• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

Caches Cripple Phenom

cosmogenesis

cosmogenesis

cosmogenesis

Soldato
Joined
15 Mar 2007
Posts
3,716
I have read the articles on Phenom and what strikes me is that native quad core yes but the cache sizes are a small company offering. Intel offer 4MB per core whilst AMD offer L2+L3 combines solution which I am sure is slower and makes for a less responsive machine.

THG you the new AMD overclocking tool to get 3Ghz on air stable which gives some improvement.

On the pricing of phenom, they are quad core and not everything is games related and hence they are not badly priced are they ?
 
IMO the memory bandwidth is denting performance.

Once more mobo's are out, bioses updated, memory compatibilities discovered I think we'll see this puppy flying better than the scrambling of benchmarks show now.
 
Frank_Rizzo said:
IMO the memory bandwidth is denting performance.

Once more mobo's are out, bioses updated, memory compatibilities discovered I think we'll see this puppy flying better than the scrambling of benchmarks show now.
Click to expand...

For AMD's sake I better hope you are right. :rolleyes:
 
No come on Vir, the fact that Phenom can be dropped into many a current motherboard is more than Intel are doing. AMD are not being silly here. They are native quad core and a relatively cheap upgrade for that.
 
I noticed that, in theory it should be much higher!

The AMD cache is actually rather clever, the L3 allows the cores to individually buffer from system RAM leaving L2 for direct process queue. Theoretically a more efficient system than just large L2.

However it would seem that the low memory bandwidth scores are currently hampering the Phenom, why on earth they are so slow when it's running 3.2GHz HT speeds is anyone's guess. Perhaps early board BIOS teething problems?
 
theres a post about the memory bandwidth on the memory sub-forum. i linked a thread on XS in it (edit) http://www.xtremesystems.org/forums/showthread.php?t=166222 (edit), in the thread it shows that the present benchmarks seem to be at fault, in un-ganged mode at least you can get far better scores using multiple threads, but most benchs only using 1 thread = 1 64bit controler. 2 threads = both 64 bit controlers, 4 threads = fully loading both controlers and starts to take advantage of the tweaked systems more. off the top of my head the chart looks soemthing like:

read performance
1(thread) 6900mb/s
2(threads) 12000 mb/s
4(threads) 13500 mb/s
8(threads) 13000 mb/s
(edit) http://img.techpowerup.org/071119/sandra_sahara_aod.png (edit)
similar scaling for write and copy.

also, please note that running the HT bus at 50mhz or 5 ghz makes no difference to the memory bandwidth on all amd K8/K10 cpu's (asuming ram and core running same frequency) increased ht link speed gives more bandwidth and lower latency access between the K8/K10 cpu+ram combo and the rest of the computer, eg, hdd's gfx cards sound cards network cards etcetc
 
Last edited:
Yes, I didn't want to cross post but was pointing out the OP that it's more likely memory bandwidth than cache causing worse than expected figures.

If it is the case that benchmarks are unable to understand this new technology and are misreporting results then it could be the case that software (the O/S, apps, games) are doing exactly the same!

As I say, I'm pretty sure we haven't seen these cpu's fly yet. It's not so much that they are launched with the handbrake still on. More like someone forgot to put air in the tyres :rolleyes:
 
Yes well quad core must be better at somethings than dual core regardless of games performance alone. background tasks, a more responsive machine and the literally millions of business machines with power management requirements and all those open applications etc probably figure higher on the profits list then games playing and video encoding.

Maybe quad core will be better for compiling code 2
 
The cache doesn't kill it that badly.
Intel need huge amounts of cache to be able to have low latency between the memory.
AMD cant afford to build huge dies full of cache so they installed a memory controller on the core to keep latency down and manufacturing.
 
:rolleyes: yeah, they're absolutely rubbish, can't run anything on them at all. I've heard they don't even meet the minimum spec to run Vista, that's how slow they are.
 
JimmahuK said:
Whats so good about "native" Quad Core? does it actualy increase performance, overclocking potential or anything?
Click to expand...


On die core to core link, so data doesn't have to be passed around the system to get from one core to another. It will increase performance if the bus is busy or lots of data need to be sent between the core's. dont think the on die link will have any effect on overclocking.
 
cosmogenesis said:
No come on Vir, the fact that Phenom can be dropped into many a current motherboard is more than Intel are doing. AMD are not being silly here. They are native quad core and a relatively cheap upgrade for that.
Click to expand...
I think companies that are rolling out workstations tend to just replace them instead of just changing the processor in old tech.
 
Phenom is competetive with the 6600 but you get 4 cores so when software takes advantage of it the phenoms will out perform 6600. I am starting to see that benchmarks cannot account for anything to do with quad cores as the entire computing paradigm is different now surely ?
 
The ony way I've seen the L3 cache impact upon the Phenom/Barcelona's performance is a direct result of its increased latency.

The bolted-on cache has a disproportionately higher latency penalty than the L2 or (duh) L1, purely because it was a PITA for AMD to fab. However, it also appears that this latency drops (no surprises) as core speed increases.

And for those members panning the Phenom because it is obviously rubbish?

1) Design a better chip whilst the company is restructuring and not making much money;

2) LOOK AT BENCHMARKS! The chip is almost as fast as the comparative Core2Quad, so what's the big deal?

3) A BIOS update lets it run perfectly (!) happily in a standard AM2 motherboard. Will Yorkfield run on 975X?

4) Could any of said members actually spot the difference between a current AMD-based system and an Intel one?


Remember, the 'enthusiast' market is smaller than a lot of people on here realise it is. Believe it or not, the incredibly vast majority of people don't even know what SuperPi is, let alone care what the results mean.

If Mr. Bloggs walks into a shop or goes online looking for a computer to play games with, he will, in all likelihoold, buy a large monitor (i.e. bigger than 17") and as a result, want a suitably beefy graphics card to play with. At any resolution above 1024x768, playing modern games with IQ settings, the CPU does not make much difference at all, to the point at which a 4GHz Yorkfield shows a 1fps increase over an Athlon FX-62.

Yes, a Yorkfield will be faster running Cinebench, SuperPi and DIVX encoding (but curiously, not WMV), but the people who care about that are not who Intel/AMD are targetting.

Of course, this excludes the server and laptop markets...
 
Duke said:
I think companies that are rolling out workstations tend to just replace them instead of just changing the processor in old tech.
Click to expand...

True, but the likes of Dell will have supply-chain comittments meaning they'll be turning out AM2 machines for a while yet. Either for contractual or logistical reasons they won't be able to change platform in a large way quickly.

Companies will buy new machines, but the chances are they'll be using the same brand as before.
 
yes, basically sisoft, which also had issues reading bandwidth on nf2, and maybe intel's first dual channel, can't detect/doesn't use the 2nd connection. now if this is effecting some other benchmarks and some things aren't using enough threads is questionable, i don't think its an issue really.

as for cache, core 2 duo uses massive cache to counter latency, to a point, k8 also used little cache, thats the design of the chip. core 2 duo wouldn't work great with 512kb cache per core, and phenom's most likely would get little extra speed out of 12mb cache.

as mrthingyx pointed out, you will rarely be able to see the difference. in games, there is NO difference, the biggest difference will be 4-5% in sup commander or something, in 99.9% of games fps difference will be 2% max, and no, you can't feel that.

you could take the CEO of intel, or maybe better bet is the head designer. blind fold him, get two computer to the start of first level of crysis with identical setups except one being a phenom and one being a penryn, and he wouldn't be able to tell you which was which. you could run the same test with a 2.4Ghz phenom vs a 4Ghz penryn, and a 2.4Ghz penryn against a 4Ghz phenom, or a phenom/penryn at 4Ghz vs a dual core at 2.4Ghz and again, the guy wouldn't be able to tell you what cpu he was playing on.

as far as it goes, for 95% of us, thats the end result of which chip you buy. i've got a couple intels, i'm switching one system for a phenom, probably my main overclocking/gaming rig, purely because i want to play with it. i know it won't affect my framerate in the slightest.
 
sisoft ... can't detect/doesn't use the 2nd connection. now if this is effecting some other benchmarks and some things aren't using enough threads is questionable, i don't think its an issue really
Click to expand...

If sisoft can't/isn't detecting something it should, then are O/S's, apps / bioses doing the same?

What if crysis doesn't use the memory in the most effective way (in which the new architecture was designed too), what if my customed compiled apps don't either, what if Vista XYZ doesn't...

Maybe it's all down to the motherboard to do the correct signalling / protocol stuff and this is not yet happening due to a lack of products from different suppliers and bioses in safe / engineering mode rather than in the wild mode.
 
You must log in or register to reply here.
Back
Top Bottom