[B3D News]

Back in 2008, we talked a few times about AMD's Propus chip - basically a Deneb with no L3 cache and 5-10% lower performance in desktop applications - and how important it is to AMD's prospects in desktops. Now Fudzilla claims a 45W EE SKU is slated for early Q2 and others for early Q3.

Read the full news item

[OICAspork]

Have they even done a memory cell test?

Intel has already produced its first functional Westmere:

http://www.dailytech.com/First+Westm...ticle14273.htm

[Tchock]

http://itbbs.pconline.com.cn/diy/9870895.html


~5% slower than Deneb in compute intensive, cache less-sensitive (for K10, not for Core) Cinebench R10. Bad in SuperPi, though.

Estimated pricing: 115USD. I reckon 120USD is the max pricing, as AMD's P-II pricing has never overridden across series. So 6XX should cost less than 710.

[Blazkowicz]

very interesting CPU, I'll recommend it to a friend who does audio creation, with a Gigabyte GA-M56S-S3 mobo (über cheap at 53 euros and full featured with 4 PCI slots and 3 firewire ports).

If it's cheaper than an X3 710, that would mean around 50 euros cheaper than an Intel Q8200. That's a lot of raw power for media tasks at such a low price.

[mczak]

Originally Posted by Tchock

~5% slower than Deneb in compute intensive, cache less-sensitive (for K10, not for Core) Cinebench R10. Bad in SuperPi, though.

Like 15% slower than X4 with same clock and 6MB cache? Not too bad if that's about the worst performance hit. I'm wondering though, maybe memory latency is slightly better with these chips without L3 cache?
Actually though the performance hit in Cinebench R10 seems to be huge, a X4 920 seems to score around 12000, so that would be a 30% performance hit. Not so good.
Slightly disappointing though not all of Propus/Rana cores fit in 65W.
Also native dual core (Regor) is still some way off (until then still no real replacement for the aging X2) and even more disappointing no 45W version on the roadmap.

[Tchock]

^

That's 64-bit vs 32-bit Cinebench. Your Deneb score is on 64 while this Propos is on 32. Notably, AMD catches up to Intel much better on 64-bit CB, so it's interesting to see this compared to the upcoming Q7500.

Legitreviews has both sets so I used theirs.

[mczak]

Originally Posted by Tchock

^

That's 64-bit vs 32-bit Cinebench. Your Deneb score is on 64 while this Propos is on 32. Notably, AMD catches up to Intel much better on 64-bit CB, so it's interesting to see this compared to the upcoming Q7500.

Oops, you're right. There's a quite unexpectedly (?) large difference between 32bit and 64bit cinebench R10 - depending on cpu like 10-20%.
I thought though the Q7500 was just a false rumour? Otherwise though those chips will indeed have no quad-core competition from intel. Not sure such a Q7500 would make sense for intel however. These new amd chips will be significantly cheaper to produce thanks to the much smaller die (early rumours said ~140mm2 instead of 260mm2 - seems a bit exaggerated even though amd could also remove all but one of the HT links apart from cache), but there's no savings for intel (they have only 2 45nm core2 chips in production afaik, wolfdale-6m and wolfdale-3m), so maybe they'll try to compete with dual-cores instead.

Legitreviews has both sets so I used theirs.

They only seem to show times not scores though what's the factor to convert that...

[Tchock]

OOPS: See the Legitreview Phenom II 940 review.


Assume linear scaling, it's only a tad more for the real result.

[Accord1999]

Originally Posted by mczak

I thought though the Q7500 was just a false rumour? Otherwise though those chips will indeed have no quad-core competition from intel. Not sure such a Q7500 would make sense for intel however. These new amd chips will be significantly cheaper to produce thanks to the much smaller die (early rumours said ~140mm2 instead of 260mm2 - seems a bit exaggerated even though amd could also remove all but one of the HT links apart from cache), but there's no savings for intel (they have only 2 45nm core2 chips in production afaik, wolfdale-6m and wolfdale-3m), so maybe they'll try to compete with dual-cores instead.

I'd imagine if Intel releases a Q7500 or something similar, its main purpose will be to knock the competing AMD product back into price points that Intel is happy at and at a level that won't make AMD money.

[fellix]

Propus die snapped

No L3 cache at all, 512K L2 per core and only a single HT interface.

[mczak]

Originally Posted by fellix

Propus die snapped

No L3 cache at all, 512K L2 per core and only a single HT interface.

So indeed a bit larger than the initially rumoured die size (168mm^2 instead of 140mm^2). The article is a bit misleading though with comparing die size to core2 quad, since at best this will compete with q9x00 though I suspect rather q8xxx, both of which are made of wolfdale-3m not wolfdale-6m, which are only 82mm^2 instead of 107mm^2. And I dunno about comparing to E8xxx Core2 Duo, sure it may end up even slightly cheaper and certainly beat it in multithreaded apps, but the same is true for the C2Q 8xxx series...
I wonder though if the effect of the missing L3 cache (plus the fact it only has 512KB L2 cache) will hurt more on this quad core than on the dual-core Athlon II X2 in practice.

[Tchock]

The cheapest Intel C2Q8200 on Newegg is 160USD.

If AMD ever gets round their arse to release a non-EE version of Propos (currently I think only EE ones are slated), the naming/pricing convention would indicate a < 120USD price tag for 2.5-2.7Ghz. General performance figures should be Athlon II vs E5300 all over again, this time with less advantage in gaming but a huge one in price.

X3 4__s would be even cheaper, between the Phenom II X2s and Athlon IIs, so 90USD tri cores?


At least we know for now that they OC decently at least even on the energy efficient models (3.6Ghz first try on supposedly prerelease silicon by some mobo maker rep)

[Gubbi]

Are there any laptops planned with these CPUs ?

Relative latency of the main memory system more than halve when the CPU is throttled down to 0.8-1 GHz for battery operation, reducing the impact of lacking a L3.

Cheers

[Tchock]

That'll come in 2010. The September refresh of Athlon/Turion IIs will only include Regor AFAIK.

That's rather nifty, but I do wonder if the extra power consumption is worth it in this case. The current Turion X2 Ultras (Griffin) already do it with a separate NB component & power plane I think.

[mczak]

Originally Posted by Tchock

That's rather nifty, but I do wonder if the extra power consumption is worth it in this case. The current Turion X2 Ultras (Griffin) already do it with a separate NB component & power plane I think.

That's true, but I'd say this is definitely necessary, those Turion X2 Ultras can hardly compete with mobile intel celerons. Those mobile regors should be faster (both because of the core improvements and possibly slightly higher frequency) and I'd expect them to not draw more power (Griffin is still 65nm). Nothing to threaten the better mobile core 2 duo cpus, but at least something competitive at the lower end.

[Tchock]

A small update.

Looks like AMD is going for an all-out price war vs the Q8 series. They can afford to, in a sense.

[bearmoo]

The quad core Caspian should come with some transistor improvement. So should Magny-cour I guess. They both need it.

[mczak]

Originally Posted by bearmoo

The quad core Caspian should come with some transistor improvement. So should Magny-cour I guess. They both need it.

I don't follow you there. Caspian is the dual-core followup from Griffin (rumours said actually it might be shrinked Griffin but looks like mobile Regor to me). Champlain is the future quad-core mobile part (looks like mobile propos...). Magny-Cours it the monster 12-core chip (actually it's mcm with 2 sao paulo).
I'm not sure what improvements actually there are expected? Faster switching speed (hence higher possible clock), more densely packed transistors or what? There doesn't really seem to be anything on the published AMD roadmaps indicating that.

[bearmoo]

Oops I mixed up the code name and yes I meant to say Champlain. It's just that with today's 45nm AMD cpus they won't be able to make Champlain/Magny-Cours with acceptable clock speeds. So the process has to improve to make those. The slide indicate new strain techniques which If I am not mistaken can help cpus achieve clock speeds with lower voltages(You can have faster clock or same clock with lower voltage), hopefully enought to maintain a, say 2.5GHz and slip Champlain into a 35w thermal. Or I could be they are working on something like Intel's Turbo Mode to cheat the thermal wall. I could be they will have both but I might be too optimistic.

[Tchock]

Champlain's target clocks wrt Propos-energy efficient is actually okay.


2.2, 2.3Ghz - 45W, + NB.


Calpella Nehalems (Clarksfield), according to HKEPC
http://global.hkepc.com/3477

1.6Ghz 6MB - 45W - $364
1.7Ghz 8MB - 45W - $546
2.0Ghz 8MB - 55W (XE) - $1054

If Clarksfield is 45W, then AMD should have some good chances with lower (< 300USD) pricing. It's K10 and it's 2.2Ghz... it just doesn't have Turbo mode (which should be appreciated much more on the laptop segment, I guess)

[Blazkowicz]

those are very high prices (as much as a full computer), Intel wants high margin high end laptops, perhaps with an Intel SSD along it.

I believe propus will actually compete with dual core nehalem and then dual core westmere. They will best each other depending on the tasks you run.

[mczak]

Originally Posted by Blazkowicz

I believe propus will actually compete with dual core nehalem and then dual core westmere.

There will be no dual core nehalem. Havendale/Auburndale are canceled, with the westmere parts (clarkdale/arrendale) appearing a bit earlier than originally thought.

They will best each other depending on the tasks you run.

Something like a 3Ghz dual-core westmere part will obviously run circles around a 2.2Ghz propus for single/dual-threaded workloads (unless amd implements turbo-like abilities, even then it can't win). Even with heavily multithreaded apps it could be quite close. And I fear in terms of perf/watt it will not compete neither (of course, you could say that's unfair a 45nm part against a 32nm part...). Maybe those improvements GF mentioned indeed help, but I dunno I wouldn't expect drastic improvements from it.