[LegionHW]CPU scaling with the Radeon 5970

http://www.legionhardware.com/document.php?id=869&p=0

The only thing lacking is the i5 in the article.All in all good read for the potential HD5970 and 58x0CF buyers.

Quote:

---

Originally Posted by Conclusion

The Phenom II X4 results were quite different to those recorded when testing with the Core i7 processors, though this was not necessarily a bad thing. When operating at lower clock speeds, the Phenom II X4 did not fair all that well, as we saw a sharp decline in performance. However when clocked at 3.0GHz and beyond, the Phenom II X4 really picked up the pace, and in many cases was able to outclass the Core i7.

In games such as Wolfenstein, Call of Duty: Modern Warfare 2, Tom Clancy’s H.A.W.X, BattleForge and Far Cry 2 the Phenom II X4 processors were actually faster when clocked up near 4GHz! This is quite amazing as out of the 9 games tested, the Phenom II X4 series was faster than the Core i7’s in 5 of them. Although the margins were very limited, the Phenom II X4 was found to be faster, and had it just managed to match the Core i7 series with the Radeon HD 5970, we would have been impressed.

While the Phenom II X4 matched the Core i7 in Crysis Warhead, the only games where it failed was Company of Heroes Opposing Fronts, Left 4 Dead 2 and Batman Arkham Asylum. The Phenom II X4 was noticeably slower in these games, making the Core i7 the better choice here. Still, for the most part we found the Phenom II X4 to be every bit as good as the Core i7 processors when gaming with the new Radeon HD 5970.

Having said that, we recommend that AMD users looking at buying this powerful graphics card make sure that they have a Phenom II X4 processor that is clocked at 3.0GHz or greater. Most Phenom II X4 processors are capable of overclocking to 3.0GHz and beyond, while the more high-end options, such as the Phenom II X4 955 and 965 processors, come clocked at 3.2GHz and 3.4GHz respectively.

While we hardly expect there will be many users trying to pair a $600 US graphics card, such as the Radeon HD 5970, with a budget processor, it is nice to see that the sub-$200 US processors are up to the task. The Intel Core i7 920 proved to be more than powerful enough at $280 US, while the AMD Phenom II X4 955 will work just as well at $165 US, giving users plenty of great options.

Reviewed By Steven Walton

---

Quote:

---

Originally Posted by informal

http://www.legionhardware.com/document.php?id=869&p=0

The only thing lacking is the i5 in the article.All in all good read for the potential HD5970 and 58x0CF buyers.

---

Noticed that where the performance isn't limited by GPU, i7 2.0GHz is as fast as 2.6-2.8GHz phenom. I wonder what perf we should expect from even cheaper i5/i3.

Makes me wonder why I even bothered overclocking my processor to 4.2ghz while I'm still on a single GTX 285 with a 30 inch monitor LOL! :D

I just decided to keep my system at stock for now. It's so powerful i don't see any need to overclock just yet. Plenty of unused cores as well. But i have free room up to 4GHz and that should be enough for another few more years. Maybe a gfx card swap, but for now even that is usually overkill (i have a small LCD, but it does the job done).

Quote:

---

Originally Posted by kl0012

Noticed that where the performance isn't limited by GPU, i7 2.0GHz is as fast as 2.6-2.8GHz phenom. I wonder what perf we should expect from even cheaper i5/i3.

---

There seems to be some internal bottleneck in Deneb core when it runs below certain frequency.You can see Deneb outpaces i7 after certain threshold is reached(while both approach 4Ghz) in 5 out of 9 games.

I love Legion Hardware articles on bottlenecks but I wish they did a Q9650 @3.8ghz just for a reference point added into the mix.

Good stuff.
I remember reading before about 4890 CF (iirc?), 2.0GHz i7 was enough, now it's up to 2.6, slowly getting there. :up:

No minimum framerates.

What you will find is that the differences in average fps are relatively minimal. But if you look at the minimum frame rates, you will find a difference.

Bit-tech always lists the minimum frame rate in their CPU scaling reviews, and there you can always see a difference.

If one low end has got a 25fps minimum, and the higher end a 30fps minimum, then the low end CPU does not deliver a smooth gameplay experience, while the higher CPU will.

I do not understand reviewers still not see this simple truth.

Quote:

---

Originally Posted by Tim

No minimum framerates.

What you will find is that the differences in average fps are relatively minimal. But if you look at the minimum frame rates, you will find a difference.

Bit-tech always lists the minimum frame rate in their CPU scaling reviews, and there you can always see a difference.

If one low end has got a 25fps minimum, and the higher end a 30fps minimum, then the low end CPU does not deliver a smooth gameplay experience, while the higher CPU will.

I do not understand reviewers still not see this simple truth.

---

I agree!

I don't even understand whay they ever show the maximum frame rates in benchmarks.. ? :wth: As they don't matter one bit. It has always been how poorly a GPU played the game, that matters..!

Quote:

---

Originally Posted by Xoulz

I agree!

I don't even understand whay they ever show the maximum frame rates in benchmarks.. ? :wth: As they don't matter one bit. It has always been how poorly a GPU played the game, that matters..!

---

I agree, for almost a decade certain websites have done reviews like that and they are certainly enlightening.

Many people wanted to copy the gameplay scenarios in benchmarks/reviews, only to see that the actual gameplay wasn't that smooth at all, and actually had to lower one or two settings.

There are quite a few sites which consistantly prove that benchmarks should be done in the aforementioned way, one of them for instance is HardOCP, whilst they aren't everyone's favorite, they do very much give an insight into proper smooth gameplay.

You can see this particularly well in multigpu setup reviews, in particular benchmarks where you can see the minimum fps really tank in the teens due to driver issues, whilst the maximum and average fps fall terribly short of showing this issue.

That's why over time, (if you have been reading reviews long enough) you develop a preference to certain sites and their reviews.

so conclusion -> if you want a HD 5970, get a Ci7 or a P2 which is clocked at atleast 2.6ghz.

Quote:

---

Originally Posted by Hornet331

so conclusion -> if you want a HD 5970, get a Ci7 or a P2 which is clocked at atleast 2.6ghz.

---

continued: and you OC your PhenomII to ~3.8-4Ghz and system around it outpaces Ci7 based one at similar clock :)

Seems a PhII 975 idea is a good one.

Wonder what stalls deneb under the 3ghz speed??

Moral of the story either get the i7 920 or the PhII 955 both are good horses for the race but i do wish they used a i5 750 also.

Quote:

---

Originally Posted by informal

continued: and you OC your PhenomII to ~3.8-4Ghz and system around it outpaces Ci7 based one at similar clock :)

---

With one fps?
There are games that really like high clocked P2s, like FC2 and Hawx. But most of the time your running in the gpu bottleneck and i don't count differences of one fps as a win. Neither for Ci7 nor for P2.

Definitely was a surprise to see that at high clocks the phenom sometimes wins...

Quote:

---

Originally Posted by ajaidev

Seems a PhII 975 idea is a good one.

Wonder what stalls deneb under the 3ghz speed??

Moral of the story either get the i7 920 or the PhII 955 both are good horses for the race but i do wish they used a i5 750 also.

---

I think it's own core speed bottlenecks it's L3 cache speed, if they told us I7 uncore speed and Phenom II Nb speed they will see a bigger difference.

I think I7's L3 cache is 8 way, while AMD L3 cache is 48 way, clocks, latency is even different.

Quote:

---

Originally Posted by Hornet331

With one fps?
There are games that really like high clocked P2s, like FC2 and Hawx. But most of the time your running in the gpu bottleneck and i don't count differences of one fps as a win. Neither for Ci7 nor for P2.

---

so a 5970 is a gpu bottleneck ?
should not be so for I7, suppose to do better a multi-card setup.

Very interesting article, would be interesting to fill it out with some of the older core2quad chips (96xx/68xx series) for reference.

Quote:

---

Originally Posted by demonkevy666

so a 5970 is a gpu bottleneck ?
should not be so for I7, suppose to do better a multi-card setup.

---

For certain games, yes... for both P2 and Ci7, it just stops scailing past a certain clockspeed of the cpu.

Look a Farcry2 -> Its stops scaling at 2.6 ghz for Ci7 and 3.2 Ghz for P2.

Same goes for crysis 3.6ghz for P2 and 2.6ghz for Ci7.

In gaming the GPU is still the most limiting factor. When you have a current gen quadcore (Ci7 or P2) the biggest factor "limits" your gaming expirience (that and crappy profiles for xfire/sli :p: )

I wanted to see some core 2 processor comparisons

while they may not be as good compared to an i7, i wonder if both are clocked at 3.6 ghz - 4ghz how much performance is actually missing (this is good info for older systems [like mine])

Quote:

---

Originally Posted by orangekiwii

I wanted to see some core 2 processor comparisons

while they may not be as good compared to an i7, i wonder if both are clocked at 3.6 ghz - 4ghz how much performance is actually missing (this is good info for older systems [like mine])

---

http://www.legionhardware.com/document.php?id=807&p=0

there you go

As for the min. fps not being listed,author responded here:

Quote:

---

Originally Posted by Steve@ LegionHW

Last time we did an article like this there was a huge debate about not adding minimum framerates. However the problem is that we constantly saw quite a difference in performance when recording just the minimum and running the same test three times would often produce three results that were quite different. Despite this the averages did work out to be much the same.

In any case I do not believe that the Phenom II X4 processors for example are producing a noticeably lower minimum frame rate and then a noticeably higher maximum frame rate to generate the same averages as a Core i7 processor.

---

Bold text is practically summing it all up. Min. fps measurement depends on all sorts of stuff(dips due to hdd swapping,loading levels etc.) and average is much more stable .Deneb core is practically giving 99% the same results as i7 in most games and this proves that there is enough muscle in this core to feed a monster like 5970.I still remember some users claimed a while ago that when new gen of cards lunches ,the i7 will prove it is the chip to have.Although i7 is still the fastest chip overall,Phenom IIs proved to be just as good all around gaming chips as i7s,for less money.

Why would you want to know minimum framerates? You are only going to notice a dip in framerate if you get several frames in succession that are slow to render, you aren't going to notice a single slow frame. But Minimum framerate does not provide you with any information regarding the preceding or following frames, and so the only thing you can conclude from comparing two different minimum framerates is that the single slowest frame in one benchmark run was still faster to render than the single slowest frame in another benchmark run. You don't know the reason, and you don't know if it was a lengthy enough dip to be noticeable.

I would had liked to see the 975 against the 965 BE.

Quote:

---

Originally Posted by informal

There seems to be some internal bottleneck in Deneb core when it runs below certain frequency.You can see Deneb outpaces i7 after certain threshold is reached(while both approach 4Ghz) in 5 out of 9 games.

---

By a few FPS, it's almost within the noise. All he really showed is that as powerful as the 5970 is, at 2560x1600 and full AA and eye candy, the 5970 is the FPS limiter.

something is not right about these benches.

since the cpu's vary only by 200 mhz and the benches used aren't particularly accurate, how are there no paradoxical results?

one would expect to see slower cpu's just randomly beating faster versions once in a while, even if they averaged a few runs. Instead, everything fits together too neatly.
I think they fudged some of their results.

Quote:

---

Originally Posted by JumpingJack

By a few FPS, it's almost within the noise. All he really showed is that as powerful as the 5970 is, at 2560x1600 and full AA and eye candy, the 5970 is the FPS limiter.

---

Jup, maybe its time for quadfire. :p

Or reduce the res to 1920x1200, to take off a bit load from the gpu.

Do not think the speed of the processor is especially important in games where performance is determined by the processor. The most important thing hits in the cache.
It is evident how i7 will get good results even at low clock. i7's use L2 cache effectively to retrieve data in advance at light loads (eg the calculation of triangles). While Phenom may go more often to memory and find data, i7 will find it in cache.
As complexity rises the effectiveness of the prefetchers are lowered. High complexity (fragmented data, much code) need to have a cache that is able to manage that situation which is the strength of AMD.

Need better multicore support in games especially when IPC gains seem to be halting while more focus is put on bringing out more cores that can't be utilized anyway. UT3 that I play the most have a pretty nice support, going from E8400 @ 4.0GHz to i7-860 @ 4.1GHz and minimum FPS on a certain map when very crowded went from 70 to 170 FPS. When I check with Lavalys it reports 35~55% utilization for two of the cores and 70~90% for the other 2. Task Manager also shows that all cores are utilized well.

Quote:

---

Originally Posted by RPGWiZaRD

Need better multicore support in games. UT3 that I play the most have a pretty nice support, going from E8400 @ 4.0GHz to i7-860 @ 4.1GHz and minimum FPS on a certain map when very crowded went from 70 to 170 FPS. When I check with Lavalys it reports 35~55% utilization for 2 cores and 70~85% for the other 2.

---

Indeed.
Hope CryEngine 3 has a decent multicore support.

Quote:

---

Originally Posted by zalbard

Indeed.
Hope CryEngine 3 has a decent multicore support.

---

Maybe this?
http://blogs.battlefield.ea.com/batt...p-germs.aspx##

Recommended Frostbite PC Specifications for BFBC2 & BF1943
Processor: Quadcore
Main memory: 2GB
Graphics card: GeForce GTX 260
Graphics memory: 512MB
OS: Windows Vista or Windows 7

i wish they listed the uncore clock and HT clock of each overclock, or at least the bus speed X multiplier both processors were set at

Quote:

---

Originally Posted by gosh

Maybe this?
http://blogs.battlefield.ea.com/batt...p-germs.aspx##

Recommended Frostbite PC Specifications for BFBC2 & BF1943
Processor: Quadcore
Main memory: 2GB
Graphics card: GeForce GTX 260
Graphics memory: 512MB
OS: Windows Vista or Windows 7

---

Sweet!

I find this ironic as all these reviews always show [T]ard OCP's are so piss poor since all their reviews the AMD rig always seems to get half or 3/4's of the FPS an intel rig gets.

Quote:

---

Originally Posted by RPGWiZaRD

Need better multicore support in games especially when IPC gains seem to be halting while more focus is put on bringing out more cores that can't be utilized anyway. UT3 that I play the most have a pretty nice support, going from E8400 @ 4.0GHz to i7-860 @ 4.1GHz and minimum FPS on a certain map when very crowded went from 70 to 170 FPS. When I check with Lavalys it reports 35~55% utilization for two of the cores and 70~90% for the other 2. Task Manager also shows that all cores are utilized well.

---

I agree wholeheartedly.

UT3 still amazes me at how well its coded. I would argue that there are few games (if any) that rival the visuals offered by UT3 yet still yield the performance that it does. Not to mention the superb scaling as you've mentioned going from dual core to quad core.

Quote:

---

Originally Posted by Glow9

I find this ironic as all these reviews always show [T]ard OCP's are so piss poor since all their reviews the AMD rig always seems to get half or 3/4's of the FPS an intel rig gets.

---

Yeah it seems like their AMD systems are running like 2 AV scans and spyware scan along with some scientific simulation,all that while running the actual game code they are testing/reviewing :p: