intel gave a performance preview of Conroe 6 months before it was released, they knew how valuable early impressions of a yet to be released product can be. But AMD here, like with Barcelona, seems intent on keeping tight lips with BD. I think just based on that, even without pricing clues, hint at a product just barely able to keep up with SB.
What stood out for me was the "P" part. That really makes me think that BD isn't performing/clocking as well as AMD had hoped. To compete with the 2600K price range they had to up the TDP to 125W from 95W. Remember the last time AMD did this was with the Phenom 965 BE at 140W when the 955 BE only needed 125W. AMD basically upped the voltage of the 965 BE to get it to run at that speed and the CPU never overclocked any better then the 955 BE (because it basically was a 955 BE factory overclocked by AMD).
Why is P odd? Intel does the same thing with Sandy Bridge S and T versions. You have a normal and a low powered version for people willing to sacrifice a little performance for less electricity used.
Even so, if AMD can do that and it's a hit in the server market, they'll be fine.
The pricing looks really good for AMD. So good, in fact, that either it is false or Intel is in deep, deep trouble.
I don't want to read too much into it, because it could be BS, but just consider the following:
1. The FX4110 is priced the same as the i5 2500K. That means AMD can match intel core for core.
2. Given the minimal performance improvement of the 2600K over the 2500K, intel is screwed at the $300 mark. The only reason anybody would buy a 2600 over a 2500 is because they want the hyperthreading for highly threaded applications. If BD 4110 is matching the 2500K then the 8130P can be up to twice as fast as the 2500K in multithreaded scenarios, blowing the 2600K out of the water.
3. AMD is shipping 95W TDP chips across the line, that means they are doing pretty well in terms of their yields, and it could also mean quite a bit of overclocking headroom.
4. Apply all the above logic to servers and Intel is fucked.
Again, of course, the whole price list could be horseshit.
I don't want to read too much into it, because it could be BS, but just consider the following:
1. The FX4110 is priced the same as the i5 2500K. That means AMD can match intel core for core.
2. Given the minimal performance improvement of the 2600K over the 2500K, intel is screwed at the $300 mark. The only reason anybody would buy a 2600 over a 2500 is because they want the hyperthreading for highly threaded applications. If BD 4110 is matching the 2500K then the 8130P can be up to twice as fast as the 2500K in multithreaded scenarios, blowing the 2600K out of the water.
3. AMD is shipping 95W TDP chips across the line, that means they are doing pretty well in terms of their yields, and it could also mean quite a bit of overclocking headroom.
4. Apply all the above logic to servers and Intel is fucked.
Again, of course, the whole price list could be horseshit.
All the i7 9xx have 130W, and so will the upcoming Sandy Bridge-E. And AMD's current X6 are 125W mostly. IMO, it seems a pretty expected TDP for 8 cores.
Its odd because the 95W version is the normal version and the 125W "P" version is the odd ball. The SB S and T version is the other way around, lowered powered version and not the flag ship version. Like I said the last time AMD did this was with the 965 BE because the 955 BE (there flagship CPU) wasn't competitive enough with Intel's offering at the time they upped the voltage to get a 965 BE but by doing so increased the TDP. This "P" scenario sounds like exactly the same thing.
@drizek
If I remember reading correctly the 4 cores BD is clocked significantly higher then the 8 cores I believe around 3.8 GHz, while the 8 cores is around 2.8GHz (this is with all cores active). The 8 cores can run faster when some of the cores aren't in use, so there's no way the 8 core is twice as fast as the 4 core in multi-threaded situations or twice as fast as 2600k.
125W is the normal TDP for high end chips. The fact that they are shipping 95W parts is really good news. Buy a 95W, overclock it to use around ~140W, and you have a nice, fast system. I guess thats what the P version is, a 125W version of the 8110 with higher base clocks and much higher turbo.
I bet the 4110, given that it has the same TDP as the 8-core, is going to actually be a gimped 8-core. If true, either it will have ridiculously high clcokspeeds to hit 95W, or AMD isn't able to fully gate the 4 inactive cores. I seriously hope that it is the former. The latter was the case with Phenom II, where the X3s had the same TDPs as the X4s at the same clock speeds because the 4th core was still there leaking juice. Not a huge deal for desktops, totally bad news for laptops and servers if their power gating isn't working.
Their ES naming leak says that the model number indicates base clock speed. I expect the 4-core to have the highest turbo mode, but I also expect the 4110/6110/8110 to all ship at the exact same base frequency. The 8130 would be a few hundred MHz higher for the base frequency, and probably have the same turbo frequency as the 4110(which would then make it, give or take, 70% faster than the 2600K in multithreaded apps).
And, for the millionth time, since I am just wildly speculating here, everything I am saying is based just on price, based on a price list that some guy posted on the internet...
I bet the 4110, given that it has the same TDP as the 8-core, is going to actually be a gimped 8-core. If true, either it will have ridiculously high clcokspeeds to hit 95W, or AMD isn't able to fully gate the 4 inactive cores. I seriously hope that it is the former. The latter was the case with Phenom II, where the X3s had the same TDPs as the X4s at the same clock speeds because the 4th core was still there leaking juice. Not a huge deal for desktops, totally bad news for laptops and servers if their power gating isn't working.
Their ES naming leak says that the model number indicates base clock speed. I expect the 4-core to have the highest turbo mode, but I also expect the 4110/6110/8110 to all ship at the exact same base frequency. The 8130 would be a few hundred MHz higher for the base frequency, and probably have the same turbo frequency as the 4110(which would then make it, give or take, 70% faster than the 2600K in multithreaded apps).
And, for the millionth time, since I am just wildly speculating here, everything I am saying is based just on price, based on a price list that some guy posted on the internet...
As drizek said, 125W is pretty normal for high end CPUs. It's 8 cores after all, look at Intel, any 6-core above 3GHz is 130W.
And it's just designations, Intel could have marketed 65W as normal version and 95W as performance version for SB...
And 965BE was 140W from 125W, not 95->125 and this was a later edition.
Where did you get this from?
Something I just realized. If these prices are for 1k trays, than BD should demolish SB!
If the FX4110 = 190$, then I guess it will be sold to Newegg for 250$. And newegg sells it to the consumer at 300$ Same price as the 2600. Now this is just my guess but that's how it usually works. Also the FX8130P might retail for around 500$!
Last edited:
- Status
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes + WCL Discussion Threads
- Started by Tigerick
- Replies: 22K
-
News NVIDIA and Intel to Develop AI Infrastructure and Personal Computing Products- Started by poke01
- Replies: 384
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
Facebook
Twitter