QX9650 Q&A and benchmarking info.

[Amele]

I picked one of these up recently and have been starting to put it through its paces.

figured if anyone had any questions about it I could probably answer/test etc.

this is the Yorkfield quad extreme 3.00ghz (45nm). base fsb is 1333mhz. this is one of the multiplier unlocked cpu's in the current gen (the other is the QX9700) and yes, the multiplier can be partial stepped (all I did so far was half steps, if someone wants me to try something smaller than that I can.)

in case questions that require this information come up, the mainboard I have it on is a p5e3 premium, ram is gskill ddr3 1600 (7-7-7-18) 2gb x2. the gpu probably isn't relevant but I'll address that if/when it comes up.

idle temp on the cores is about 48C, at 3.0ghz load it gets to about 56C. the design temp is 64C (critical is 75C) and the design voltage is 1.272v.


currently stable at 3.6ghz (400x9.0) - at this temp it idles at about 50C and under load reaches 60C. I am confident that it can get to 3.8ghz before adjusting voltage, it may be able to reach 4.0ghz at stock. - this is on air cooling by the way.

overall first impression is very favorable, I'm having a lot of fun with the processor and it's very powerful. It's not for everyone at this time because of the price point, but it makes me confident that we'll see good performance out of the mainstream 45nm's once the product line gets expanded.

so, anything anyone is curious about with respect to this processor?

[Cephius]

Not to rain on your parade or anything, but why buy an extreme when you can overclock one of the other yorkfields and save yourself $800

Not to mention most of the 65nm quads can hit 3.6ghz on air rather easily as well, I'd be interested what you can get it up to without getting too hot, otherwise it seems like the chip is really under-performing to be honest. Also what's your ambient like? 48c on stock speeds seems pretty hot even for an extreme.

[mootsfox]

How is the GPU not relevant? FFXI is the only game I know of that can run well without a great GPU.

But if you can afford a $1,000 CPU, go get a copper pot and some LN2, lets see how high you can clock it.

In terms of CPUs, I'm unsure if there will even be another 45nm before Nehalem (which will be 45nm, but a completely different architecture).

[Amele]

Not to rain on your parade or anything, but why buy an extreme when you can overclock one of the other yorkfields and save yourself $800

Not to mention most of the 65nm quads can hit 3.6ghz on air rather easily as well, I'd be interested what you can get it up to without getting too hot, otherwise it seems like the chip is really under-performing to be honest. Also what's your ambient like? 48c on stock speeds seems pretty hot even for an extreme.

well, the first reason is the larger cache; the second is the low end yorkfields don't get to 4.0+ (I'm expecting to be able to bench this at 4.4 on air - I won't run full time on that though, since I'm going to liquid after I finish benching on air). the big advantage to this chip over say, a Q9300 is that this one is multiplier unlocked and the lower end chips aren't. unlocked multipliers makes doing overclocking much simpler and much easier to fine tune.

the second is the 65nms take a voltage bump to hit 3.6, this one did it on stock voltages without any issues - I'll have no issues getting to 4.0; after that I'll need to actually pay attention to temps.

and again, 48C is the core temp, not the ambient. case ambient is about 32C. (core temp as measured at Tjunction.)


gpu wasn't relevant to the discussion I was planning to have. but since you asked: 9800 GX2 (512mb x2). I haven't 3dmark'd yet.

I am planning to go to a phase change/peltier eventually, I wanted to bench it on air first.

[Avanarius]

I'll never understand the point of big number benchmarks, show me some game fps

[Cephius]

I'll never understand the point of big number benchmarks, show me some game fps

Benchmarks are mostly for stability testing and to see what (if any) difference minor tweaks make. If all you care about is how you can run a game then yeah capping out your FPSs is all you need. But most enthusiasts want to push their system to the limits and benchmarks are a great way to measure it while tweaking.

[Amele]

I'll never understand the point of big number benchmarks, show me some game fps

ok, sure. what games would you like to see?

[Avanarius]

I'll never understand the point of big number benchmarks, show me some game fps

ok, sure. what games would you like to see?

call of duty 4 and mass effect preferably at 1920x1200

[Amele]

I'll never understand the point of big number benchmarks, show me some game fps

ok, sure. what games would you like to see?

call of duty 4 and mass effect preferably at 1920x1200

would 1600x1200 be ok? I don't have a widescreen - I'll need to borrow a copy of mass effect from a friend, I can do some measurements on Cod4 tonight

[Avanarius]

1600x1200 would be kinda meh; see if you can cap it out at 60fps. I'd assume you'd easily be able to

really though if you just want to test the cpu bottleneck just go to 1024x768

[mootsfox]

I'll never understand the point of big number benchmarks, show me some game fps

ok, sure. what games would you like to see?

call of duty 4 and mass effect preferably at 1920x1200

I'd rather see it at 640x480, 800x600 or whatever the smallest resolution is.

[Snowknight26]

Don't do high resolutions because then your video card will be the bottleneck, unlike the lower resolutions where the CPU is the bottleneck.

[Amele]

1600x1200 would be kinda meh; see if you can cap it out at 60fps. I'd assume you'd easily be able to

really though if you just want to test the cpu bottleneck just go to 1024x768

1600x1200 4xAA vsync disabled: min 47, max 191, average 102.2; played the bog since it has alot of flame and particle effects. (level takes a little over 30 minutes for me on hardened)

other settings were all yes/normal/insane as appropriate. it crashed once - I'm not quite stable at 3.8 on stock voltage I guess haha. gonna increment voltage and go to 4.0 and see what happens.

I'll do another test when I have a widescreen display if you'd like. (or alternately I can run a 1920x1080 on my hdtv - but that'd be vsync'd by necessity)

sisandra aggregate benchmark is 9769. doing 56762 MOPS; 8.34GB/s ram throughput. doing 32MB super pi in about 12 minutes.


I've got salvage but I'll do a half hour at the lowest res (640x480 I think) after the event.

[jkim1162]

Hmm few questions for you, what are you using to test for stability?
Need to run Orthos or Prime95 on small fft for at least 8 hours (some people will say less others will say more).

Did you recalibrate core temp to base temperatures off of tjmax 95.

Good luck on your road to 4GHz+.

I've had fun getting my xeon e3100 to 4.275GHz, can't wait til nahalem desktop goes mainstream to play with.

[Amele]

Hmm few questions for you, what are you using to test for stability?
Need to run Orthos or Prime95 on small fft for at least 8 hours (some people will say less others will say more).

Did you recalibrate core temp to base temperatures off of tjmax 95.

Good luck on your road to 4GHz+.

I've had fun getting my xeon e3100 to 4.275GHz, can't wait til nahalem desktop goes mainstream to play with.

prime95, 6 hours since I'm impatient. failed twice at 3.8 around the 3 hour mark - I bumped the voltage one increment (like 0.0625V) and it smoothed out. large FFT fails like no other because I don't have ECC ram atm.. oversight on my part; I thought the gskill I was buying was ECC'd turned out it wasn't. won't be an issue for standard operation. (I'm not overclocking the ram anyway, the sticks are just a bit big not to have ECC)

yes, I recalibrated off of tjmax 95. that's a good point. - I had to adjust the baseline temps up from what they were reporting about 10C; (known issue for intel chips heh) for people comparing to other 45nm or late gen 65s that didn't do this, my core temps would read uncalibrated at about 38C idle and not more than 54C under load.

I was able to boot it at 4.0ghz on stock volts with C1E disabled but.. I couldn't run shit without crashes and blue screens - I don't expect I'll need to go above 1.3V to reach it

then we'll see about a phase change liquid setup and aim for 5.0ghz+



for the questions about 640x480 cod4: min: 115 max: 307 average: 197. full settings except for resolution. I couldn't tolerate playing the level for more than a few minutes though at that res lol.

[mootsfox]

prime95, 6 hours since I'm impatient. failed twice at 3.8 around the 3 hour mark - I bumped the voltage one increment (like 0.0625V) and it smoothed out. large FFT fails like no other because I don't have ECC ram atm.. oversight on my part; I thought the gskill I was buying was ECC'd turned out it wasn't. won't be an issue for standard operation. (I'm not overclocking the ram anyway, the sticks are just a bit big not to have ECC)

Yeah, well ECC is for server systems...

Good luck finding FB-DIMMs at DDR2-800 in 2GB sticks for under $100-150 per stick.

I don't even think ECC RAM would run in your motherboard, and if it did, it would probably default out to non-ECC.

If you are using the 2x2GB DDR2-1000 kit from G.Skill (seems likely) you need to bump your RAM voltage up to 2.1v, they aren't made to run at 1.8v which is the stock/JEDEC for DDR2.

[Amele]

prime95, 6 hours since I'm impatient. failed twice at 3.8 around the 3 hour mark - I bumped the voltage one increment (like 0.0625V) and it smoothed out. large FFT fails like no other because I don't have ECC ram atm.. oversight on my part; I thought the gskill I was buying was ECC'd turned out it wasn't. won't be an issue for standard operation. (I'm not overclocking the ram anyway, the sticks are just a bit big not to have ECC)

Yeah, well ECC is for server systems...

Good luck finding FB-DIMMs at DDR2-800 in 2GB sticks for under $100-150 per stick.

I don't even think ECC RAM would run in your motherboard, and if it did, it would probably default out to non-ECC.

If you are using the 2x2GB DDR2-1000 kit from G.Skill (seems likely) you need to bump your RAM voltage up to 2.1v, they aren't made to run at 1.8v which is the stock/JEDEC for DDR2.

it's a DDR3 board (I went with an x48 chipset), but you're right, I checked the motherboard specs and while the x48 chipset 'supports ECC ram' it doesn't actually support ECC. heh. I've typically gone with ECC memory in the past for stability reasons, I should've paid closer attention to the support comment on this particular board. - although further research shows that generally speaking, DDR3 with ECC is vanishingly rare to begin with so I can see why intel didn't support it here.


this is DDR3 1600 7-7-7-18 2x2GB. it's 1.9V in this configuration (which is higher than the stock/JEDEC which is 1.5 for DDR3.) the ram also supports slower configurations at 1.5V for compatibility but I don't have a need to run there.

in the process of doing some research to get to the bottom of this I called g.skill support and they mentioned a specific issue with the p5e3 and asus recommended configuration for 2 DIMMs that causes ram errors in 1600 speed configuration that can be resolved by swapping to a different 2T configuration (using slots 1 and 3 instead of the recommended 2 and 4) I'm going to make that change tonight and see if it solves my problem.

[mootsfox]

ASUS is picky at best with it's RAM. Some sticks play nice and others will only cause problems. I use a ASUS board myself, (P35) and I got lucky with the RAM I used.

For a short amount of time I used a Maximus Formula Extreme (X38) and a 2GB DDR3-1333 kit from Corsair that actually worked very well. I sold it because I didn't think I'd ever go crossfire, and my P5K-e was already doing 500+ FSB.

But since you have an X48, you should sell the 9800GX2 and buy a pair of HD4870's to test in crossfire. I think that would be more fun then overclocking your CPU (as we know how high it goes and performs) until you get a phase change unit.

Back to RAM, even supported RAM kits with an ASUS board can cause problems. If you can contact ASUS tech support and complain enough they might upgrade you, which should have -fewer- issues with RAM. ASUS's RMA center is wonderful, and they usually handle them without needing any purchase info. Failing that, just downclock and loosen the RAM timings, you're going for max overclock on the CPU, don't let the RAM be a limiting factor.

[Amele]

ASUS is picky at best with it's RAM. Some sticks play nice and others will only cause problems. I use a ASUS board myself, (P35) and I got lucky with the RAM I used.

For a short amount of time I used a Maximus Formula Extreme (X38) and a 2GB DDR3-1333 kit from Corsair that actually worked very well. I sold it because I didn't think I'd ever go crossfire, and my P5K-e was already doing 500+ FSB.

But since you have an X48, you should sell the 9800GX2 and buy a pair of HD4870's to test in crossfire. I think that would be more fun then overclocking your CPU (as we know how high it goes and performs) until you get a phase change unit.

Back to RAM, even supported RAM kits with an ASUS board can cause problems. If you can contact ASUS tech support and complain enough they might upgrade you, which should have -fewer- issues with RAM. ASUS's RMA center is wonderful, and they usually handle them without needing any purchase info. Failing that, just downclock and loosen the RAM timings, you're going for max overclock on the CPU, don't let the RAM be a limiting factor.

upgrade me to what? p45? would actually slow down the fsb (I'm not overclocking the ram or northbridge at all, this is the design spec) unless asus has released a p45 chipset 1600fsb board that I'm not aware of yet. It is something to consider if I can't get the current ram issues resolved to my satisfaction though. - I may down clock at some point too to see if I can get higher at 334 instead of 400 base but I haven't hit a wall yet at 400 base so I'm not concerned.

I had considered the crossfire solution - but I prefer nvidia and based on early reviews, the 9800GX2 outperforms the 4870 crossfire setup at the resolutions I play at.* as fun as it would be to play with a crossfire setup, I wasn't going to pay ~100$ more for a weaker solution. (I have plenty of disposable income as seen by the particular cpu I selected; but let's be honest here for a minute: paying a premium for a small performance gain is very different than paying a premium for no gain at all.)


*because of the way the 9800GX2 ram is laid out, the card has issues with resolutions in the 2560x1600 range - the frames get too big since the ram is setup basically as 2x512 separately rather than 1024 common.

[mootsfox]

ASUS is picky at best with it's RAM. Some sticks play nice and others will only cause problems. I use a ASUS board myself, (P35) and I got lucky with the RAM I used.

For a short amount of time I used a Maximus Formula Extreme (X38) and a 2GB DDR3-1333 kit from Corsair that actually worked very well. I sold it because I didn't think I'd ever go crossfire, and my P5K-e was already doing 500+ FSB.

But since you have an X48, you should sell the 9800GX2 and buy a pair of HD4870's to test in crossfire. I think that would be more fun then overclocking your CPU (as we know how high it goes and performs) until you get a phase change unit.

Back to RAM, even supported RAM kits with an ASUS board can cause problems. If you can contact ASUS tech support and complain enough they might upgrade you, which should have -fewer- issues with RAM. ASUS's RMA center is wonderful, and they usually handle them without needing any purchase info. Failing that, just downclock and loosen the RAM timings, you're going for max overclock on the CPU, don't let the RAM be a limiting factor.

upgrade me to what? p45? would actually slow down the fsb (I'm not overclocking the ram or northbridge at all, this is the design spec) unless asus has released a p45 chipset 1600fsb board that I'm not aware of yet. It is something to consider if I can't get the current ram issues resolved to my satisfaction though. - I may down clock at some point too to see if I can get higher at 334 instead of 400 base but I haven't hit a wall yet at 400 base so I'm not concerned.

I had considered the crossfire solution - but I prefer nvidia and based on early reviews, the 9800GX2 outperforms the 4870 crossfire setup at the resolutions I play at.* as fun as it would be to play with a crossfire setup, I wasn't going to pay ~100$ more for a weaker solution. (I have plenty of disposable income as seen by the particular cpu I selected; but let's be honest here for a minute: paying a premium for a small performance gain is very different than paying a premium for no gain at all.)


*because of the way the 9800GX2 ram is laid out, the card has issues with resolutions in the 2560x1600 range - the frames get too big since the ram is setup basically as 2x512 separately rather than 1024 common.

Well, on a CPU with an unlocked multi, how high the motherboard can go is kind of a moot point, eh?

Does you EE max at 13x or 15x? I never could remember with the unlocked chips. At 15x, you can run a 333/1333 FSB and hit 5GHz... Either way, a P35/45, X38/48 can easily do 480+ with a quad in almost every case. You don't need a $350 motherboard to hit high overclocks.

What resolution do you play at? I'm pretty sure crossfired HD4870's beat the 9800GX2 across the board in everything.

Take a look at this review: http://www.hardwarecanucks.com/forum/ha ... eview.html
Keep in mind that they are only testing one card... and that the HD4870 is $300, less than half the cost of the GTX280. (For this example, I consider the 9800GX2 to be on par with the GTX280, +/- 10%)