AMD Phenom II 940 "Xtremely" Benchmarked

[Malovane]

High res gaming has never been about the CPU though, and its always inevitable that the CPUs bunch up when the GPU does the brunt of the work.

Fair point about the synthetics, I don't pay much attention to them myself personally, but check out the 3D rendering benches, Deneb is still somewhat lacking in that respect.

Pretty much what we can gather so far is that Deneb is a decent gaming CPU but slightly lacking in other areas compared to Core 2 and especially i7.

Well, that could be part of AMD's long term tactics at play. I don't think they see the need to optimize their CPU's as much for rendering and encoding applications if they are going the fusion route. They (and Nvidia) will push for more GPU usage in applications, which would utterly demolish any CPU based processing of the same work. Will they succeed? I don't know... but it seems like both AMD and Nvidia will be pushing that theory heavily this year and the next. They just need to get their basic programs working right to demo it, and this latest AVIVO release was a poor attempt.

As for high res gaming and the CPU... that may have been true in the past. Phenom I, and to a larger extent i7 show that it is not all CPU and GPU bottlenecks, and that a high bandwidth IMC can in fact alleviate some issues in games.

 

[fazers_on_stun]

http://www.xtremesystems.org/forums/showthread.php?t=213200&page=5

Shootout @ 3.7GHz - Deneb vs Kentsfield vs Yorkfield vs Bloomfield

http://i279.photobucket.com/albums/.../misc/deneb_vs_kentsfield_yorkfield_bloom.png
http://i279.photobucket.com/albums/...t/misc/deneb_vs_kentsfield_yorkfield_bl-1.png

Slightly short of Kentsfield levels of performance, gotta say thats rather disappointing, especially after all the hype from the fanboys @ XS about 15 - 20% gains from a die shrink... yeah right.

Hmm, I'd say we should wait a few more days until the NDA lifts and we can see comparisons from AT and Tom's and other sites, before passing final judgment on Deneb. My own opinion is that it'll turn out somewhere between a Kentie and a Yorkie, but then I'm a known Intel fanboy!

Unfortunately for AMD, this may be their finest for the next 2-3 years, since their own roadmaps show Bulldozer pushed back to 2011. Hopefully the 6-core version plus some clock and steppings tweaks will keep them in the game when Westmere comes out in a year.

 

[WR]

Anyone have new ideas to explain why i7 performs so poorly in GPU-limited gaming? It seems in almost every non-gaming benchmark, i7 leads by a considerable margin and doesn't hurt from HT. Then in high-res gaming, i7 loses not just to P2 but to Q6600, and HT's detrimental impact is seen left and right.

Is it a problem with the X58/PCIe pathway, or is the issue inside the CPU? If inside the CPU, as suggested by the effect of toggling HT, why don't other benchmarks detect the issue?

For now, it isn't a convincing point to say the P2 beats i7 in some benchmark when the under-$200 Q6600 also beats i7 in that same bench.

 

[JAYDEEJOHN]

Ill post this here as well, this is P1 vs Q6600, clock for clock, and shows its 1% slower than Deneb. http://www.xtremesystems.org/forums/showthread.php?t=213642

Dont know about everyone else, But I suspect we need more benches done, as P1 being only 1% slower just doesnt add up

 

[roofus]

i thought the NDA was supposed to be lifted weeks ago. now its a day to day pursuit to see what the actual performance is. preliminarily it looks good but nothing has emerged beyond the site in Greece and the guys at xtremesystems (which may be right on or may be alot of fluff).
more benches from places and people that do not have a vested interest in making it look good or bad.

 

[fazers_on_stun]

Anyone have new ideas to explain why i7 performs so poorly in GPU-limited gaming? It seems in almost every non-gaming benchmark, i7 leads by a considerable margin and doesn't hurt from HT. Then in high-res gaming, i7 loses not just to P2 but to Q6600, and HT's detrimental impact is seen left and right.

Is it a problem with the X58/PCIe pathway, or is the issue inside the CPU? If inside the CPU, as suggested by the effect of toggling HT, why don't other benchmarks detect the issue?

For now, it isn't a convincing point to say the P2 beats i7 in some benchmark when the under-$200 Q6600 also beats i7 in that same bench.

I wouldn't say it "performs so poorly" - rather, it doesn't show the same improvement as in other benches, where the GPU appears to be the bottleneck anyway.

As for why this occurs, I dunno - clearly HT is not needed for single-threaded games and would hinder performance, whereas Turboboost would be useful there. I think Intel tried to cover both low and high-threaded scenarios with i7, although maybe not as successfully as we would have liked. Also, doesn't the addition of L3 cache raise the overall memory latency?

 

[fazers_on_stun]

i thought the NDA was supposed to be lifted weeks ago. now its a day to day pursuit to see what the actual performance is. preliminarily it looks good but nothing has emerged beyond the site in Greece and the guys at xtremesystems (which may be right on or may be alot of fluff).
more benches from places and people that do not have a vested interest in making it look good or bad.

I thought the NDA was supposed to be lifted with the introduction at CES, which is tomorrow IIRC. So maybe this time tomorrow Tom's will have some reviews out Certainly by Friday I would guess.

 

[JAYDEEJOHN]

Exactly. Perf at this point isnt and wont be fully known. Clocks is another thing, and what weve seen so far is pretty conclusive, but just not so regarding perf

 

[roofus]

if p2 doesnt scale well with multiple GPU's it will never know what my desk looks like. being that there was already a link with it running on a 780a board with SLI and it did quite well i think it most likely does. this is the time AMD should be flooding the tech sites with products to bench. let everyone drool over them prior to release. i thought they waited an awful long time with the i7's but this leaves everyone to speculate how good or bad it is. not doing their product justice IMHO.

 

[JAYDEEJOHN]

Theyve given out the clocks and ocing abilities. I agree, Id like tosee more for future ideas on porchases as well. Im thinking AMD is still somewhat playing it close to the vest so to speak, since the Barcy debarcle. I think we will still see some interesting results yet, and it may not just be clocks and simple perf.

 

[roofus]

they probably are being cautious but in fairness they got a bad rap with the Barcy. it was not as dismal as it was made out to be. their own hype and then not being able to deliver to the level of hype definitely hurt so it kind of makes sense.

 

[epsilon84]

Ill post this here as well, this is P1 vs Q6600, clock for clock, and shows its 1% slower than Deneb. http://www.xtremesystems.org/forums/showthread.php?t=213642

Dont know about everyone else, But I suspect we need more benches done, as P1 being only 1% slower just doesnt add up

No, it shows Agena 4% behind Kentsfield, you can't take existing Deneb results and extrapolate it into new data since the tests are totally different. And FYI, the Q6600 is tested on an old P965 chipset mobo, newer chipsets such as the P35 have much improved memory performance and are in general a few percent quicker: http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=2991&p=1

 

[JAYDEEJOHN]

And you know for a fact everything youve seen on P2 is on the best boards with the best bios? Plus, 2+2 does equal 4, as you can see a 4% overall in real world apps, when you can just go here http://translate.google.com/translate?sourceid=navclient&hl=en&u=http%3a%2f%2fwww%2edinoxpc%2ecom%2fTests%2farticoli%2farticolo%2findex%2easp%3fid%3d866
And see that P1 isnt close to 1% within P2

 

[roofus]

we haven't even seen the p2's unchained yet. they are running them on existing platforms so far as i have seen.

 

[caamsa]

I am as excited about the Phenom II launch as the next person. But I certainly am not overly concerned with most of these benchmark comparisons. I imagine the CPU will do rather well and if AMD can keep the prices right I am sure a lot of enthusiasts will pick one up to play around with since it seems to OC well.

Go back a year ago when AMD launched the first Phenom which many think was a flop. Search the forums and you will find a plethora of posts stating that AMD will go out of business blah blah blah no one will buy them etc......Well it is now a year later and AMD is still hanging in there and will continue to hang in there despite the economy. Their GPU division is doing rather well.

AMD is no longer number 1 and had not been for a while.....so what, who cares.....nothing wrong with number 2. Intell is a huge company.....they should be number one......shouldn't they?

AMD will continue to do its best and provide good solid CPU's that are competitive with Intell. That is why I purchased my K6-2 back in 1999. It was a great chip in the super 7 shuttle board and was much cheaper than the Intell solution at the time.

 

[epsilon84]

And you know for a fact everything youve seen on P2 is on the best boards with the best bios? Plus, 2+2 does equal 4, as you can see a 4% overall in real world apps, when you can just go here http://translate.google.com/translate?sourceid=navclient&hl=en&u=http%3a%2f%2fwww%2edinoxpc%2ecom%2fTests%2farticoli%2farticolo%2findex%2easp%3fid%3d866
And see that P1 isnt close to 1% within P2

No mate, you can't mix and match data from totally different benchmarks and make conclusions based on that. I don't know how much clearer I can make it, but those two tests are totally unrelated. One is a mix of synthetic/gaming/3d rendering and the other is entirely real world apps.

If you also can't see the relevance of using a modern chipset (P35/P45/X48/X58) that offers up to 5% performance gains over the 2.5 year old P965, then so be it. While we're at it, why not use 2.5 year old AM2 mobos for Phenom testing, since that is obviously not a problem for you. 😉

Anyway, here is a 'proper' Phenom II review, its in Italian but benchmark charts are universal.

http://www.dinoxpc.com/Tests/articoli/articolo/index.asp?id=866&p=1&Phenom+II+940+Deneb%3A+AMD+finalmente+a+45nm

Of particular interest to you would be this page, which shows the gains from PI to PII:
http://www.dinoxpc.com/Tests/articoli/articolo/index.asp?id=866&p=4&Phenom+II+940+Deneb%3A+AMD+finalmente+a+45nm

The biggest gains (10%+) are in low res gaming, but in everything else its around a 5% improvement overall.

 

[JAYDEEJOHN]

First off, they couldnt oc past 3.6, so its not the best review. I understand about the chipset, but if thats what you have, thats what youll get, regardless. Im sure even tho, as in the same link you gave as I did, that better boards and chipsets are on the horizon for P2, as well as bios', but we have to take what we have now right? So, ultimately, the AM3 with the 8xx chipset is going to be much faster too. I understand what youre saying, but the only constant here is, the P2 is more similar to the P1, and itll be more than 1% faster on the same board and setup as the P1, so, its easier to make such a comparison, and valid, tho not exact

 

[epsilon84]

First off, they couldnt oc past 3.6, so its not the best review. I understand about the chipset, but if thats what you have, thats what youll get, regardless. Im sure even tho, as in the same link you gave as I did, that better boards and chipsets are on the horizon for P2, as well as bios', but we have to take what we have now right? So, ultimately, the AM3 with the 8xx chipset is going to be much faster too. I understand what youre saying, but the only constant here is, the P2 is more similar to the P1, and itll be more than 1% faster on the same board and setup as the P1, so, its easier to make such a comparison, and valid, tho not exact

Um, how does the overclocking results effect the quality of a review?! Maybe not all PIIs will hit 4GHz effortlessly on air as some people would have you believe?

As for the chipset, it actually makes a bigger difference for the Intel CPUs (except i7) because the memory controller is off die, and as I showed earlier P35 is ~5% quicker than P965 in memory performance. It makes less difference for AMD because of the IMC, I doubt AM3 will be much faster than AM2+ since its essentally for DDR3 compatibility, I know AMD are claiming a 5% increase with AM3 though so we'll see.

 

[roofus]

even 3.6 is pretty much average OC for an Intel quad (i7's seem to be capable of more with less effort) so even if that is the "norm" that isn't bad at all.

 

[JAYDEEJOHN]

But, as you said, maybe all of then cant get to 4Ghz, also, youd know that AMD gains alot from their chipsets as well. Id say its more than just 5% between a 6xxx and a 7xx chipset. So, does this site know this and other things as well when it comes to AMD, or, are they so used to Intel, they dont know the nuances required to actually make a AMD cpu show its best?

 

[roofus]

very true. just as the other site used an ancient 965 chipset there are alot of gaps that require people with a good understanding of what it takes to get the CPU's on a somewhat level playing field to show some decent results.

 

[amdfangirl]

I'm gonna wait for THG to post its review before I make a decision. Lots of sites post false data to gain publicity... If I remember correctly, there were many sites claiming the the Phenom beat the Core 2 Quad (Kenty) by like 20%.

 

[roofus]

yea there are some suspect numbers and comparisons but i think overall it isn't hard to gain some insight to what they have. they shrunk the die, lowered the thermals, gained overclock headroom. those are the givens. the actual comparisons are hard to get a grip on until we see some familiar, knowledgeable, and reputable sites get their paws on them.

 

[JAYDEEJOHN]

Yea, if theyve only seen the server apps, and extrapolate from there, then it shows how reliable those sites are. They certainly werent doing their jobs, and AMDs hype machine at the time didnt help any at all

 

[amdfangirl]

AMD's chips have always been more about the servers than the desktops. We'll see in a couple of days.