phaxmohdem
Golden Member
Hehe yes I know. I'm personally hoping Intel has set aside at least one of those Pressies capable of running said speeds, to restart the sun should it ever decide to die out.
Hehe yes I know. I'm personally hoping Intel has set aside at least one of those Pressies capable of running said speeds, to restart the sun should it ever decide to die out.
You obviously didn't read the rest of it. He did stress test it, but he refuses to run SuperPI, because he doesn't like that benchmark. 3.2 was the stable setting he got, not 3.5, he just posted what his max was. It isn't stable enough at that setting yet to run benchmarks. (He said he planned on tweaking it to get it stable for 3.5 benches though.)
- Sep 28, 2005
- 21,131
- 3,667
- 126
sorry when i posted that the thread i read was only 5 posts long, i didnt go back to it after that! I was keeping up with that thread tho, and i was the reason why he gave the no superpi reason. :T
BTW im known as NaeKuh over there. And dont think im disrespecting him. When i read his first post thats how it sounded to me.
And lopri
Yeah tony is an overclocker. You really need to check out his TEC chiller. He couldnt say anything about it, told us hush hush, but WOW. Thats all i need to say. Too bad it got deleted along with a lot of old TEC threads.Also the WR is in Phenom Clock. I dont think anyone has gotten 3.5ghz yet, or thats what the thread was going on.
error8
Diamond Member
- Nov 28, 2007
- 3,204
- 0
- 76
We all admit the superiority of the Intel CPUs, their overclocking and performance over AMD. There is nothing that can really convince us that AMD is better. Because it surely isn't. But this thread shows that AMD starts doing something better then just promises.
- Dec 30, 2006
- 11,366
- 2
- 0
How the hell did P4 get into this. My internet PC very old very fast when new. I would love to show you guys my old benchies. Their were NO AMD FX51 that could spank it ever.
I cann't show them and most these mods know why I can't show as do some of you.
I don't know how a P4C would scale @ 45nm . But I bet it would knock our socks off.
I cann't show them and most these mods know why I can't show as do some of you.
I don't know how a P4C would scale @ 45nm . But I bet it would knock our socks off.
- Dec 30, 2006
- 11,366
- 2
- 0
Originally posted by: error8
We all admit the superiority of the Intel CPUs, their overclocking and performance over AMD. There is nothing that can really convince us that AMD is better. Because it surely isn't. But this thread shows that AMD starts doing something better then just promises.
:thumbsup:
- Sep 28, 2005
- 21,131
- 3,667
- 126
Originally posted by: error8
We all admit the superiority of the Intel CPUs, their overclocking and performance over AMD. There is nothing that can really convince us that AMD is better. Because it surely isn't. But this thread shows that AMD starts doing something better then just promises.
A very very big fat AMEN to this!
Nicely said!
lopri
Elite Member
- Jul 27, 2002
- 13,329
- 709
- 126
Before yesterday, no one would have expected Phenom to overclock anywhere near 3.5GHz. What we'd heard from actual users were how hard it is to achieve 2.6~2.8GHz. I am following the thread @XS and Tony says 3.2GHz was easy and totally stable. He is currently trying to get things stabilized @3.4GHz.
When I saw that post first of course it was quite a shocker just to see a Phenom @3.5GHz but like I mentioned above, I noticed right away the 'unusual' L1/L2 numbers, which is more like those of C2D, from Everest. That was the next shocker. :laugh: The L3 numers are still sh*tty and I do not know how the NB (integrated memory controller) affect the performance, but it looks apparently more difficult to clock NB than the core itself. I wonder if the NB is the key to performance. Tony was able to get the NB up to 2.7GHz when the core was @3.5GHz. The subsequent numbers he posted are also very good almost nearing those of Kentsfield. (Cinebench and 3DMark06)
When I saw that post first of course it was quite a shocker just to see a Phenom @3.5GHz but like I mentioned above, I noticed right away the 'unusual' L1/L2 numbers, which is more like those of C2D, from Everest. That was the next shocker. :laugh: The L3 numers are still sh*tty and I do not know how the NB (integrated memory controller) affect the performance, but it looks apparently more difficult to clock NB than the core itself. I wonder if the NB is the key to performance. Tony was able to get the NB up to 2.7GHz when the core was @3.5GHz. The subsequent numbers he posted are also very good almost nearing those of Kentsfield. (Cinebench and 3DMark06)
Well the frequency of the L3 cache is equal to that of the NB, so NB speed can make a difference in many situations. Leaving the L3 at a fixed frequency while just overclocking the core is not going to be the best way to increase performance. That's one of the reasons sites reviewing the Phenom get poor results at higher frequencies; they don't seem to release that overclocking K10 involves more than the core frequency.
Dadofamunky
Platinum Member
- Jan 4, 2005
- 2,184
- 0
- 0
Dadofamunky
Platinum Member
- Jan 4, 2005
- 2,184
- 0
- 0
I wouldn't take it seriously unless he ran the usual stress tests first. Still, it's nice to see it's possible. I WANT to see AMD compete!
lopri
Elite Member
- Jul 27, 2002
- 13,329
- 709
- 126
The Everest shot is quite telling. For folks who missed from above,
http://www.ocztechnologyforum....om_3500MHZ_everest.png
A few observations:
1. If we compare the L2 and L3, the L3 is approx. half the speed of L2 when it comes to Write and Copy. However, for some reason the Read bandwidth shrinks to 1/3.
2. The slow Read bandwidth of L3 somehow looks to match the memory Read performance. This gives me an impression that the memory Read performance would be linear to the Read performance of L3.
3. Looking again, all memory performance seems to relate to L3 performance, except the latency. (Read=Read, Write=2xWrite, Copy=Copy)
4. If that's the case, what dictates the Phenom performance could be the slow L3, except where random access matters (latency of L3 is still decent).
5. Phenom's L1/L2 performance is at least x1.5 that of A64 and about equal to that of Core 2. (perhaps thanks to the 128bit execution units vs 64bit?)
6. If L3 is what holds Phenom's performance, it makes perfect sense that AMD is enhancing the L3 with upcoming Shanghai. It reportedly will have 6MB shared L3. (vs. 2MB of Barcelona)
http://www.ocztechnologyforum....om_3500MHZ_everest.png
A few observations:
1. If we compare the L2 and L3, the L3 is approx. half the speed of L2 when it comes to Write and Copy. However, for some reason the Read bandwidth shrinks to 1/3.
2. The slow Read bandwidth of L3 somehow looks to match the memory Read performance. This gives me an impression that the memory Read performance would be linear to the Read performance of L3.
3. Looking again, all memory performance seems to relate to L3 performance, except the latency. (Read=Read, Write=2xWrite, Copy=Copy)
4. If that's the case, what dictates the Phenom performance could be the slow L3, except where random access matters (latency of L3 is still decent).
5. Phenom's L1/L2 performance is at least x1.5 that of A64 and about equal to that of Core 2. (perhaps thanks to the 128bit execution units vs 64bit?)
6. If L3 is what holds Phenom's performance, it makes perfect sense that AMD is enhancing the L3 with upcoming Shanghai. It reportedly will have 6MB shared L3. (vs. 2MB of Barcelona)
I don't disagree with any of your analyses, including the snippet above, but I did want to comment that historically when one triples the size of an on-die cache the speeds (in terms of core clockcycle defined latency) do not typically improve as well.
If AMD is able to triple the L3$-size and improve latency or bandwith or both then it would be pretty amazing.
And YES Intel did an awesome job with their P4->C2D transition with L2$ performance but consider that Intel's cache design team itself is likely around twice the headcount of AMD's entire K10 design team...you get what you pay for (are you listening Hector? hello? is this thing on?)
<YAWN>
It's nice to see Phenom clocked up, but he's using water.
Let's see standard air cooling results.
It's nice to see Phenom clocked up, but he's using water.
Let's see standard air cooling results.
What you say can be true in that yes a manufacturing process can be so bad that it becomes the limiting factor in determining sram speed...but that same process and its crappiness would also limit clockspeed of the logic transistors too...so you'd still expect to get roughly same latency (in terms of core clocks) but everything would run slowly.
But if you make improvements in process manufacturing that improve sram speed then usually those same improvements help improve core logic speed too (not always but close) so your sram improves and your core improves...so your latency (in core clocks) remains the same.
It takes design (ports, n-ways, etc) to determine whether a cache will improve speed over the core clocks.
Cookie Monster
Diamond Member
- May 7, 2005
- 5,161
- 32
- 86
Tony was also said that there was no improvements after 2.4GHz for the NB/L3 clock (so anything higher resulted in similiar results).
But if these NB/L3 clocks pretty high (well higher then 1.8GHz), what is keeping AMD from upping these clocks (something that was originally planned)? because of increase in power consumption/heat and that the tradeoff between the two are too large?
But if these NB/L3 clocks pretty high (well higher then 1.8GHz), what is keeping AMD from upping these clocks (something that was originally planned)? because of increase in power consumption/heat and that the tradeoff between the two are too large?
RiverRicer
Member
- Aug 28, 2007
- 28
- 2
- 71
I believe that AMD will improve on the current SRAM based L3 cache by moving to Z-RAM, which it licensed in 2005 and again in 2006. Z-Ram may not be incumbent until 32nm, or possibly the second iteration of 45nm when metal high-k gates and ultra-low k interconnects are due.
RiverRicer
Member
- Aug 28, 2007
- 28
- 2
- 71
RiverRicer
Member
- Aug 28, 2007
- 28
- 2
- 71
@Idontcare
The link you provided was interesting, however it seemed rather biased toward the negative regarding issues with the technology. It appears that the article was somewhat older than the posted date in that the Gen 2 Z-RAM had been cooking for some time and AMD already licensed it in Dec. 2006.
This link might be of interest. The one comment received on the article from Sang U. Kim, Ph.D is included. He lost me after the first sentence!
http://www.semiconductor.net/i...icleid=CA6512566#93486
The link you provided was interesting, however it seemed rather biased toward the negative regarding issues with the technology. It appears that the article was somewhat older than the posted date in that the Gen 2 Z-RAM had been cooking for some time and AMD already licensed it in Dec. 2006.
This link might be of interest. The one comment received on the article from Sang U. Kim, Ph.D is included. He lost me after the first sentence!
http://www.semiconductor.net/i...icleid=CA6512566#93486
As an aside, if you check newegg reviews for the just released 9850s every Tom, Dick and Harry is running at 3.2 on air and benchmarking at 3.5, also on air. Shortest world record overclock evar.
Pretty decent overclocks for the average joe, considering the egg probably hasn't shipped those orders yet. Just imagine what the results will be like once they have parts in hand.
Pretty decent overclocks for the average joe, considering the egg probably hasn't shipped those orders yet. Just imagine what the results will be like once they have parts in hand.
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: 24K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Facebook
Twitter