![]-[itman](/forums/data/avatars/m/9/9309.jpg?1650325391){kind=avatar}
- Thread Starter
- #81
Anandtech shows very different results for HL2,
http://www.anandtech.com/cpuchipsets/intel/showdoc.aspx?i=3069&p=3
Besides this, I'll just sit this one out until we get a bit closer to release
-
Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!
Phenom/Barcelona Reviews/Previews and Pre-release Discussion
- Thread starter ]-[itman
- Start date
Anandtech shows very different results for HL2,
http://www.anandtech.com/cpuchipsets/intel/showdoc.aspx?i=3069&p=3
Besides this, I'll just sit this one out until we get a bit closer to release
This also affects the QX6850 which is clearly being bottlenecked by the single 8800 GTX.
You also find the 3GHz 6000+ being slower than the 2.33GHz E6550 in the same two game benches.
Only one socket was occupied:
http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=3092&p=4
Therefore the impact of the second socket doesn't come into play. So in essence, the K10 is up against a 2GHz FX system and the fair comparison is with a Intel quad-core vs an AMD FX system.
- Joined
- Jan 4, 2005
now you've gone and made me hungry!
- Joined
- Oct 1, 2004
- Location
- douglasville ga
lets see phenom on a chipset meant to show the bandwidth it can produce. it is memory starved, high timings, a socket cpu taken out.
an odd setup for sure. lets see a 3ghz amd phenom running 4gb of ddr2 pc8500 with the ram screaming from a good overclock. nice preview with a server platform, but Im still gonna wait for the real thing.
btw these arent even the final revisions to ship out. these are the ones for the big stock holders.
an odd setup for sure. lets see a 3ghz amd phenom running 4gb of ddr2 pc8500 with the ram screaming from a good overclock. nice preview with a server platform, but Im still gonna wait for the real thing.
btw these arent even the final revisions to ship out. these are the ones for the big stock holders.
- Thread Starter
- #86
I showed earlier (maybe in another thread) that the same scaling applies with lower clocked Core2's where the GTX hasn't been maxed out. The K10 scales very well with speed increases, that was one of the only things we kept hearing from people with early samples before the NDA was lifted.
Not sure what this has to do with the discussion at hand. It is true that the 6000+ lost to a much lower clocked Core2, but I also showed that given anandtech's K8 to K10 comparison, a Phenom cpu would gain enough to tie or overtake Core2 clock for clock.
You are correct there, I missed that difference from their server review. I do still stand behind my comparisons because the K8 to K10 comparison was opteron to opteron, not FX to opteron. Adding a second socket will add some latency for certain applications, but neither oblivion nor HL2 are multi-threaded, so only 1 core would be used, no latency penalties for the second socket. Therefore the fair comparison would be A64 to Phenom which I based my calculations from.
Based on the applications used in the preview, it also scales worse than the C2 Quads.
Games are memory latency sensitive:
http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=2879&p=6
The Quad FX loses 6-10% in Oblivion and HL2 when the second socket gets filled. That is the nature of the Opteron platform, each additional filled socket increases latency.
Based on all that, the 2GHz Phenom will perform more like a theoretical 1.86GHz Kentsfield on games.
- Thread Starter
- #88
Examples? I've already shown the K10 scales better in those two games, so I'll have to ask you to be a bit more specific.
Depends on the game, some are a lot more sensitive than others.
You're comparing two different platforms with two different chipsets. There's no way to read into all the factors restricting these two platforms. The editors at Anandtech are obviously very aware of the added latency of the second socket. They are also probably using Windows Vista in their Phenom preview(I did not see that mentioned) whereas the FX system was tested with WindowsXP (Vista handles multi-core scheduling better). All we have is Anandtech's proposals of % improvements of Phenom over A64 with a stipulation that this is probably the minimum we should expect.
Last edited:
- Joined
- Nov 17, 2002
I request that Intel fans stop inciting flames. Its gonna get really old really fast.
Here's a nice link that should be interesting (or maybe get interesting). First capable OC'er I've seen with a set. Just starting skimming it myself.
http://www.xtremesystems.org/forums/showthread.php?t=158051
Here's a nice link that should be interesting (or maybe get interesting). First capable OC'er I've seen with a set. Just starting skimming it myself.
http://www.xtremesystems.org/forums/showthread.php?t=158051
The K10 example used a far lower resolution of 1024x768, making it easier to scale. Plus a 2GHz K10 starts at a much lower performance base. If you do the calculations, scaling from the Q6600 to QX6850 is better than the 2GHz K10 to 2.5GHz K10 for Sysmark, iTunes and Lightwave. The K10 only scales better in WME and roughly the same in 3Dsmax.
It's not an OS issue, it's a platform characteristic. It's precisely why a single dual-core Opteron outperforms two single-core Opteron on virtually every application despite a significant disadvantage in memory bandwidth.
But this, combined with Techreport's review with more desktop and workstation like applications shows Phenom doesn't have enough improvements to overcome C2D's performance advantage per clock in games.
- Joined
- Apr 22, 2005
- Location
- A Labyrinth
I request that Intel fans stop inciting flames. Its gonna get really old really fast.
Here's a nice link that should be interesting (or maybe get interesting). First capable OC'er I've seen with a set. Just starting skimming it myself.
http://www.xtremesystems.org/forums/showthread.php?t=158051
I second this, go spend more productive time learning what Penryn and Neha will offer.
- Thread Starter
- #92
Sysmark is essentially a tie (<2%), and K10 scales better in lightwave (77.6% efficiency versus 70.2%), WME and both games. So really, itunes is the only intel win.
I was simply saying with a more efficient OS with scheduling the difference should not be as large. Either way, like I said, Anandtech would be well aware of this before drawing their conclusion.
We still only have results from a server platform with workstation applications. We really don't know how much improvement there will be given the proper platform and faster, less latent RAM.
- Joined
- Feb 26, 2003
LOL 8 vs 2...LOL wow, some people will grasp at anything to support the "under dog" oh wait, AMD hasnt been an underdog, i am going to buy intel and fight the man!! myself.
- Joined
- Nov 17, 2002
I can say that after 7 pages, that thread I posted is pretty thin on results. Some talk of steppings from people under NDA and who actually have the cpu's towards the end, some random benchmarks scattered around in the middle, but s7eph hasn't managed to get his up and running yet, let alone OC'd. He is still wrestling with Asus for a bios on his L1N64.
- Joined
- Nov 28, 2001
- Location
- in a magical field
The Intel server platforms with FB-DIMMs have additional latency over unbuffered desktop DDR2 as well.
I don't know why there's any debating really, pretty much every review has similar conclusions: K10 is a solid improvement over K8 in many ways and is (potentially?) very competitive with C2D architecture but ultimately doesn't win against C2D overall when clockspeeds are a factor as well. IPC and clock-for-clock are nice geek details to know they just don't tell the complete story. This is pretty typical for an AMD launch, putting aside the Netbust era post-Northwood vs K7 when Prescott couldn't touch A64: as AMD improves the clock scaling things will get closer it's just not the C2D-destroying godsend from the get-go that fanboys hoped. The only possible wrench in the works is how much AMD and IBM can squeeze out of 65nm SOI, ever notice that the highest-speed A64s are still 90nm? It's nice and competitive and has advantages in power use but it's back to 'wait and see' (and hope for the fanboys)
I don't know why there's any debating really, pretty much every review has similar conclusions: K10 is a solid improvement over K8 in many ways and is (potentially?) very competitive with C2D architecture but ultimately doesn't win against C2D overall when clockspeeds are a factor as well. IPC and clock-for-clock are nice geek details to know they just don't tell the complete story. This is pretty typical for an AMD launch, putting aside the Netbust era post-Northwood vs K7 when Prescott couldn't touch A64: as AMD improves the clock scaling things will get closer it's just not the C2D-destroying godsend from the get-go that fanboys hoped. The only possible wrench in the works is how much AMD and IBM can squeeze out of 65nm SOI, ever notice that the highest-speed A64s are still 90nm? It's nice and competitive and has advantages in power use but it's back to 'wait and see' (and hope for the fanboys)
Last edited:
- Joined
- May 26, 2004
- Location
- Japan, Tokyo, Ueno.
Well but cause of the marketing 40% faster, the world's first native quad, add the C2D killer avatars, 30K 3DM or even dave_graham's comment on XS a bit more was expected 18 months after C2D.
They been only bashing Intels not real quad while missed out on a market what's about 5-10% of AMD's worth.
Explains why did they become so open with their future plans recently about bulldozer SSE5 fusion.
They been only bashing Intels not real quad while missed out on a market what's about 5-10% of AMD's worth.
Explains why did they become so open with their future plans recently about bulldozer SSE5 fusion.
- Joined
- Apr 22, 2005
- Location
- A Labyrinth
I can say that after 7 pages, that thread I posted is pretty thin on results. Some talk of steppings from people under NDA and who actually have the cpu's towards the end, some random benchmarks scattered around in the middle, but s7eph hasn't managed to get his up and running yet, let alone OC'd. He is still wrestling with Asus for a bios on his L1N64.
I watched the "yawn" intro seminar running tonight and posted it here but they did show some benches with impressing "looking" results. I think by weeks end we will have an idea what the platform will do. Hope S7 can get his board cooperating.
- Joined
- Jul 27, 2005
- Location
- Austin, Texas
It's server chip and it is much improved over its opteron predecessor. Why so much fuss?
I don't know why anandtech did not use the Asus 4x4 board instead, since that board is the high end gaming platform. It would have made more sense as far as performance in desktop is concerned. Does anyone know why that board was not used?
I don't know why anandtech did not use the Asus 4x4 board instead, since that board is the high end gaming platform. It would have made more sense as far as performance in desktop is concerned. Does anyone know why that board was not used?
The Q6600->QX6800 scaling is more like 85% in Lightwave, performance increases by 21% with a 25% change in clockspeed. And like I said before, the K10 example used a much lower resolution for the games.
It's not scheduling, it's the fact that memory latency is slower with the second socket. No OS can fix it.
We have AMD server platform vs Intel server platform on Techreport with desktop and workstation like applications.
- Joined
- Feb 17, 2003
- Location
- NYC, New Orleans, Tokyo
I believe these chips will be better once they move to a 45nm process. They seem weak to me as I was expecting more. However I'll wait until I see some new chipsets that may take advantage of these cpu.
Similar threads
