Sign in with
Sign up | Sign in
Your question

Anandtech has an overclocked Opteron test.

Last response: in CPUs
Share
September 5, 2003 1:03:50 AM

<A HREF="http://www.anandtech.com/cpu/showdoc.html?i=1856" target="_new">Link</A>

The title says "Athlon64 Preview" but it's an Opteron overclocked to 2.0 GHz. Interesting and the gaming performance is stellar. Just hope the Athlon64 does as well as its Opteron sibling.

"We are Microsoft, resistance is futile." - Bill Gates, 2015.
September 5, 2003 1:25:29 AM

Nice read but, who believes that an fsb of 222 with a mult of 9 will give the same results as 200 X 10? It may be ok math but it sure aint cpu sense.
September 5, 2003 1:38:30 AM

Why are they overclocking? Aren't 2 GHz Opterons out already?
Related resources
September 5, 2003 2:13:52 AM

It's gaming processor, period.

Still, I'll leave my amazement to the gaming section, but I ain't much impressed for anything else. ESPCIALLY considering:
1)A64s are likely to begin Single Channel. If the FX is out at the same time, then this point is moot so eliminate point 1.
2)This is with an OCed bus.
3)They did not even include a 3.2GHZ CPU. THIS IS the competition, not a damn 3GHZ.

I still think 2.2GHZ is the margin of competition, along with HT.
Still, if that is the performance to expect, at the same price of the 3GHZ Intel, it just might be ok. But we'll have to see. I just don't think yet the 3200+ rating is deserved again if its workstation perf is about equal to a 3GHZ P4.

--
<A HREF="http://www.lochel.com/THGC/album.html" target="_new"><font color=blue><b>Are you ugly and looking into showing your mug? Then the THGC Album is the right place for you!</b></font color=blue></A>
September 5, 2003 3:19:28 AM

Some thoughts…

Although the bus was overclocked to 222mhz its still running cas2.5 memory and thusly has a latency response time of 177mhz cas2 memory. In terms of random access to memory, single channel DDR400 cas2 would be faster. Remember the XP was, and to some extent still is, competitive for games when using single channel memory.

Hyper-threading does almost nothing for games.

Dichromatic for your viewing plesure...
September 5, 2003 4:10:13 AM

Erm......the HT bus was overclocked. Memory is not on the HT bus, it's on the built-in memory controller. The 222MHz peripheral HT bus is overkill and probably won't matter. However, I wonder whether an overclock of the CPU core frequency overclocked the onboard memory controller as well.

"We are Microsoft, resistance is futile." - Bill Gates, 2015.
September 5, 2003 6:22:28 AM

Doh...

Dichromatic for your viewing plesure...
September 5, 2003 5:25:17 PM

AMD should seriously issue a big document explaining EVERYTHING about the HT principle and perhaps even explain THEIR view of a CPU so that we understand from scratch their clocking methods and what FSB is in their eyes with HT now.

--
<A HREF="http://www.lochel.com/THGC/album.html" target="_new"><font color=blue><b>Are you ugly and looking into showing your mug? Then the THGC Album is the right place for you!</b></font color=blue></A>
September 5, 2003 5:26:21 PM

I feel HT (Hyper Threading) can make a big difference. The CPU can work significantly more (have you read Bruce Gain's article on THG about HT gaming benefits for AI?) in AI and perform amazing AI functions. Additionally it can finally act as a much more efficient rendering co-processor to the GPU. I see it that way personally.

--
<A HREF="http://www.lochel.com/THGC/album.html" target="_new"><font color=blue><b>Are you ugly and looking into showing your mug? Then the THGC Album is the right place for you!</b></font color=blue></A>
September 5, 2003 8:15:25 PM

Quote:
I feel HT (Hyper Threading) can make a big difference. The CPU can work significantly more (have you read Bruce Gain's article on THG about HT gaming benefits for AI?) in AI and perform amazing AI functions. Additionally it can finally act as a much more efficient rendering co-processor to the GPU. I see it that way personally.

Which is perfectly within your right to see it that way. :)  As for me I thought that the HT article you refer to was a complete joke from a programmer's standpoint and that HT, while being a cool technology for making a PC more responsive wile multitasking, is absolutely worthless in most of the situations where people seem to think it should provide a benefit.

<pre><A HREF="http://ars.userfriendly.org/cartoons/?id=20030905" target="_new"><font color=black>People don't understand how hard being a dark god can be. - Hastur</font color=black></A></pre><p>
September 5, 2003 10:50:43 PM

I know what you mean Eden..and that is exactly what hypertransport does..its like hyperthreaing but on hardware level...(it is able to perform more than one read/write operation on the bus). That means if the clock frequency is higher then it can perform even more operations on the bus.

P.S: memory bus has nothing to do with HTB.

---
If you go to work and your name is on the door, you're rich. If your name is on your desk, you're middle class. If your name is on your shirt, you're poor!
<P ID="edit"><FONT SIZE=-1><EM>Edited by pirox on 09/05/03 07:25 PM.</EM></FONT></P>
September 6, 2003 2:05:07 AM

Problem is, since the K7 never needed much bandwidth, it is only safe to assume that HT is made for serious server racks.

To me, HTransport isn't that much for home use, and I'd be hard pressed to find any performance increase it provides for our usage needs. It may help hardware transfer data faster, who knows, but HTransport is like all other Northbridge to South technologies really, it just makes peripherals talk faster.

HThreading OTOH promises a lot if the programmers were willing and if Intel would perhaps consider proving toolkits to help increase productivity. For Slvr_Phoenix as a programmer, coding many threads and making sure they can go in parallel is just not funny. I'm sure if Intel worked hard on making proper HT Compilers or HT Programmer kits, it could do wonders.

--
<A HREF="http://www.lochel.com/THGC/album.html" target="_new"><font color=blue><b>Are you ugly and looking into showing your mug? Then the THGC Album is the right place for you!</b></font color=blue></A>
September 6, 2003 2:08:30 AM

The way I see it, GPUs are not made to waste time on AI, right?
At least not until they move a dedicated Hardware AI unit in there.

The CPU has been the center of AI processing. The Pentium 4 can gain a lot from proper Hyper Threading usage. Now think of how much gaming like FPS games or strategy games can benefit if the extra space is used by extra threads for AI calculations. They CAN make gameplay better.

It is ideal though, granted, and not always a possibility yet.

--
<A HREF="http://www.lochel.com/THGC/album.html" target="_new"><font color=blue><b>Are you ugly and looking into showing your mug? Then the THGC Album is the right place for you!</b></font color=blue></A>
September 6, 2003 2:46:03 AM

Certainly as games get more complex, more processor resources will be thrown at it. GPUs have removed what used to be a major burden, the transformation pipeline. The CPU is not left off the hook; there is plenty of geometry setup to burden the CPU. When the CPU is called on to do its part and feed the GPU, it must be done as fast as possible. Once this is done, it is freed to apply its self to other tasks, such as AI, as the GPU takes over the grand task of rendering. If Hyper-Threading is introduced, it has the effect of smoothing out all the processing, which may slow down the critical processing it takes to load the geometry to the GPU.

It’s like putting a car in overdrive. You may get better mileage but it’s not as responsive.

Dichromatic for your viewing plesure...
September 6, 2003 2:52:42 AM

I don't quite follow your logic comparison.

It seemed to me those who did write for HT had made sure Windows XP had full control of threads and managed properly and will do more as Prescott introduces an enhanced HT (additionally proof that the slowdown is no longer there by less responsiveness, is with WinXP eliminating the huge performance problems the Windows 2000 core had with HT)

To me it seems that WinXP and upcoming versions and Prescott will make Hyper Threading even more viable and programmer-friendly so that it DOES make sure any processing goes smoothly and well-controlled.

--
<A HREF="http://www.lochel.com/THGC/album.html" target="_new"><font color=blue><b>Are you ugly and looking into showing your mug? Then the THGC Album is the right place for you!</b></font color=blue></A>
September 6, 2003 3:16:53 AM

The logic is as simple as this. If you have one task to do quickly, you will do it quickest if you have exclusive use of all the resources. If the CPU gets the information to the GPU 0.01% slower because another non-crucial operation was using a small amount of the CPU during a critical time, it is falling that far behind every rendering and all those little time slices add up. Some tasks, especially those that have to do with rendering, just don’t overlap.

Once you start dividing the pie, your biggest slice is less than whole

Dichromatic for your viewing plesure...
September 6, 2003 6:29:38 PM

Yes but as I said, it has to be sure it can do that task. HT is still young, and I assume Intel will consider the gaming community and think about how to make extra threads get inserted whenever the PIPELINE HAS SPACE. I don't understand how a programmer can't figure out how his game must properly send tasks to each Processing Unit and be able to manage rightfully when HT can insert its functions into the game.

It just wouldn't be used for games if the engineers didn't think of a way to.

--
<A HREF="http://www.lochel.com/THGC/album.html" target="_new"><font color=blue><b>Are you ugly and looking into showing your mug? Then the THGC Album is the right place for you!</b></font color=blue></A>
September 6, 2003 7:41:25 PM

Let me deconstruct a typical frame cycle.

The following tasks are necessary for a game to run.

I - Input
A - AI / Physics
S - Sound
G - Geomerty Setup
T - Geomerty Transfer
R - GPU Rendering

In a typical cycle IASG is done then T producing an R cycle. During this R time the next IASG is processed and the cycle repeats. If IASGT <R the system is GPU limited. If IASGT>R the system is CPU limited. Since I(input) must come before A which is done before G these cannot overlap. S(Sound) can overlap but sound is processed in kHz and doesn’t have much of an effect. So the only true overlap you have is TR over IAG. Yet T is a critical operation and R is done on the GPU and doesn’t affect the CPU. So you get T over IAG which has the affect of delaying the R.

I’m sure there is some situation where Hyper-Threading could improve calculations, but for a typical game I doubt it.

Dichromatic for your viewing plesure...
!